Syntactic Parsing of Web Queries with Question Intent

Proceedings of NAACL-HLT 2016, pages 670–680, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Syntactic Parsing of Web Queries with Question Intent Yuval Pinter1, Roi Reichart1,2, and Idan Szpektor1 1Yahoo Research, Haifa 31905, Israel, {yuvalp, roiri, idan}@yahoo-inc.com 2Faculty of Industrial Engineering and Management, Technion, IIT Abstract Accurate automatic processing of Web queries is im- portant for high-quality information retrieval from the Web. While the syntactic structure of a large portion of these queries is trivial, the structure of queries with question intent is much richer. In this paper we therefore address the task of statistical syntactic parsing of such queries. We first show that the standard dependency grammar does not ac- count for the full range of syntactic structures man- ifested by queries with question intent. To allevi- ate this issue we extend the dependency grammar to account for segments – independent syntactic units within a potentially larger syntactic structure. We then propose two distant supervision approaches for the task. Both algorithms do not require manually parsed queries for training. Instead, they are trained on millions of (query, page title) pairs from the Com- munity Question Answering (CQA) domain, where the CQA page was clicked by the user who initiated the query in a search engine. Experiments on a new treebank1 consisting of 5,000 Web queries from the CQA domain, manually parsed using the proposed grammar, show that our algorithms outperform alter- native approaches trained on various sources: tens of thousands of manually parsed OntoNotes sentences, millions of unlabeled CQA queries and thousands of manually segmented CQA queries. 1 Introduction As the World Wide Web grows in volume, it encompasses ever-increasing amounts of text. A major gateway to this invaluable resource is Web queries which users compose to guide a search engine in retrieving the information they desire to inspect. Automatic processing of Web queries is therefore of utmost importance. 1This treebank has been released via Yahoo’s Webscope pro- gram: webscope.sandbox.yahoo.com. Previous research (Bergsma and Wang, 2007; Barr et al., 2008) suggested that many Web queries are trivial in structure, usually embodying entity lookup, e.g. “frozen” or “condos in NY”. However, with the increasing popular- ity of Community Question Answering (CQA) sites, such as Yahoo Answers, StackOverflow and social QA forums, more Web queries encompass information needs related to questions that these sites can answer. We found that this subcategory of queries, which we call CQA queries (following Liu et al. (2011; Carmel et al. (2014)), exhibits a wide range of structures. This suggests that the process- ing of such queries, which constitute ∼10% of all queries issued to search engines (White et al., 2015), may benefit from syntactic analysis (Tsur et al., 2016). Recent progress in statistical parsing (Choi et al., 2015) has resulted in models that are both fast, pars- ing several hundred sentences per second, and accurate. These parsers, however, still suffer from the problem of domain adaptation (McClosky et al., 2010), excelling mostly when their training and test domains are similar. This problem …

Content truncated to 500 words