- It discovered that I am a geek (Computer Programmer, New Scientist magazine, Battlestar Galactica, Air and Space Museum, Kubrik, etc.). Not difficult to figure it out based on my answers to some of the questions.
- It recommended me with a few books that I think I would like (Catch-22, Budha's Brain), but it basically failed with more "personal stuff", not in a million years I would like the Spinach Salad and Basketball is not an sport I am interested in. However my main issue with the system is the amount of irrelevant things it presents, such as what kind of toothpaste I would like or what kind of wine varietals I should choose. Recommendations not only should be accurate but relevant to the context.
- Answering questions got boring after a while. I wish I had a way to give it my facebook and twitter accounts and getting the system to figure the answers itself.
More comments once I play with Hunch a little more.
PS: +1 Point for recommending me Gustav Mahler as composer
With data and analytics, the value is found in crossing boundaries and silos. In order for a system to better "understand" me, it needs to be able to access information from the many domains of my personal life: work, personal, health, education, leisure, etc.
As more of our daily activities are captured in digital form, cross-silo data sharing and analytics is somewhat inevitable.
Privacy issues are huge of course...
It appeared to be a combination of recommendations based on the topic already followed by your twitter friends on quora, but also keyword search on twitter account names. For example, as I follow the @googleio account, quora suggested I followed the topic Google I/O. (I should say the idea of quora using account name search is pure speculation by me but you could see how this extra dimension could work.)
I'm going to keep answering questions and continue to 'teach hunch about me' and slowly refine it's recommendations to the 'true me'. From a marketing perspective, I'm curious about how they would classify or profile me based on my responses.
As far as an educational use goes, I think Hunch recommendations could be used a self analysis/reflective tool. The recommendations could be used to initiate a piece of creative writing such as a 'self portrait', 'how I think I see myself and how others really see me' or something like that.
I'm curious to find out how other LAK11 participants are going with Hunch.
However, after answering some more questions, it did turn out some interesting suggestions of sources I'm not yet familiar with such as "The Atlantic".
I believe it could be used as an exploratory tool, to find new resources or start a discussion, although I prefer Amazon's much more accurate suggestions (based on more and better information I believe).
I don't really like Hunch recommendations, maybe it's too US-centered? I know it's my fault, but I've never heard of some of the recommendations before. I logged in through Facebook, does anybody know if Hunch uses my "contextual" information (gender, age, address, and so?)
Nevertheless I have to say that some of the recommendations are pretty accurate, i.e. Stanislaw Lem in first position of Authors or xkcd in Humor Sites (although everybody loves xkcd, I guess).
I would like to be able to ask Hunch why I'm recommended a specific item, not exactly which questions lead me there, but more or less why, it would be useful.
Regarding Hunch uses in education, I think that building informal user profiles when learners enroll into a course (or a degree) would be useful to promote engagement in some learning activities if these are related to topics that are interesting to a particular learner, that is, adapting such activities to specific topics, like programming a maze (if you like computer science and mazes) or calculating orbits (if you like math and astronomy).
We are all products of the cultures that shaped us. If we are influenced by one culture, the job is hard enough. There are those of us who have grown up in two (or more!) countries during their formative years, so our outlook is an amalgam of the "home" culture and the "away" culture(s). It's hard to program for that!
The recommendations that it gave me seemed to be pretty random too. The occasional hit and then a lot of missers. I had the ambition to try out the top 5 music albums it would recommend me, but couldn't bear the thought of listening to all that rock. This did sneak a little thought into my head: could it be that I am very special? Am I so ecclectic that I can defeat all data mining effort. Am I the Napoleon Dynamite of people? Of course I am not, but the question remains: does this work better for some people than for others.
One other thing that I noticed how the site seemed to use some of the tricks of an astrologer: who wouldn't like "Insalata Caprese", seems like a safe recommendation to me.
In the learning domain I could see an application as an Electronich Performance Support System. It would know what I need in my work and could recommend the right website to order business cards (when it sees I go to a conference) or an interesting resource relating to the work that I am doing. Kind of like a new version of Clippy, but one that works.
BTW, In an earlier blogpost I have written about how recommendation systems could turn us all into mussels (although I don't really believe that).
What didn't make it very useful to me is that it didn't ask me anything about my professional life.
Perhaps if I continue to answer questions I will begin to see some recommendations that are not already known to me, and hence be encouraged to wade into previously untested water.
This reminds me of several "personality quizzes" that I've taken over the years when my organization was looking into Employee Behavioral traits or attempting to identify candidates for management. In those cases I always felt like there was an inherent tendency to provide an answer that would be to my own benefit... a "right answer" that someone was expecting. The outcome of this quiz was less significant to me so I felt no urge to embellish.
(Sorry for my bad English)
The first thing I find is that I have to register to access the account through Facebook or Twitter. Do not allow other records, and certainly ensure that your data is more reliable if they come from your network.
I get a screen that'll show you a picture where I have to give permission to that application to dispose of my account information even when not connected and that of my friends with their data and their relationship with me which does not seem right of entry because I do not know for what purpose they will use that information, and I think they should not possess.
Then when you accept, you make a series of 20 personal questions that appear mixed between marketing to sell various products, and personal on religion and politics are very sensitive issues.
In any social network or blog I contribute the data I want, without coercion. I think a student can not you do this for a long series of data that might be interesting to define a student profile. These data flowing through the network after an indefinitely without any control on our part.
My reaction has been negative, I think after this game is a collection of data (data mining) for sale to companies in various sectors. This platform specifically for teaching do not know what to make, really.
There may be other points of view and I would not have noticed them, so I hope someone gives me another point of view.
In the opening of the book Supercrunchers, the author describes how analysis of soil, moisture, etc data provides as effective (or better) analysis of wine quality than a sommelier. Suggesting that analytics can do a better job with tasks that we deem uniquely human is disorienting (at least for me). I hope we can discuss some of the tensions around this disorientation as the course progresses.
As George mentions, having the right variables (soil, moisture, etc. in the case of wine) could provide better recommendations than any human could do with access to limited knowledge. We (as a species) are more predictable than we thought.
I remember talking with Erik Duval and Wayne Hodgings about the Snowflake number (the minimun number of items in a personalized list that make us unique) and how everybody would think that they have a higher Snowflake number than they actually have. Finding out that we are more similar that different would be a good thing in my opinion.
The main issue with recommender systems, being for music, learning or anything else is to find the right variables. There is where extensive data mining and analytics could help us. Understanding which features in a learning material make it more effective for certain students and situations or which learning approach works better given the user and the context are open research question. We have tried several years to find an answer with theory and human experts and we have not succeed. Let's get the help of data and computers to see if we have better luck.
I have been involved in efforts like book tastes suggesting music tastse. It all comes down to asking enough poeple what books they like and what music they like and then and relating the two bodies of knowledge together so the next time someone says they like book X, Y, and Z, they get suggests song, A, B, and C based on some statistical algorithms based on clustering. (I haven't written these algorithms, I have been involved in trianing the system to accurately predict relationship population.)
In my field ins emantics, we hope to make artificalial intelligence as least as accurate as human classifiers (if not at least 10% or better) to take some of the arbitrariness out of classifying or recommending. Typically, the trianing coprus needs to be as much as 10% of the corpus as a whole to accurate traing the system to make accurate predictions. How accurate is accurate? Some of the studies I have looked at are shooting for 75% to 95% accuracy with somewhere in the 80s being the average for accuracy. For many enterprises, their systems are currently unable to have any accuracy so they rely on humans as knowledge bases and interpreters of knowledge and recommenders. Thus information gets siloed and protected and oftern poor business decisions result.
In learning circles, I see huge potential for a system that knows what the course obejctives are, tests look like and how I responded, assingment constructs and my submitted essays, readings, can track when I access a reading, if I made any notes or discussion posts, analyze those posts, and if I score badly on quizzes, suggest exact topics in a textbook that I need to review or articles that ineed to read, remind me when I have need to do any of the above based on the amount of time it would typically take to read, absorbe, reflect, and then be ready to be tested on a concept. And yes, I am asking the system to be my Mother becasue my mother got me to school on time, ensured I did my homework, and encouraged me to excel. So the LMS of the future should not be HAL but rather "MOM" for Massive Online Memory.
I mean, if we accept or not that humans are predictable according to, or to be more extreme, consist of, a collection of patterns, will ultimately depend on our philosophical beliefs of who we are as human beings.
[I understand you never said that we consist of our patterns, but I am afraid that this will lead to that, eventually]
I can accept that my behavior online, the clicks I do every day will tell a lot about me, but I am not prepared to accept that it will "define me" because I can change and because it's just the behavior, not the real me you are analysing.
Disclaimer: I still haven't been able to watch the elluminate recordings or read the papers. I will definitely get that done by today, so sorry if you already talked about this.
Isn't this an example of the Cum Hoc Fallacy?
Just because data correlates does not mean it is causal. I may be wrong on this.
I've read that Google uses the same mechanism to improve search results.
Is the intention here that we think of how this mechanism could be used to direct people to relevant learning?
Began answering questions as instructed .. skipping those questions where I could have easily given more than one answer. After answering about 20 questions it was already prepared to give me a list of results .. but I refrained from viewing them and went on to answer more. After answering what I thought was a sizeable number of questions, it then made it pretty clear to me that it had identified a set of recommendations for me. A rather strange list I thought... consisting of
- fitness programs
- fashion designers
- mac software
- New York Times Bestsellers
- GPS devices
- Beer types
- Music Festivals
- TV Shows
Results for me?
Only two on the list even appealed to me. These were in the "art" and "tv shows" categories.. and to my surprise they were both remarkably accurate in providing a rather comprehensive list. In the art category, of the first 20 listed, I'd say more than 75% I knew and was prepared to note how I liked them. In the "tv shows" category, I'd say five of the first ten were accurate - the other five? Wouldn't touch with a barge pole. Further investigation? I've been labelled an "optimist" ..and a newbie (with a pair of nappies as a logo - cute) I've also earned 293 banjos (still trying to figure that one out).
I went on to examine GPS just to see what they would come up with (I already have two GPS - a handheld and one built into my phone) .. but could see nothing that appealed and wondered why it was on the list. Can't imagine that this sort of thing can predict that I have 2 such devices and am in the market for more (which I'm not btw).
I also found out that I have given 119 "THAY" answers (another acronym I'm intrigued to learn more about)..
All in all .. I'm left with wanting to know more about how these sorts of "educated" guesses have been arrived at. I suspect cross relational analysis of my answers with those of others before me.. and their further selection of "likes" and "dislikes" to suggest that I statistically share a very similar profile and thus may share in these same statistically noted "likes" and "dislikes".
How can this model be used for teaching / learning?
Perhaps a similar approach could be explored to identify student interests and learning styles at the beginning of a course to then direct students in
- their choice of topics and/or methods to study a set of course related concepts / objectives
- their choice of concepts / objectives to help identify and realize a personalized course of study
- their choice of classmates to work collaborative on course work
- a possible course outline that others of a similar profile have found useful
Perhaps during the course, a similar "hunch" tool could be used to
- solicit likes and dislikes on course materials and learning design to test instructional designers on their own assumptions about the appropriateness of design strategies .. ideally even identifying areas for review
I'm realizing just how much of a naive and trusting soul I am here to think that the presentation of such a tool in such a context must have been thoroughly pre-screened and judged appropriate by those that have "designed" the course to then result in my immediate jump in with both feet to set to work on completing the survey and then "allow"ing access to my Twitter account.
However, after reading some of the earlier comments (most notable being Merce Galan's), I'm now sensing a possible personal need to be perhaps a bit more skeptical about the open solicitation of data on me in any context (i.e. "allow" access to my Twitter account for the exercise) .. when I really don't have a clear idea as to how "it" "may" be used. Nor can I perhaps safely assume that even a course designer may know all of the details about how such a site may be collecting and using such data (when I don't see any mention of nor any note of disclosure from in this case "Hunch" as to what they are up to). It's easy to want to think that this is a possible service to me when in fact it may be a disguised form of "data mining" and thus a potential intrusion.
All of that raced through my head prior to me allowing access to a Twitter account. My own quickly considered conclusion was that I am generally guarded enough in my management of the Twitter account to avoid undesired exposure; I made similar calculated decisions about which questions to answer during the Hunch survey. It appears that I was able to respond to the questions in a way that provided a snapshot of my general interests without (i hope) opening the door to unsavory characters. Only time will tell if I see an increase in spam to the associated email account.
In lieu of my "reflections" and perhaps to address any new questions that had arisen out of it, I decided to hunt for information on Hunch that gives some form of disclosure on how information might be used now and in the future ..
I found the following.
Hunch's ambitious mission is to build a 'taste graph' of the entire web, connecting every person on the web with their affinity for anything, from books to electronic gadgets to fashion or vacation spots. Hunch is at the forefront of combining algorithmic machine learning with user-curated content, with the goal of providing better recommendations for everyone.
Hunch provides personalized recommendations on tens of thousands of topics on Hunch.com and is now partnering with other companies to power custom recommendations on 3rd-party sites and applications.
Hunch was started by clever folks who were exploring how machine learning could be used to provide smart, taste-driven, highly-customized recommendations.
Hunch is located in New York City.
.. and terms of service?
Tons of "you" statements .. but not much on Hunch and how it will use the data until I got to this paragraph
How your personal information is used..
and finally discovered what THAY means (Teach Hunch About You)
.. here's where we find out in great detail what can and may happen to data..
We may share with third parties certain pieces of aggregated, non-personal information such as the number of users who engaged in a particular topic, correlations between question answers and item preferences, and how many users choose to buy a recommended item. Such information does not identify you individually.
We may employ other companies and people to perform tasks on our behalf and sometimes need to share your information with them to provide products or services to you. Examples include analyzing data, providing marketing assistance, and providing customer service.
In some cases, we may choose to buy or sell assets. In these types of transactions, customer information is typically one of the business assets that is transferred. Moreover, if Hunch, or substantially all of its assets, were acquired, customer information would be one of the assets that is transferred.
Reviewing, Updating or Deleting Your Personal Information
- Update or correct your personal profile information, photo, or email preferences
- Delete some or all of your Teach Hunch About You (THAY) answers to past questions
- Disable the 'remember answers' feature for future questions
- Block other users from seeing your previously answered questions by setting your THAY answers to 'private' mode
- Delete your account completely
My contribution here, following JIm, Stephan and Merce, is to raise the consideration that the FB/twitter generation is less concerned about privacy than their, er, elders in general. What if the data-mining marketers know what I like? Isn't it a positive to be bombarded with ads for products I'm interested in rather than a random assortment, given their presence in our lives?
Therein lies the root of our mistrust. Are there any among us who have not been bombarded with unwanted advertisements for "performance enhancement products" (nudge nudge wink wink), a Nigerian banking scam, phishing sites, and the like? It's not always easy to discern a legitimate offer from an illegal scheme, so I tend to err on the side of caution.
You are correct that the FB gen is less concerned about privacy. I routinely Google job candidates before interviewing and have in fact discarded some from consideration after learning too much about their personal lives via their unsecured sites.
With all that said -- does it make the task of gathering data easier? I assume it does. Those who willingly open up about their lives in an unstructured format like a FB post or tweet are probably more likely to respond honestly to a structured survey. Through this course I am hoping to gain more insight into current and emerging practices for mining unstructured discussions for meaningful trends.
even the Facebook "free" MBA application expects something in return:
"Joining the Global M.B.A. is like joining any other Facebook application — would-be students first have to give access to their name, profile picture, Facebook ID and list of friends". ( 28 Nov 2010 "Poking, tagging and now landing an MBA" http://www.nytimes.com/2010/11/29/education/29iht-educlede29.html?pagewanted=1&_r=1 )
Mr. Tweet http://mrtweet.net was sort of similar to Hunch in that it searched for people sharing opinions and similar outlooks. Reviewed at: http://www.whoisandrewwee.com/social-networking/new-twitter-analytics-tool-mr-tweet-debuts-and-a-review/
I have just published a blogpost with a quote from you, I hope that is okay? The first part is on privacy, the second part refers to you (will link that part here)
Sarah Haavind, a participant of LAK11, mentions two interesting point as well: the FB and twitter users no longer clutch to privacy (is this really true and has anyone researched this or the reasons behind it?) and secondly she adds to the privacy-commercial link: “What if the data-mining marketers know what I like? Isn't it a positive to be bombarded with ads for products I'm interested in rather than a random assortment, given their presence in our lives?”. She has a point there, but… how can small, very targeted businesses come up through the data analytics that is available?
If I had to choose, I would also like smaller companies to be able to take part in this ad-data-world. I am not interested in starbucks, I am interested in the local coffeeshop burning its own coffee beans, just because I like a variety of tastes. If I go to Ethiopia, I want to listen to local contemporary music, I am not interested in the hashed music that can be found everywhere. So to that effect, I would like it better if I could choose localized data from my profile of ‘local business’ then the big mainstream stuff. But do I need to open up my private life to get access to such time-saving and potentially interesting stuff?
Having said this, gathering and mining educational data might help us a long way to get quick access to some sustainable outcomes (learner preferences, learner critical thinking skills…). But of course, like all data mining, the proof is in the algorithm. If the algorithm is based on a wrong hypothesis, will the results then still be useful? And if data is sold to the highest bidder, then the ethics of that highest bidder better be in sync with mine.
1. the issue of Privacy
2. "bombardment" of ads
As far as Privacy goes, we don't really have privacy; what we DO have is a faucet and we can choose how much to turn the faucet, and in what area. If people we know got together, they COULD piece together info about us, even though we might have given each member a different piece of the puzzle. The internet, and web 2.0 services, make this much easier.
I don't have any delusions of Privacy when using these services; after all they do get hacked, apps get written that take advantage of the info I've stored in private areas, and ToS do change and once private info becomes public. What I want (as a "net-generation" member) is finer control over who I give my info to, the shelf-life of that information, and how these people/apps/places can use my info.
As far as ads go, I don't want to be bombarded with ads - period. Neither ads that are pertinent to me, nor ads that are not. Watching Hulu (a video streaming service in the US), I get A LOT of ads that are not pertinent to me, BUT they are amusing in their own right. I get ads for car insurance that are pretty funny (allstate with "mayhem"), I've seen some pretty imaginative HP printer ads, and of course the "Hello, I'm a Mac" ads. Some (or all) of these don't pertain to me, but I watch them anyway for their artistic (or humor) value.
The product isn't the only thing that matters - presentation is equally important. I'd be happy to watch ads for products that don't pertain to me if the ad is interesting.
It can be use as a tool to obtain recommendations, opinions or knowledge about a topic. It can be use to do classroom research about topics. As much as it is align with the class objectives the results can be more specific and useful. Hunch can be use to know about people thoughts of preferences.
As the say in the Web page:
Hunch personalizes the internet by getting to know you and then making smart recommendations about what you might like.
Hunch's ambitious mission is to build a 'taste graph' of the entire web, connecting every person on the web with their affinity for anything, from books to electronic gadgets to fashion or vacation spots. Hunch is at the forefront of combining algorithmic machine learning with user-curated content, with the goal of providing better recommendations for everyone.Reference: http://hunch.com/info/about/
(sorry, english is not my first language)
Q1: Hunch recommendations - accurate?
I wish to say that it's merely suggesting, out of the analysis done across different aspects of data it retrieves. I noticed that the Books recommended by Hunch are more of those related to the 'idols' or 'people who influence me' in my FB, instead of my choice of answers in the questions.
Even the TV shows it recommends are based on the "type of series" shown, which are family weekly series (probably influenced by my list in FB) instead of my usual preference of movies or non-series base. Recommendations are good, no doubt, but not really what I would go for. I mean, "Star Trek" and the "Big Bang Theory"? Come on... I know I idolise Stephen Hawkings but that doesn't mean I would prefer to watch such series - it's totally dirrent aspect of idolising a person.
Q2: The educational uses of a Hunch-like tool for learning?
One way I could figure, regardless whether it's accurate or not, is the fact that I could predict the type of students I would face in my class. It's something I would do in my own physical class - e.g: I asked my class today (first class with full attendance this sem), of their programme background, just to know what kind of audience I'm facing so that I can relate to them later in my class with my examples, in order for them to understand better on my teachings.
I guess Hunch can be applied in the same manner - not necessarily to be accurate all the time, but averagely acceptable to kickstart a whole new venture of knowing the people you're dealing with in learning and teaching.
Oh ya, it's also about trust. But then again, we can't rely fully on the analysis of Hunch to trust it more than the learners/colleagues. If I put myself in my students' shoes, I would probably believe and rely on data and suggestions given by Hunch to decide whether to trust the student 'next to me'. But as an adult learner and teacher, and also non-digital native, I believe that technology is merely the art of humans, so why must you really rely on it without venturing personally yourself to know for sure whether to trust the person or not.
As mentioned earlier, Hunch predictions/analysis is merely to kickstart whatever you want to do (or decide) next... It's like doing research - it may (or may not be) start from your own "hunch", with some facts lying around in your head, which needs to be sorted out in order to make it more justified and makes more sense.
As a lecturer who is known to be people-person, I believe that Hunch can be used to know another person in order to ease later communication, conveying of message, and setting boundaries to areas within the scope of understanding of the audience/others.
Hope I'm not off-track in my answers. ;D
Kuala Lumpur Time: 2.43AM
1. I was just interested in OAthing and not having Hutch access my other information (as another learner pointed out!)
2. My Fb account actually does not have that much info about my likes/dislikes. I had a lot of info a few versions ago (before they switched to the "be a fan of the [insert TV show] Page" but I did not like the new paradigm.
Hutch was a hit n' miss for me. Some things were accurate (I think I answered 75 questions), like the TV shows that I would like and cuisines that I would enjoy, but other things were not (like European cities to visit). I would visit the cities that were listed, but they are not at the top of my (current) list :-)
As far as academic uses go, working in higher education, I think if we had a system like this in place, where students could connect to their Fb account (presuming that there is valuable data to be mined there), if the learner could start answering questions as they enter college and continually answer questions while they are students, faculty members could have a lot of data on their hands that can help shape the course that students are taking; using such data to come up with examples that are meaningful and relevant to the learner.
I can see the value of a tool like this in that it can expose you to new items that you may not have been aware of otherwise. The big challenge for a tool like this will be to be able to passively mine the results it needs to build your profile for two reasons:
1) Users may not be willing to spend lots of time answering questions. I stopped at 20 just because I didn't have a lot of time to sit and answer questions.
2) Depending on the question, users may answer what they want to tell people they do, vs. what they actually do. I may want to read the books from the Governor General's list, but if I only read the latest pulp SF book, my answers would not be accurate.
It definitely told me what I already know, but it and the discussion here started me thinking about whether something like this tool could be specifically targeted in a way that would be more helpful.
For instance, I could see a Hunch for LAK11 and other MOOCs to help me initially navigate the nonlocalized blogs, tweets, forum discussions, and so on in these experiences. Then it could learn about me by what I choose to respond to, which course readings I've accessed, which Elluminate sessions I've listened to (and whether live or recorded).
Also, I had no problem accessing Hunch without a Twitter or Facebook login; you don't have to provide that information.
I have to admit that I backed out of Hunch as soon as access was requested to Twitter and FB, and I can glean the intent and effect from others' postings. It seems to be a very similar intent as with my Amazon account where I am offered recommendations each time I enter the site. I appreciate the effort of what they are trying to do, but they often recommend books that I have already purchased.
What has been reinforced for me is that the strategy and algorithm behind data mining efforts are really the key to the quality and usefulness of the output. It doesn't just happen. A lot of work has to go into the design and modelling of data in order to get useful results. A good needs analysis is the key.
- rather than be required to answer lots of questions about preferences you aren't interested in knowing more about, you choose the genre of music (topic) you want more information about, and get output/music with additional background information about the musician/artist as well as the music
- you can just get started with what you want to explore right away, even though Pandora is also based on marketing and has a business plan that pushes music to consumers in hopes you will probably like and potentially want to buy (In fact, if you choose an artist, they will not just play that artist for you because they are hoping to expand your (consumer) interests).
- the thumbs up/thumbs down response is broken down into a variety of musical attributes of the song to identify patterns of consistency across an individual's musical preferences. I find the "Why this song is playing" explanations fascinating -- much more sophisticated than amazon's report of why I get certain email announcements of new products/books from them
- Pandora somehow mixes immersion/experience with new knowledge (new music) with marketing in a much more subtle and entertaining, pleasant, learner-centered way...
Additionally, I would have thought a service such as Kyntex http://www.kynetx.com/ or the seemingly defunct Attention Profiling Mark-up Language http://www.apml.areyoupayingattention.com/ would have a far greater educational potential that Hunch.
Kyntex is especially interesting as it uses identity cards where uses shares certain to information about themselves to select providers that they subscribe to. This information as well as certain contextual data, web page visited, time and date - to provide recommendations and information.
Pandora has a deep "expert-mediated" knowledge of the music in its database. See: http://www.pandora.com/corporate/mgp for more...
Playing around with Hunch was fun.The answers may be accurate or not but I don´t think that´s the main point. Using it in a class with adolescents may be interesting. If some questions don´t apply, they may skip them. In an EFL class students would profit by doing a reading Comp exercise and then sharing the answers and holding discussions in pairs or groups. It may lead to some interesting blog entries, therefore allowing them to practise writing and finding their voice.
Now, how is the info going to be used in Internet?
I would rather remain with a system of silos than allow this software with such a close consideration of my interests. It is a useful service but far too personal for my taste. For me the privacy is more important than any potential revelations that the service could provide.
I do see some applications (theoretical)for a service *like* this - privacy concerns aside for the time being:
1. If these recommendations are refined over time, this could be a way to customize your learning across institutions/offerings, etc. Rather than think about learning in a course or program, think about owning your own learning throughout your life. Or perhaps higher ed will evolve to a point where students can pick and choose courses from a variety of schools, where ever you want, fulfilling degree requirements based on their own personalized combination of learning products, specializations and contexts.
2. In an organizational setting, as a performance support system - someone else mentioned EPSS - if I wanted to figure out how to do something, having a system which is smart enough to recommend where/how I could find the answers might be a real time saver and help me learn something faster.
3. If the service could throw you "wild cards" boosting your serendipitous learning, or offer you stretch learning, that might be really interesting, too. The danger of an "intelligent" service that recommends things based on what you've answered is that you may miss out on those things you've learned on a whim, through chance or something else.
Hope these points are valid even though I did not sign on to hunch!
I do agree with all of your points in that they could definitely increase relevant finding performance of all of us as learners.
One remark though, in my institute we are discussing the 'Future of education', meaning how the institute will restructure itself as an educational institute. The most difficult part is for them to be willing to be open, and to give outside students access to some materials. In fact, turning f-2-f content into online content is seen as a problem, all because of old school thinking (I think): research needs to be closed, we need to have students, ....
So although opening up seems the most logical step to me as well, to get those students that best fit (parts) of your program, the human protective impulse will be a difficult nut to crack.
Ironically, many students are starting to go that way already, I wrote a blogpost on this here.
Thx for the great post
What I found, is that in addition to faculty sometimes not wanting this info to become public, students also sometimes don't want to be the ones doing the work. Some students feel like they've paid for the course and why should others have access to the materials that they paid for (viewing perhaps education as a collection of materials rather than a process of learning) OR students don't have sufficient motivation to contribute information during the course or after the course is over.
Here's the wiki I've created: http://pocketid.wikispaces.com
Still a work in progress. Now that I've graduated from the program, it's not as easy to go back and rework all this material without help from others :-)
I have no idea if this is the future or not, but wonder if a more cooperative model might be explored. It's not just open or closed, I think there are other options. Right now, if each school is competing for students, who represent $, I would think that very prestigious schools would do well, very specialized institutions, would survive, but others would struggle and close. I wonder if a few schools combining forces and thinking of new business models, like a cafeteria style selection to degrees using individually managed tools (reducing/cost-sharing administrative costs) is viable. You are so right that the human fear of change is a powerful force.
However, this is a tangent to the original task of hunch, so I'll stop there. I am going to read your blog post now. Thanks for the response.
I was reading yesterday that learners:
* know what they know
* know what they don't know
* don't know what they know
* don't know what they don't know
This last bullet is the most problematic one, and one that set curricula aim to remedy. If students don't know what they don't know, the problem with picking and choosing your own courses means that you might end up with a perpetual blind spot that you aren't even aware of.
I think that educational institutions can do A LOT with the first 3 bullets by allowing students to customize a set (but not set-in-stone) curriculum so that they don't have to repeat things they know, and they don't have to repeat things they know in order to know that they know them.
The whole data thing is a very institutionally driven concept, wondering if we looked at it from another perspective if we'd see it differently.
I just thought of it as kind of like my personal learning records, almost like my personal learning network.
In my eclectic life, here are some perspectives that I draw upon:
I am very involved in my kid's schools (primary/middle years) and volunteer regularly. So much of what happens in any given school year is teacher dependent. Kids aren't empowered to own their own learning. As they move into higher ed, they have more control over what they learn, but not complete control.
If they move into workplaces (this is where I work), there is no interest in what/how they learned in schools, but many orgs do have their own system to classify your personality type/style. Even there, you are a worker who receives workplace learning. But, your learning records and even performance feedback could be part of your overall data portfolio.
If you are pursuing pro D of your own, THEN you get to be in the driver's seat, but you are likely to fall into your 3 bullets described.
Maybe a hunch like system with a rypple.com type addition could take your cumulative learning data - trending, pattern recognition - and give you suggestions based on more input sources than your own answers to questions.
Maybe I'm way off-base, but thanks for letting me pontificate anyway! Appreciate your response.
The first is that it is fiendishly hard to match paths of learners because a) there is no guarantee that others took the best path in the first place (because, unlike in matters of taste, they don't know what they don't know) and b) the combination of prior history and future goals is very difficult to model, so it is much harder to match people accurately. Most systems that have been built for this so far, including my own, are more to do with riding a wave than following a path, with sights set on the next thing rather than dealing with the complete learning trajectory.
The second (related) issue is that, unlike for questions of taste, where previous likes are a good predictor of future likes, once we have learned something we no longer have any need to learn the same thing (as a rule).
Just a couple of thoughts. I'm catching up with this conversation so will probably have more to say on the subject!
Hunch is a recommender system that uses your likes/dislikes to discover new relevant information objects for you. I do believe in this approach and at the same time I find this (as Stephen Dixon mentions) too simple an approach. Even though "when there is interest, learning will happen" (Arthur C Clarke), I think "stuff you like" is sometimes entirely different than "stuff you need". Edu-Hunch (or something) will keep you satisfied with recommendations about stuff you are likely to like, but will it give you an answer to a question you have? Also, "liking" itself is a very limited way of expressing your thoughts on something. You might like pizza, but no anchovis. If you dislike a Pizza Anchovis, what does the system store? On the other hand, some interesting analogies and recommendations can be done with relatively simple like/dislike statements and multiple choice answers.
If you take a Hunch-like system, but rather than expressing your interests, you are proving your expertise by answering questions, you might be able to combine interests with knowledge, a much more powerful combination in learning environments.
Some thoughts on Hunch and how something like it could become useful for education:
- Hunch should incorporate serendipity.
- As I mention above, I think that an Edu-Hunch should be combined with tools focusing on expertise/knowledge.
- We should not forget the medium and form. A very interesting topic can be delivered in a very uninteresting way, and vice versa.
I gotta go sleep now. I might add something tomorrow.
That being said, in general, I find applications that use self-assessment to build predictions/recommendations to be limited to a large degree by the respondent's self-awareness (and, of course, the level of nuances permitted/detected in responses). No brainer, I know, but just as true as ever.
pre Hunch play thoughts.
Determine if it has allowed all my current open silos to cross reference openly. More to me is best. What will aggregation accuracy be? Has Hunch applied unknown filters or has Hunch an unstated commerciaal intent bias? Curious.
Hunch learning will be best informed (changed/modified/corrected/deleted/improved/supported/justified etc) if an open data deluge (positive) flows (via white list intent filters) through.
Would a transparent legislated intent 'service' offering open data discovery, with a shapable filter, be mandatory for and on ourselves?
Be back post play to report.
Ok, back from playing. Overall, that was crap. Hunch gives data a bad name.
After supplying 647 THAY answers, I'm convinced input is redundant to any subsequent Hunch suggestions. Premeditated marketing if I was a skeptic.
Their commercial sales intent of most (all?) suggestions spoils any potential 'complex data for good' nuance that may exist elsewhere.
Hunch should have mined ample evidence of my specific interests with an even cursory scrape of already available open data. I have that permission out there.
Cookie cutter junk mail suggestions were returned. At least a Horoscope is usually a laugh. This wasn't.
As I scrolled through "show more", obviously all manual input data (647 remember Hunch?) was disregarded as Hunch still pushed suggestions I'd explicity informed them I don't or can't do. (FB games, USA only product WTF?)
But they wriggle off the hook with
Hunch does not evaluate or guarantee the accuracy of any advice or content on the Hunch site Well meh, true dat.
Forgot to ad how this may impact school student learning.
Kids will flock to interfaces they like to use and give up data readily. Critical literacy teachable moments abound.
The benign interface of data collectors posing as something else raises ethical questions. Farmville is not just a game. not at the back end anyway. Nor is Hunch.
Should transparent intent of data collectors/miners be controlled through legislation? Who watchers the watchers? Grist for the mill.
Creo que si bien facilita la búsqueda de algo cercano limita la exploración de posibiliddes no exploradas, como un análisis discriminante.
Al buscar playing encontré muchas concepciones asociadas para las cuales si existen palabras diferentes en español. Lo que me permitió conocer los diferentes significados en inglés y las opciones asociadas a la palabra.
En español tenemos una expresión, "en gustos se rompen géneros", que nos da la idea de que cuando menos los esperamos encontramos en las generalidades las especificidades.
Me hago una pregunta ¿en la medida en la que se expanda la base de conocimientos aumentará su grado de certeza y confiabilidad o su complejidad y variabilidad?
It seems as if you were traveling a route that is defining the new questions and new possible answer. The answers proposed graduation admit any and every society or culture gives it variety and diversity.
I think if it makes finding something close limits posibiliddes exploring unexplored, such as discriminant analysis.
When searching I found many concepts associated with playing for them if there are different words in Spanish. It allowed me to understand the different meanings in English and the options associated with the word.
In Spanish we have an expression, "in breaking gender preferences, " which gives us an idea of when we least expect to find in the general specifics.
I do a question how far to expand the knowledge base will increase the degree of certainty and reliability or complexity and variability?
Applications for learning: I'm interested in the idea of programming adaptive learning environments and the tech that might be used to do this. Hunch seems an interesting tool you might use if you could train it to identify a learner's interests and background to suggest enrichment activities and resources. The downfall, of course, is when the tech is only marginally accurate and pigeon-holes a learner into a course of study that isn't really responsive to his/her needs. The learner follows a machine-suggested program and, because you don't know what you don't know. is led down a path guided by false assumptions. However, if you can blend tech like Hunch with some human elements (e.g., a discussion forum with teacher/facilitators like this one who curate and filter content), it could actually fairly useful. Also, if the learning program is one where the learner is involved in experiences that are pretty close to real world performances, the learner, to some extent, identifies what he/she doesn't know when he/she has difficulty completing a task. Perhaps a Hunch-like system could also be used then to suggest avenues for further exploration, tailored to specific learning difficulties.
I was immediately on guard about access through Twitter and Facebook and Hunch's disclaimers are just amazing! I don't want to open the email gates to unwanted advertisements or sales pitches: why wouldn't students (and their parents) feel the same way? I think that teachers have no moral right to expose children to this type of data mining- by asking them to participate you are validating the site. I know that some of my adult students would flatly refuse to be involved. I'm an individual and while I am unavoidably a statistic, I would not want and do not choose that an algorithm determine or influence my personal actions and choices or worse, have some other person tell me that it should, although I do realise that my life might well be changed by a statistic generated by a population census, for example.
The idea is interesting, even though it sounds like a up-market survey from a popular women's magazine, and seems as though it could identify trends (though as some respondants have said who wouldn't want to eat Italian food and drive a Porsche?). However, I could not take what it says about me seriously- it seems a bit like an astrological chart (horoscope is right, Diane) pigeonholing an individual according to some pre-determined model. Bit creepy, really- shades of Aldous Huxley.
Hunch is not the answer! (What was the question, again??)
The blended tech/human interface you mention is of interest. Transitioning steps before machine learning will be significant.
How 'big' would data have to become before it starts to return learning scale?
Learner data exhausts (needs) will become richer and cleaner when fed from disparate but open silos.
That then raises disturbing ethical/privacy concerns if the intent of miners is opaque. Posing as a benign game but repurposing and on selling.
Synchronous live data analysis then feeds this repurposed exhaust back into networks as evidence. Networks become self informed during the learning using the blended tech/human interface you mention.
Learning modifications, additions, adjustments can then be made once informed patterns becomes clear.
I follow the remark that it is US-centered, but also that it is rather old school in its questions (putting a robot next to a male or female toilet icon is not new, at least put in a third gender), similar with the geographic linking (that is 20th century thinking).
Additionally, I felt like selecting 'wrong' answers to get a more differentiated result, I did not want the results they gave me as I am an eclectic person and it only featured (US) limited top 5 (because I am quickly bored by repetition).
And as some of you mentioned before, what is the relationship, what is the data-mining algorithm? And why are algorithms mostly for mainstreams?
In the end I got 'firefly' in as a series, okay so that works for me and so I stopped.
If data is used for educational purposes, it should be open and transparent, so it can be viewed in a way that is favorable for the learner (holistic, specific, links...).
I also do not belief hunch is without commercial idea, I think that the suggested magazines for instance might be in the algorithm to come forward. Some sort of hidden advertisement. Well that is what I would do if I were into profit: get advertisers in and tell them that if they pay X dollars, I would tweek the algorithm.
As such, this is my result, so how diverse are our results :-)
|Van Blogger Pictures|
The fact that you are supposed to receive suggestions specifically designed for you creates a state of mind prone to the self-fulfillment of the prophecy. I might agree or not that my favorite sport is soccer, but that statement has the same validity for basketball, biking, etc.
Some suggestions are curious, and might make you try some music composer, writer, or shampoo brand. But the amount of information to receive in return for the amount of answers you provide does not pay off.
The most troubling aspect are the questions about political/religious believes. I would say those (and some others) are fairly delicate topics. Seeing them next to question about the first thing I use when I shower seems to be deliberately created to lower your reluctance and gather this information. Paranoid, may be.
As for its use in a learning scenario, as is, I'm even more skeptical. The main reason is the scope of a learning experience. In a learning scenario there is typically a fairly delimited topic or area. A platform like hunch should have more specific questions to provide also much more specific recommendations. This would take us to the issue of specificity. The more specific the topic, the more specific the recommendations, I think the less effective the approach shown by hunch.
Could the Hunch approach be modified to easily adapt to a specific topic in a learning scenario? Tricky and challenging, but interesting nonetheless.
a) Identify learning styles via completion of variously presented mini-learning activities
b) Suggest learning content which matches learning styles and objectives
In this way, the learner would determine what they want to learn, with edu-hunch then suggesting how and where to learn it best.
I do think that it's a narrow way of defining how one learns. Much more goes into learning that just learning styles.
Hunch (the web-based recommendation service) seems to make the assumption that I have one, persistent and more or less predictable set of preferences. That's just not true for me, and I suspect it's not true for a lot of other people too. I don't work the same way as I play. Or do I? I remember reading in another thread in this forum that it's a bit unnerving to see a machine be able to predict what I like or dislike. It's certainly worth thinking more about this sentiment!
Great points here: "I wonder if there is merit in thinking about the effects of, say, learners switching to a very different "profile" depending on context. I wonder if there's some middle ground between big categories and continuously changing profiles."
My colleague at TEKRI - Jon Dron - has been working on what he calls a "context switcher" for allowing individuals to change their information flows and social interactions based on their context. We are running an elgg install (called the Landing). We're using this as a social network site to create a virtual learning/social commons. As more of our social identities bleed into each other, we need a means to filter and share information and activity based on context. Jon's work is quite forward thinking in this regard...and your comments of balancing fluidity (profiles) and rigidity (categories) fits nicely in the discussion of context switching.
RE context switching, absolutely: we are different personas with different needs at different times when we have different intentions and different contexts. To assume that we are a single kind of thing, socially and personally, is just wrong. Quite a bit of the work I have done on recommendation strategies is to try to provide recommendations which do not rely so much on a mined model, but which allow people to create their own disembodied user models. A very simple example would be the use of tags to describe what it is that we value in a resource. Once we have done that, others who value the same things can explicitly choose that facet of their interests to find stuff that helps. Combining this with implicit and explicit ratings in the context of such a tag can lead to recommendations that relate to a particular need in a particular context, rather than an assumption that we are a simple flat set of attributes.
Another approach is to use scrutable learner models, as Judy Kay and colleagues have done. Scrutable systems let you vary the parameters that are used to define you. I quite like that but it runs into difficulties on implementation because it requires people to understand how the user model is being used to provide recommendations, which adds layers of complexity most people are not keen to deal with.
Using clusters such as learning styles might be one way to help simplify that though I am *very* wary of stereotyping of this nature, especially in the light of Coffield et al's wonderful and thorough demolition job on learning style theories a few year's back. Essentially, learning styles are mostly stuff and nonsense and, even when they can reliably identify sets of related learners, offer no reliable means to apply that knowledge to achieve better learning. However, maybe there is scope for emergent, bottom up clusters to be used, identifying similarities through statistical/network analysis techniques like k-means, though it would be hard to give labels that people could reliably identify with using such a technique. Interesting possibility though.
Also difficult for me to sit for long periods of time poring over content -- I need to move and do.
Knowing this about myself, I seek out the types of learning experiences that work for me and tend to avoid those I find cumbersome.
Doesn't this mean I learn in a different way (dare I say 'according to a different style') than someone having the opposite inclinations?
I take your point about the non-au-courant-ness of learning styles -- and I think this is true when considering instructor or institution perspectives in building learning experiences for students.
The situation I'm describing is a learner planning their own learning -- in which case the institutional perspective itself is no longer au courant...but being selective in how I choose to learn what I want to learn certainly is.
The hesitation I had signing up for hunch, playing with the tool along with reading everyone feedback has me thinking two of common barriers that I consistently see slowing down BI projects - privacy and imperfect interpretation of the data.
I found it interesting to hear the various feelings and perspectives on Hunch's request to access personal data via facebook. With such a variety of individual comfort levels and concerns regarding sharing personal information, it is understandable why organizations struggle and are cautious with "breaking down silos" and sharing personal data of their constituents.
Personally I was very uncomfortable with Hunch's request to access and use my personal data, so I used an inactive twitter account that has no personal information. As organizations, I guess we need to do the same thing. Recognize the risks of breaking down the silos and them minimize those risks as we make use of some wonderful tools that have potential to help our organizations and people that our programs are trying to serve.
INTERPRETATION OF DATA
From our sample of users, Hunch's predictive accuracy is hit and miss. From my limited use (I answered 50 questions), Hunch did a pretty good job. I was amused that even though I indicated I did not watch TV and did not read many book, that the first two prediction themes displayed were TV shows and books. I was, however, surprised to see the Daily Show and the Colbert Report on the list of TV show (as these are the only two shows that I ever watch). As Stephen D, has stated 'people are far more complicated than what Hunch can unravel' But as Shahrinaz has indicated, this doesn't mean that these tools cannot be useful.
This week’s readings mention the fantastic work that Campbell has done creating Signals at Purdue (dashboard that provides students with a predictive course success score after the 10th day of a course). Initial feedback regarding the use of Signals from Purdue faculty focused on the need to understand how to interpret and act on the information provided by Signals. So I think that a lot of the issues around interpretation of data will come out of educating data consumers and providing best practices on how to use aggregate/predictive type data/information.
I've heard that many traditional cultures have/had strict rules around the sharing of traditional knowledge. Certain teachers are selected to be 'Knowledge Stewards' and they decide who will carry certain knowledge forward. Contrasting this framework is the 'Information Democracy' culture, where data is freely available for all to interpret and participate in the creation of information/knowledge. Although I have worked in BI for the last 15 years and tend to be supportive of the latter framework, I also am cautious by nature and recognize and appreciate having Knowledge Stewards to direct the show. I'll look forward to more discussions and activities that help me to reconcile interpretation, privacy and data access/sharing.
Here's another incredible project that provides aggregate metrics on what people are feeling by gender, age, location and time. It has been around for some time . . . but if you haven't seen it before, it's worth spending 10 or 15 minutes playing around - http://wefeelfine.org/.
One of the articles this week mentions that data access is one of the biggest barriers to making progress in data mining in the education sector. If it is more difficult to get organization comfortable with sharing data than getting individuals to share their personal data via tools like facebook and twitter, should we be looking at building some of our predictive models outside our individual institutions, allowing students to opt in with their own personal data and then allowing them when to decide whether or not to share it with an instructor.
For me the recommendations weren't accurate at all. That's probably because it is designed for persons in the US and I am in Europe, but some other issues as well. The program recommended me, e.g., only veggies as meals (because I am female?) and MAC products (because I was too fast in answering the questions and accidently clicked that I am a MAC-user which I am not – that’s how you produce false results ;-).
I am also very much concerned with the privacy of my data, so I logged in via Twitter (which I rarely ever use) and after using the tool, I denied Hunch access to my Twitter info (you can do that in your profile).
Looking at the responses here, I noticed that some people didn’t want to use Hunch in the first place and others wouldn't use it again. However, somewhere out there are a lot of people who use it. Would be interesting to know who the users are. The young ones who aren't concerned so much with privacy? Or the ones who don't know that or how their data is used? Or those who just don’t care about their data being used?
Why would I need a tool to tell me what interests me? Maybe because the amount of information and products being offered to us today makes it impossible to keep an overview of everything that’s available. On the other hand, things I don’t know about, I won’t miss...
Thinking about the use of analytics for education, the one thing that comes to my mind is the creation of personalized learning environments on the web, something such as netvibes, but maybe going a bit further and suggesting you new topics you might be interested in or schools to attend that best match your profile.
For the truly paranoid:
- Make sure you sign out of Facebook and Twitter and delete all your cookies
- Create a throw away email address
- Sign up to Hunch and play around
- Dress up as somebody from one of the other two genders than yours (thanks Ignatia)
- Go to a public Internet computer and wear gloves
However, it found many of its recommendations hard to evaluate because most of they are beyond my prior experience or knowledge. Also, in some cases, recommendations it provides for a specific topic tend to be the most well-accepted options--rather than based on my answers I guess.
Another thing is, it's a one-way recommendation, rather than an iterative loop that enables users to provide feedback on recommendations. I need to understand how did Hunch recognize those initial patterns or models, and whether/how does it keep improving them.
And I wonder Who am I to give the particulars of my friends without their knowing, to a private company that openly publishes its data?
On the other hand, I agree with people saying that the questions are mainly aimed at an American audience because he missed me the answers.
But I keep turning to see what this program can bring to teaching and learning.
I tried it out today and thought the site did a great job. The user interface is excellent and the questions were direct and revealing. My recommendations seem to be in the ball park for me and there are a few books/tv shows that I wasn't aware of - must find more time in day.
On the education implications, I'd like to see if these responses from students could yield any additional predictive power beyond current state. Other psychological research has often found very modest effect sizes for these types of self-report indicies on student success. Still the experience was fun and it might build student's investment in their learning community - depending on how it would be implemented.
For example, numerous sites exist that help people find new music (pandora), new books (amazon), relationships (eharmony), etc. In spite of these developments, most training and education initiatives have done a very poor job of moving away from the "one educator, 30+ student" model.
I'm interested in:
1. How can we come to understand our learners
2. How can we provide learners with learning relationship suggestions (i.e. "since you're connected to John, You might find Susan to be a good person to follow for chemistry discussions").
3. How can we provide learners with content suggestions that accounts for their current state of conceptual development and fills knowledge gaps.
Recommender systems (for content or relationships) is a small part of analytics, but it is one that is developing fairly quickly in non-education fields. As such, it is an easy jumping-off point in moving toward learning analytics adoption.
So the idea behind Hutch is it's predictive power - I can see this harnessed to be a powerful assessment tool. As you answer questions correctly, incorrectly, or in a specific manner, a student could get new questions that were tailored to reinforce "correct" knowledge, and if you get an answer wrong, you get a new question that is on the same topic or approach that is slightly easier, to get you back on track.
Hmm. I think I'm rambling. But this could be an example of mastery learning. It would take a lot of time to set up, but perhaps the system could learn itself based on which questions students get right or wrong.
I answered much more than 20 questions perhaps because they are accessible or because I had the LAK11 motivation!
The recommendations were OK and it was pleasant that it detected that I work in several languages.
For learning I can only think of helping to join students in a MOOC by groups of affinities, by objectives or by deepness on certain knowledge.
A good experience but not a revolution!
The speed with which Hunch created a profile based on 'partial' information was impressive. The results really were not particularly accurate.
I tend to agree with those who suggest that the results reflect the design of the tool and the interests and priorities of the designers.
As far as the educational uses of a tool like this, I am not so sure yet. It seems more focused on delivering recommendations for consuming products from the marketplace. It does make me wonder if the same kind of profile that is slowly developed and gathered in Hunch could be done in a more traditional academic setting, based on learning experiences. This kind of profile would likely require a different sets of questions.
Still, looking through the Education and Career recommendations, the list is pretty deep and interesting and certainly offers a lot of possibilities worth exploring in more detail. I could really see counselors in secondary education incorporating it. The school where I work, the students go through a guidance seminar for a semester, during the year leading up to their applications to university (16-17 years-old). This might be really useful to students in that context, as they are still trying to get a better sense of where they will be heading in the near future.
After reading almost 74 posts, I decide following one of the advice to create a new twitter account just for fun and answer a lot of questions, analyze the results I found that hunch identify some tv shows that I like, but nothing else.
This tools could be use as a learning adventure trying to identify how our digital footprints give away our marketing and buying potential, identify things that people from certain background or age could have in common. If you have a class and want to discover things they have in common you could use this tools, and prepare a group chart of some of the answers.
Then discuss the importance of our connections, the power of social networking and the future in commerce and education.
My first impression is this same kind of application can be very useful if integrated to virtual learning environment, since it can increase the connectivity and interaction with content and learners.
In virtual learning environments questions related to the context of course can help assess the level of student knowledge and suggest content (articles, books, podcasts, etc.) and activities, for example.
The data we provide to virtual environments may make them more semantic and intelligent.
I do not have a Facebook or a Twitter account so I used the old-fashioned way to create an account on Hunch. I was curious to see how well Hunch would "describe" me without information other than the one obtained with 20 questions. Plus, I tried to make things challenging for them by not answering the direct questions on age or sex (which they probably figured out anyway based on my other answers...).
My impressions on Hunch
Like a lot of people in this forum, I found that the recommendations were generally not accurate, and that they were very commercial and US-centric. I actually find that very interesting, because it reflects exactly my concern about data mining in general: can't results be biased by the database itself? If Hunch included authors, writers, movies, or products from outside the Western/US culture in its recommendations pool, would its results be more accurate? Also, Hunch's own biases in interpreting my answers can influence the results: being concerned by the environment doesn't mean I want to drive a Prius... But I guess it's all based on probabilities. Programming algorithms to find relationships between data and draw conclusions must be extremely difficult!
Also, it seems to me that the database Hunch is building by asking us all those questions will be much more interesting: businesses will be happy to use it to perform market researches! (It made me think of those endless questionnaires on brand products they used to send you by the mail...) Although they might end up with the same "representation" issue if most of the accounts belong to Western/US citizens...
Use in Education
I wouldn't recommend Hunch itself because it's too commercial, but I agree with many participants that a similar tool could help define learning styles, match students for teamwork, etc. Also, I like the idea of using such a tool to recommend rather than to impose: students need guidance in their learning process, but they also need to be in control. I think making recommendations and offering choices based on that is an additional step towards personalized learning experiences.
There are three kinds of profile data that one generates when using web apps:
1) Unconscious - for example, when you play a game (or use a simulation), the app can learn a lot about you without asking you to reflect on yourself.
2) Conscious/spontaneous - thumbs up/down, rating 1-5 stars, etc.; you are asked to reflect on yourself, but not deeply
3) Conscious/deliberate - written comments or other constructed responses where you are asked to present yourself as you wish to be perceived.
Hunch works since it's operating at the second level and only making recommendations about things that are entertaining. They are neither life-changing recommendation (as a good set of questions in the third category would lead to) nor are they correcting your mental model (as a good series of questions in the first category support).
For education, you need all three categories of questions. The first category is embedded formative assessment and the third category is more like e-portfolios. The second category looks more like interim or benchmark assessments - that neither change your thinking nor express your deep goals.
Hunch : Social Networks :: Chapter Quizzes : Education
(light assessment leading to light guidance).
First of all, there's a lot of posts to go through at this stage!! Lot's of different opinions, observations and discussions....great! One I haven't seen so far (but might have missed!) was CREDIBILITY!
For me, a tool like this must demonstrate that the data model in the black box is credible (the model might be incorrect, and the data as well, but I still want it to be credible and objective).
The point where Hunch lost my credibility was after answering the question on MAC vs PC (Sorry, I never use MAC, nor have an iPhone!), and getting almost all Apple PC recommendations in the Desktop Computer category(4 out of 5). This demonstrated to me that it's more a marketing tool than really data analysis, and put all their answers in a different perspective. Was that and other results generated due to marketing contributions, or do I really need to start using Apple? Would that book be really interesting, or is Hunch payed by the publicer? Hunch should not attempt to re-establish my credibility and it's most likely the won't bother...
On the positive side, it did generated some typical "he, that's me!" suggestions as well as some potential new discoveries. But mostly in the non-commercial topics.
I really liked the professions it suggested to me.
I think a Hunch-like tool would be useful if the questions can be appropriately designed to predict the learners' learning styles, the level of social interactions, the propensity for openness, active engagement in learning communities and sharing.
- Three magazines that habitually I read. Two that not.
- Five TV shows, one that I know and I do not like and four that not (and I have no interest in them)
- Five books that have nothing to do with my personal tastes,
- One of the holiday destinations is my city: Barcelona
- Only recommend me Italian and Indian food (not are my favorites)
It is a total failure. :D
I have reconnected wiht Facebook and now recommended to me:
- Reading National Geographic
- High-end cars,
- Gourmet Wines
- Classical music
- Holidays in very selected destinations.
Umm, I think I prefer the connection with FB. ;)
Educational use? Oooopps! None.
It works in a simple way - it asks you twenty questions, and based on your responses, recommends items from various categories (magazines, games, food, etc) that are suited for you. Amazon has been doing this for some time now, and Facebook and Twitter attempt to do something similar. They recommends friends.
I chose not to get very deep in the understanding of how recommendation engines work. At least not till I wrote this post. We are in Tuesday already; nearly half the week is over. So it is a good idea for me to finish this post, so that I can go back and study if my initial thoughts on this have any basis at all.
First few thoughts on Hunch, per se (I'll follow this up with a few thoughts on general recommendation engines):
- As some of my classmates have noted, requiring a Twitter or a Facebook OAuth is a put off. I think the primary reason for this is that you are not sure before you "get into Hunch" what it is about - and given that you possibly have a lot of personal information (at least in Facebook), you'd think twice before using FB to sign in. For the common user, who is concerned about privacy, it s not a very useful method - a direct signup feature would be useful.
- Hunch is very American. So if you are not an American, it doesn't work very well for you. These are cultural aspects which Hunch probably did not consider. If I state that I am a non-vegetarian, I get recommendations of steak-houses, but I do not eat red meat, for example. Further, when you look at the recommendations, even if I want to act on them, chances are - they are not available in my country. As a Mac user, Hunch made some very interesting recommendations regarding Mac products, however, music via iTunes is not sold in India. Apples knows why, we don't. There are no Apples stores in India, either.
- Next, the commercial angle; to an extent related to the regional and cultural aspect. The eCommerce sites are all US-specific - even for books. The most popular online bookstore, Flipkart.com, doesn't find a mention.
- Finally, the quality of the recommendations - they are far from relevant. Now, when I look at the results, I might say, hey, I find that interesting, but when I see a majority of the recommendations far from what I might really like, the algorithm fails.
Personally, if I were Amazon, I'd rather allow folks to link my goodreads.com site and extract information from there, than just map to who bought other books when they bought the book that I am viewing right now. Goodreads.com seems to work on a data set that has been created by the user, rather than the system. When I do a book-compatibility check with friends on goodreads.com, I'd like to believe I have a more closer understanding of recommendations from friends: I'd trust good reads.com more than I'd trust Amazon.com. The fact that someone bought some books is less valuable to me, than the fact that someone liked (or disliked) the books - explained with a reason.
Recommendations as they exist today, I conclude (as of this post), are slaves to the limitation of databases and the relationships that software architects can define, within the construct of a (software) program. Which means that soft-attributes that are more human and analogue, so to speak aren't a component of the mapping exercise.
In education, this takes on a different dimension. When learning online, It is quite possible (and easy) to recommend what content other students have liked, rated, which you might consider for your own consumption. The inherent problem in this is similar to (what I think) is called (Iterative) Economic Inequality: The rich getting richer. Allow me to explain this through an example:
iTunes has a built-in smart-playlist called "Top 25 Most Played." It is a simple algorithm: it takes the 25 songs with the most play-counts. If you use this play-list often enough, the playlist does not undergo any significant change. Simply because the playcounts of these 25 songs increase (because you keep listening to these songs). The only way to break the authority of this play list is to choose songs that you want to hear. Over a period of time, this "smart playlist" changes. I refer back to my earlier statement of this kind of recommendation is slavish to the limitation of the imagination and the construct of the program (or the programmer.)
When you add the variables of: learning styles (for those of us who still believe in it), variable pace of learning, environments in which we learn, cultural contexts in which we learn, and the economic situations which we prepare for, recommendations and similar gimmickry are fairly redundant.
In (final) conclusion, I think that recommendations, used the way they are today serve little (and often, negative) benefits to learning. There needs to be either a radical shift in the way we think about this model, or it needs to encompass several more variables (which sounds infeasible to me) for it to make any sense in learning and teaching.
The algorithms are not well posed. Should be much more delimited. But in this case does not matter because the objective is to collect data for commercial.
I think his purpose is to promote Appel products because this it is the same recommendation for all users.
This was a very interesting activity indeed. Most of the recommendations that Hunch made to me were inaccurate. I think this was because Hunch tried to profile me instead of trying to find out what I am interested in. However, the concept of a tool that recommends stuff is quite good.
A few questions that came to my mind when I was playing around with Hunch were:
Why does it ask irrelevant questions like, “Do you normally eat breakfast during the week?” to recommend new books to me? What kind of correlation is Hunch trying to draw? It is almost impossible to profile the kind of person I am, in order to recommend a book to me. However, you can always ask the kind of books I am interested in reading and areas that interest me and then help me discover different aspects of the area of interest. This, in fact, could have a lot of use in education.
In a learning set up, trying to profile a learner may not be a very good idea. This is because the purpose of the Internet should be to break the learners away from buckets and not put them in some. We could perhaps envision a tool that:
1. Allows the learners to select the categories in which they’d like recommendations. For example, learners may want to find Experts, Books, Curriculum, Communities, Research Groups, Discussion Groups, Competitions etc.
2. As a next step the system can ask them questions about their areas of interest, people they know who are interested in a similar thing, books they’ve read related to their area of interest, questions related to their level of proficiency in the field etc.. Based on this information the tool could make informed recommendations.This might turn out to be like an assisted Google search or could be like a mind map that the learners may expand upon based on the recommendations being made to them.
It recognizes that I am a technology and science fiction geek and try to sell me things that are related to it. But it doesn´t goes beyond.
In that sense, these algorithms could be used to identify the educational needs of students, especially in the professional environment to develop learning paths. Although it depends on how are the questions and the theory behind these questions.
But I did not like hunch.