Effects of Social Approval Votes on Search Performance

In this paper we develop a Social Information Retrieval model that incorporates different types of social approval votes for documents in a collection. The approvals reflect a level of endorsement by the community related to the collection and can be interpreted as trust, relevance, recommendation, and similar. They can come from perceived authorities, such as recognized experts and professional associations, or from aggregated opinions of a wider community, representing popular approval. We conducted preliminary experiments to incorporate social approval votes into search over 42,000 books by training neural networks. Using a set of 250 search topics with partial relevance judgments from non-expert users, we observe that the votes reflecting a broad appeal are most effective. We hypothesize that such sources of approval are more compatible with the general nature of the relevance judgments used in the experiments.