Synthèse

So far, I’ve liked everything I’ve seen and read by Nicholas Carr (author of “The Big Switch: rewiring the world from Edison to Google”). I was interested and challenged by his recent article in The Atlantic “Is Google Making us Stupid“.  His basic thesis is that the information overload that results from the availability of huge amounts of data from search engines is making us unable to read closely and think deeply.

As part of the five-year research program, [scholars from the university of London] examined computer logs documenting the behavior of visitors to two popular research sites, one operated by the British Library and one by a U.K. educational consortium, that provide access to journal articles, e-books, and other sources of written information. They found that people using the sites exhibited “a form of skimming activity,” hopping from one source to another and rarely returning to any source they’d already visited. They typically read no more than one or two pages of an article or book before they would “bounce” out to another site.

[See this report for more details.]  He claims that this persistent kind of behaviour has rewired our brains, thanks to its recently discovered plasticity in adulthood.

I’m wondering whether this observation bodes well for Recommender Systems.  One reason that is often given for building recommenders, particularly in the context of a digital library, is that it is supposed to help address the information overload problem. However, one can easily argue that the converse is true.

Given that we are overwhelmed by mountains of information and that high precision search helps us find the needle in the haystack, the point of a recommender system is to help us find more needles that we didn’t know we were looking for but are (potentially) just as useful.

Thus, a recommender actually adds to the information overload problem and thus exacerbates the attention deficit problem that Carr complains about in his article.  Says Carr:

[Google's] assumption that we’d all “be better off” if our brains were supplemented, or even replaced, by an artificial intelligence is unsettling.

So I’m discouraged that Carr doesn’t think that AI should help people find stuff, but perhaps he’s right – perhaps we should focus on what’s in front of us instead of fretting about whether we’ve reached 100% recall.  Of course, that won’t help you get any grants if the reviewers find you’ve missed something….

A recent post by David Rothman points us to MiSearch, the “Adaptive PubMed Search Tool”. There is a paper about this system in the journal Bioinformatics.

MiSearch uses implicit profiling in a way that bears some resemblance to what I think an article recommender should do.  However, instead of “recommending”, it uses a content-based algorithm to re-ranking search results. From the help file:

MiSearch uses a classification algorithm based on MeSH term, substance names and author names associated with citations. Two sets are defined. One is the set of articles you have previously clicked on to view. The other is all of PubMed. For each citation in the retrieval set, the algorithm calculates the likelihood that the citation is a member of these two sets. Article having the highest likelihood of belonging to the set of articles you have viewed are ranked at the top of the list.

And from the paper mentioned above:

MiSearch is using query expansion with probabilistic weighting of terms derived from the implicitly defined relevant document set.

My concern with re-ranking search results in this way is that it isn’t clear why the re-ranking has happened in the way it has or, indeed, how the user can control it.  Explanations and user-interfaces that provide them are key to the usability of such systems, I think.