A probabilistic graphical model for learning as search

This work describes ongoing research on web information retrieval using queries which require prolonged browser interactions in exploring, revising, resubmitting, and studying results produced by the search engine. Over the last two decades or so, web search has improved significantly in a lot of areas, but has not made much progress at the difficult end of the online search spectrum — one that is precisely characterized by queries with complex information needs. Such queries, we call learning queries, require extended browser interaction by the user for satisfactory results. The main goal of the research is to address the difficulty of learning queries by lessening browser interactions through the implementation of a probabilistic graphical model. The expected outcome is an improvement in relevance, inspired by automatic discovery of collaborative search structures in conjunction with leveraging current search engine capabilities. To attain the project goal, we design and implement a browser extension that collects post-query browsing details based on browser interactions by users. The post-query browsing details are used to create a database of query-indexed trees for validating the model proposed. The work presented here describes the design and implementation of browser extensions and the project methodology.