Matching sets of parse trees for answering multi-sentence questions

Proceedings of Recent Advances in Natural Language Processing, pages 285–293, Hissar, Bulgaria, 7-13 September 2013. Matching sets of parse trees for answering multi-sentence questions Boris Galitsky Knowledge Trail Inc. San Jose USA bgalitsky@hotmail.com Dmitry Ilvovsky, Sergey Kuznetsov, Fedor Strok Higher School of Economics, Moscow Russia dilv_ru@yahoo.com; skuznetsov@hse.ru; fdr.strok@gmail.com Abstract The problem of answering multi-sentence questions is addressed in a number of products and services-related domains. A candidate set of answers, obtained by a keyword search, is re-ranked by matching the set of parse trees of an answer with that of the question. To do that, a graph representation and learning technique for parse structures for paragraphs of text have been developed. Parse Thicket (PT) as a set of syntactic parse trees augmented by a number of arcs for inter-sentence word-word relations such as co-reference and taxonomic relations is introduced. These arcs are also derived from other sources, including Speech Act and Rhetoric Structure theories. The proposed approach is subject to evaluation in the product search and recommendation domain, where search queries include multiple sentences. An open source plugin for SOLR is developed so that the proposed technology can be easily integrated with industrial search engines. 1 Introduction Modern search engines are not very good at tackling queries consisting of multiple sentences. They either find very similar documents, if they are available, or very dissimilar ones, so that search results are not very useful to the user. This is due to the fact that for multi-sentences queries it is rather hard to learn ranking based on user clicks, since the number of longer queries is practically unlimited. Hence we need a linguistic technology, which would rank candidate answers based on structural similarity between the question and the answer. In this study we build a graph-based representation for a paragraph of text so that we can track the structural difference between these paragraphs, taking into account not only parse trees, but the whole discourse as well. Paragraphs of text as queries appear in the search-based recommendation domains (Montaner et al., 2003; Bhasker and Srikumar 2010; Thorsten, 2012). Recommendation agents track user chats, user postings on blogs and forums, user comments on shopping sites, and suggest web documents and their snippets, relevant to a purchase decisions. To do that, these recommendation agents need to take portions of text, produce a search engine query, run it against a search engine API such as Bing or Yahoo, and filter out the search results which are determined to be irrelevant to a purchase decision. The last step is critical for a sensible functionality of a recommendation agent, and poor relevance would lead to a lost trust in the recommendation engine. Hence an accurate assessment of similarity between two portions of text is critical to a successful use of recommendation agents. Parse trees have become a standard form of representing the syntactic structures of sentences (Abney, 1991; Punyakanok et al., 2005; Domingos and Poon, 2009). In this study we will attempt to represent a linguistic structure of …

Content truncated to 500 words