Recommender Systems and Collaborative Filtering

07 May 2022 16:31

Yet Another Inadequate Placeholder

An ubiquitous application of data mining.

One of the things I find technically interesting here is the large role for factor models. This makes all kinds of sense, starting with the potential simplicity of the technique --- you can (I have!) teach it to undergrads whose prior training ends at linear regression. (Needless to say industrial-strength implementations are a bit more involved, but the ideas are the same.) But, unlike many other applications of factor models, here people seem un-interested in "reifying" the factors, and indeed uninterested in interpreting them.

--- Some day, an essay will be written about how we got from the collectivist optimism of "collaborative filtering" (as in Shardanand and Maes 1995) to being followed around the Internet by intrusive and inappropriate ads. This is not that day.

See also: Collective Cognition; Factor Models