Improving Web Search Ranking by Incorporating Structured Annotation of Queries

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 468–478, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Improving Web Search Ranking by Incorporating Structured Annotation of Queries* Xiao Ding1, Zhicheng Dou2, Bing Qin1, Ting Liu1, Ji-Rong Wen3 1Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China 2Microsoft Research Asia, Beijing 100190, China 3Renmin University of China, Beijing, China 1{xding, qinb, tliu}@ir.hit.edu.cn; 2zhichdou@microsoft.com; 3jirong.wen@gmail.com Abstract Web users are increasingly looking for structured data, such as lyrics, job, or recipes, using unstructured queries on the web. However, retrieving relevant results from such data is a challenging problem due to the unstructured language of the web queries. In this paper, we propose a method to improve web search ranking by detecting Structured Annotation of queries based on top search results. In a structured annotation, the original query is split into different units that are associated with semantic attributes in the corresponding domain. We evaluate our techniques using real world queries and achieve significant improvement. 1 Introduction Search engines are getting more sophisticated by utilizing information from multiple diverse sources. One such valuable source of information is structured and semi-structured data, which is not very difficult to access, owing to information extraction (Wong et al., 2009; Etzioni et al., 2008; Zhai and Liu 2006) and semantic web efforts.  *Work was done when the first author was visiting Microsoft Research Asia Driving the web search evolution are the user needs. Users usually have a template in mind when formulating queries to search for information. Agarwal et al., (2010) surveyed a search log of 15 million queries from a commercial search engine. They found that 90% of queries follow certain templates. For example, by issuing the query “taylor swift lyrics falling in love”, the users are actually seeking for the lyrics of the song “Mary's Song (oh my my my)” by artist Taylor Swift. The words “falling in love” are actually part of the lyrics they are searching for. However, some top search results are irrelevant to the query, although they contain all the query terms. For example, the first top search result shown in Figure 1(a) does not contain the required lyrics. It just contains the lyrics of another song of Taylor Swift, rather than the song that users are seeking. A possible way to solve the above ranking problem is to understand the underlying query structure. For example, after recognizing that “taylor swift” is an artist name and “falling in love” are part of the lyrics, we can improve the ranking by comparing the structured query with the corresponding structured data in documents (shown in Figure 1(b)). Some previous studies investigated how to extract structured information from user queries, such as query segmentation (Bergsma and Wang, 2007). The task of query segmentation is to separate the query words into 468 disjointed segments so that each segment maps to a semantic unit (Li et al., 2011). For example, the segmentation of the …

Content truncated to 500 words