Structure Cognizant Pseudo Relevance Feedback

International Joint Conference on Natural Language Processing, pages 982–986, Nagoya, Japan, 14-18 October 2013. Structure Cognizant Pseudo Relevance Feedback Arjun Atreya V, Yogesh Kakde, Pushpak Bhattacharyya, Ganesh Ramakrishnan CSE Department, IIT Bombay, Mumbai {arjun,pb,ganesh}@cse.iitb.ac.in,yrkakde@gmail.com Abstract We propose a structure cognizant frame- work for pseudo relevance feedback (PRF). This has an application, for ex- ample, in selecting expansion terms for general search from subsets such as Wikipedia, wherein documents typically have a minimally fixed set of fields, viz., Title, Body, Infobox and Categories. In existing approaches to PRF based expan- sion, weights of expansion terms do not depend on their field(s) of origin. This, we feel, is a weakness of current PRF ap- proaches. We propose a per field EM for- mulation for finding the importance of the expansion terms, in line with traditional PRF. However, the final weight of an ex- pansion term is found by weighting these importance based on whether the term be- longs to the title, the body, the infobox or the category field(s). In our experi- ments with four languages, viz., English, Spanish, Finnish and Hindi, we find that this structure-aware PRF yields a 2% to 30% improvement in performance (MAP) over the vanilla PRF. We conduct ablation tests to evaluate the importance of vari- ous fields. As expected, results from these tests emphasize the importance of fields in the order of title, body, categories and in- fobox. 1 Introduction The ruling paradigm for Information retrieval (IR) (Manning et al., 2009) is Pseudo Relevance feed- back (PRF). In PRF, an assumption is made that the top retrieved documents are relevant to the query for picking expansion terms. Zhai and Laf- ferty (2001) show that using pseudo relevance feedback on monolingual retrieval improves the overall result considerably over the retrieval with- out PRF. In case of retrieval for languages with lit- tle web content, Chinnakotla et al., (2010) show that taking help of another language to expand query helps in better performance. The motivation for our work is as follows. Ev- ery document in the web collection has certain structure associated with it viz., title, body, links, etc. Each of these fields has different level of importance in the document. For instance, docu- ment title broadly describes the whole document, whereas the body of the document contains the de- tails. Content in these fields have different scales of contribution in uniquely representing that doc- ument in the collection. Hence it is important to consider the structure of a document while extract- ing expansion terms from it. Structure based PRF, of course, draws on the basic theory of PRF as in Zhai and Lafferty (2001), which is based on expectation maximiza- tion (EM). We formulate a per field EM to get the weights of expansion terms and subsequently take their weighted sum in a spirit similar to mixture models. 2 Related Work Approaches based on the use of external resources like wordnet for query expansion, though ex- tensively studied, have been eventually dropped (Gong et al., 2005; Qiu …

Content truncated to 500 words