Paper 2010/245 Original: eprint.iacr.org/2010/245 ↗

Towards a Theory of Trust Based Collaborative Search

Abstract

Trust Based Collaborative Search is an interactive metasearch engine, presenting the user with clusters of results, based not only on the similarity of content, but also on the similarity of the recommending agents. The theory presented here is broad enough to cover search, browsing, recommendations, demographic profiling, and consumer targeting. We use the term search as an example. We developed a novel general trust theory. In this context, as a special case, we equate trust between agents with the similarity between their search-behaviors. The theory suggests that clusters should be close to maximal similarity within a tolerance dictated by the amount of uncertainty about the vectors of probabilities of attributes representing queries, pages and agents. In addition, we give a new theoretical analysis of clustering tolerances, enabling more judicial decisions about optimal tolerances. Specifically, we show that tolerances should at least be divided by a constant>1 as we descend from one layer in the hierarchical clustering to the next. We also show a promising connection between collaborative search and cryptography: A query plays the role of a cryptogram, the search engine is the cryptanalyst, and the user's intention is the cleartext. Shannon's unicity distance is the length of the search. It is needed to quantify the clustering-tolerance.