Semantic Tagging of Web Search Queries

Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pages 861–869, Suntec, Singapore, 2-7 August 2009. c⃝2009 ACL and AFNLP Semantic Tagging of Web Search Queries Mehdi Manshadi Xiao Li University of Rochester Microsoft Research Rochester, NY Redmond, WA mehdih@cs.rochester.edu xiaol@microsoft.com Abstract We present a novel approach to parse web search queries for the purpose of automatic tagging of the queries. We will define a set of probabilistic context-free rules, which generates bags (i.e. multi-sets) of words. Us- ing this new type of rule in combination with the traditional probabilistic phrase structure rules, we define a hybrid grammar, which treats each search query as a bag of chunks (i.e. phrases). A hybrid probabilistic parser is used to parse the queries. In order to take contextual information into account, a discriminative model is used on top of the parser to re-rank the n-best parse trees gen- erated by the parser. Experiments show that our approach outperforms a basic model, which is based on Conditional Random Fields. 1 Introduction Understanding users’ intent from web search queries is an important step in designing an intel- ligent search engine. While it remains a chal- lenge to have a scientific definition of ''intent'', many efforts have been devoted to automatically mapping queries into different domains i.e. topi- cal classes such as product, job and travel (Broder et al. 2007; Li et al. 2008). This work goes beyond query-level classification. We as- sume that the queries are already classified into the correct domain and investigate the problem of semantic tagging at the word level, which is to assign a label from a set of pre-defined semantic labels (specific to the domain) to every word in the query. For example, a search query in the product domain can be tagged as: cheap garmin streetpilot c340 gps | | | | | SortOrder Brand Model Model Type Many specialized search engines build their in- dexes directly from relational databases, which contain highly structured information. Given a query tagged with the semantic labels, a search engine is able to compare the values of semantic labels in the query (e.g., Brand = “garmin”) with its counterpart values in documents, thereby pro- viding users with more relevant search results. Despite this importance, there has been rela- tively little published work on semantic tagging of web search queries. Allan and Raghavan (2002) and Barr et al. (2008) study the linguistic structure of queries by performing part-of-speech tagging. Pasca et al. (2007) use queries as a source of knowledge for extracting prominent attributes for semantic concepts. On the other hand, there has been much work on extracting structured information from larger text segments, such as addresses (Kushmerick 2001), bibliographic citations (McCallum et al. 1999), and classified advertisements (Grenager et al. 2005), among many others. The most widely used approaches to these problems have been sequential models including hidden Markov models (HMMs), maximum entropy Markov mod- els (MEMMs) (Mccallum 2000), and conditional random fields (CRFs) (Lafferty et …

Content truncated to 500 words