The downside of collaborative filtering

Yesterday’s San Jose Mercury News carried an article on recommending systems and collaborative filtering. (You can get the article, actually from the Los Angeles Times, here although you may need to register.) The article pointed out some downsides to collaborative filtering. I wondered which of these downsides would transfer to the scholarly world, and how.

The first downside is dumbness. I’ve noticed this for years. I use Amazon almost exclusively to buy things for other people – I usually know what I want to buy for them, and value someone else shipping something for me (as opposed to getting to UPS or the post office myself – something I hardly ever do). What I do not value are the “recommendations.” The recommendations, in my case, are based on occasional purchases for a variety of people. Yes, I know I can improve my recommendations, but that’s never my task or my interest. I want to get in and get out.

The article in the Merc cites another kind of dumbness, outlined in the following scenarios; you are going someplace (let’s say Italy) and you buy some travel guidebooks. You take your trip and come back, only to be pelted with more Italian guidebook recommendations.

I think the first kind of dumbness (recommendations, but not for me) wouldn’t necessarily transfer to the scholarly world since most people don’t do a lot of research on behalf of others. The second kind of dumbness (I’m so over that, thank you very much) could apply, particularly for students who have new research topics imposed upon them frequently. Wrong recommendations can be treated by working with the collaborative filtering software – what I never have time to do with Amazon, editing preferences and expressing lack of interest in Italy. In the case of “I’m so over that,” or the student with many topics, having multiple profiles within an account may be a way to go.

The other downside of collaborative filtering, narrowness, could have implications for scholarship that I think are worth exploring.

In the marketplace, collaborative filtering provides a service to consumers by helping to narrow millions of choices down to a few. A list of a few well-targeted recommendations also helps to increase sales, and also helps consumers find items they would not have otherwise found. I’ve had this experience on Netflix, finding and enjoying movies I’d never heard of. But narrowing down comes at a price, described in the article as “society [balkanizing] into groups with obscure interests.” While groups with obscure interests sounds, on the face of it, like scholarship itself, there is a definite negative aspect to this. What scholar wants to only read one side of an argument, one interpretation of events, one school of thought? How many scholarly careers have been changed by serendipity? This could come from a conversation at a cocktail party, or could also stem a confusing moment in the catalog, a wrong turn in the stacks. How do we interject chance and opportunity into recommending systems? Maybe recommending systems will make cocktail parties even more important.

In any case, all this has me thinking. Time to start planning that trip to Italy.

One Comment on “The downside of collaborative filtering”

  1. I totally agree about the danger of receiving only what we want to hear/read/see. The essence of the scholarly world is to try to arrive at something resembling at least a part of truth, so if we aren’t exposed to things we don’t like or aren’t interested in we have blocked out all chances for kismet to intervene and shake us up. It might make for a very tidy world, but it is a very small one.

Comments are closed.