Social Filtering and Social Reality

In this paper, we argue that most current social information ltering approaches may bene t from more seriously taking into account the peculiarities of human cognition and human social behavior since current approaches only consider de-contextualized ratings. Social ltering systems exploit ratings provided by users in order to compute recommendations for other users. Typically, these ratings are detached from the situation and the social embedding in which they have been provided. Recent research on human cognition and behavior suggests that actions should not be viewed in isolation from the situation in which they occur (thus, the term \situated actions"). Accounting for the situation and the social embedding requires support for exploiting the situation rather than abstracting away the situation. In respect to exploiting the social embedding of ratings, we discuss the need for two related basic research directions. First, a self-organizing network of users trusting each other should be explored as a basis for true socially embedded ltering. Second, the suitability of collaborative ltering techniques as a tool for maintaining the focus of Usenet discussion groups by exposing spam and other clear o-topic postings should be investigated