Unsupervised Detection and Promotion of Authoritative Domains for Medical Queries in Web Search

Medical or Health related search queries constitute a significant portion of the total number of queries searched everyday on the web. For health queries, the authenticity or authoritativeness of search results is of utmost importance besides relevance. So far, research in automatic detection of authoritative sources on the web has mainly focused on a) link structure based approaches and b) supervised approaches for predicting trustworthiness. However, the aforementioned approaches have some inherent limitations. For example, several content farm and low quality sites artificially boost their link-based authority rankings by forming a syndicate of highly interlinked domains and content which is algorithmically hard to detect. Moreover, the number of positively labeled training samples available for learning trustworthiness is also limited when compared to the size of the web. In this paper, we propose a novel unsupervised approach to detect and promote authoritative domains in health segment using click-through data. We argue that standard IR metrics such as NDCG are relevance-centric and hence are not suitable for evaluating authority. We propose a new authority-centric evaluation metric based on side-by-side judgment of results. Using real world search query sets, we evaluate our approach both quantitatively and qualitatively and show that it succeeds in significantly improving the authoritativeness of results when compared to a standard web ranking baseline. ∗Corresponding Author