A Guide to Recommender Systems
We're running a special series on recommendation technologies and in this post we look at the different approaches - including a look at how Amazon and Google use recommendations. The Wikipedia entry defines "recommender systems" as "a specific type of information filtering (IF) technique that attempts to present information items (movies, music, books, news, images, web pages, etc.) that are likely of interest to the user." That entry goes on to note that recommendations are generally based on an "information item (the content-based approach) or the user's social environment (the collaborative filtering approach)." We think there's also a personalization approach, which Google in particular is focused on. We explore some of these concepts below.
In a recent post, Xavier Vespa of the blog HyveUp analyzed 3 different approaches to recommendation engines on the Web. He identified that Pandora used "deep structural analysis of an item" for its recommendations, Strands focused on "intensive social behavior analysis around an item" and Aggregate Knowledge did "structural analysis of an item, paired with behavioral analysis around the item".
A couple of years ago, Alex Iskold outlined what he saw as the 4 main approaches to recommendations:
- Personalized recommendation - recommend things based on the individual's past behavior
- Social recommendation - recommend things based on the past behavior of similar users
- Item recommendation - recommend things based on the item itself
- A combination of the three approaches above
Amazon: King of Recommendations
In that post, Alex analyzed what Amazon.com - probably the canonical example of recommendations technology on the Web - used to power it's recommendations. Unsurprisingly, he found that Amazon used all 3 approaches (personalized, social and item). Amazon's system is very sophisticated, but at heart all of its recommendations "are based on individual behavior, plus either the item itself or behavior of other people on Amazon." What's more, the aim of it all is to get you to add more things to your shopping cart.
As Xavier identified, other newer Internet companies have tended to focus on specific methods of recommendation. For Pandora, it is a deeper analysis of the item (using its "gene" theory); Strands has taken a boatload of VC money to try and become the number 1 social recommendations provider on the planet; and Aggregate Knowledge is taking more of a behavioral approach.
Google: Focus on Personalized Recommendations
The most successful Internet company of this era has without a doubt been Google. It too has been using recommendation technologies to improve its core search product. There are two ways that Google does this:
- Google customizes your search results "when possible" based on your location and/or recent search activity;
- When you're signed in to your Google Account, you "may see even more relevant, useful results based on your web history."
So Google is using both your location and your personal search history to make its search results supposedly stronger. This is very much the 'personalized recommendation' approach - and indeed personalization has been a buzzword for Google in recent years. However, the two other types of recommendation are also present in Google's core search product:
- Google's search algorithm PageRank is basically dependent on social recommendations - i.e. who links to a webpage;
- Google also does item recommendations with its "Did you mean" feature.
There are surely other ways recommendations technologies are being deployed in Google search - not to mention the range of other products Google has. Google News, its start page iGoogle, and its ecommerce site Froogle all have recommendation features.
ReadWriteWeb Resources for Recommendation Technologies
Let us know what types of recommendation technologies other companies are using. We also invite you to explore using our custom ReadWriteWeb Resources:
- RWW Recommendation Industry Feed Favorites OPML file (save link)
- RWW Recommendation Industry Feeds - Best of Feed (copy and paste to your reader)
- Click to preview the above feeds before subscribing (pop-up window)
- RWW Recommendation Site Search (Visit and Bookmark)
Share
Comments
Subscribe to comments for this post OR Subscribe to comments for all ReadWriteWeb posts
I would like to reproduce your paper in the blog http://europa-eu-audience.typepad.com
of course with source indication
Regards
Jacques
For book, movies and such there is also a social recommendation system, where websites recommend items based on bloggers recommendation.
Lets say 100 bloggers recommended some book, the recommendation aggregator could post it on his website.
We are using the "combination approach" at theSUGGESTR.com to local search recommendations. We look at your previous ratings, ratings of similar users and then factor in similar item (businesses in our case) comparisons as well. We use a genetic system to continually tweak the weighting of our 3 inputs. As we've gathered more data the weights that yield the best results have change significantly. We also plan on experimenting with some new approaches once (if) we get things moving.
I really believe that one of the next waves of sites will be those that can accurately personalize the web while not becoming too narrow. The biggest challenge for recommender systems is data. You need lots of data for accuracy and for new startups the data isn't always there. Sites like Google Base and emerging semantically marked up (microformatted sites especially) may make this easier - but there are still hurdles (performance / legal ambiguity).
The famous movie recommendation website www.moviepilot.de implemented a recomendation service based on ratings for movies.
Celebros (www.celebros.com) has an interesting recommendation engine based on numerous technologies including heuristics and language taxonomy
Chris Wright has an interesting take on music recommendation systems here.
Essentially his point is that there's a lot of music fans (like myself) who don't want to "discover" music that sounds like what we're already listening to, or what our peers are listening to. We want to hear the new, the creative, the different -- all qualities that digital recommendation systems are completely incapable of assessing.
In other words, the geeky kid that works at your local used record store is still a heckuva good resource for discovering fresh noise. Long may he live!
I am experimenting with short term profiles for personalizing search results from any search engine. You can read more about my work here: http://cs.fit.edu/~etuleu/seniorDesign/index.html
Please let me know what you think!
Thank you!
movies and such there is also a social recommendation system
SwingVine uses user-generated recommendations in a different way than listed above. There, users describe related recommendations with a story such as "People who like x will like y because so and so..."
BTW, Marcello, SwingVine is basically a way for those "geeky" kids at the local record store to share their knowledge as part of a larger community while providing greater convenience because all that knowledge is online, searchable, etc.
Reblog of this post with implications for fashion ecommerce and social shopping:
http://www.stylehop.com/blog/2009/01/27/recommender-systems-and-why-fashion-social-shopping-hasnt-worked/
This is a nice summary of approaches. One thing to add is what are the real problems that recommender systems encounter.
Large, sparse data sets is the classic issue. The other two which have not been sufficiently addressed yet are changing data (systems are biased towards the old and have difficulty showing new) and changing user preferences (today I have a particular intention). These are real issues for these systems and perhaps how well they are solved will determine their ultimate growth and use.
I'd like to get everyone's opinion on this, but I personally think recommendation systems are useless. I am an Amazon.com freak, I use it to browse books and navigate through different indices and subject categories books (and products) are placed under. There is actually the famous "Customers Who Bought This Item Also Bought" feature which is basically a God given gift. I love and live by it also.
Question: My assumption is that only the last of the examples I listed falls within a recommendation system. Navigating through indexes and subject categories of a book is a strong reason I use Amazon and these features fall more along the lines of good old organization and tagging, right?
Reading your article helped me a lot
Thanks!