User Behaviors Lend a Helping Hand: Learning Paraphrase Query Patterns from Search Log Sessions

Proceedings of COLING 2012: Technical Papers, pages 3137–3152, COLING 2012, Mumbai, December 2012. User Behaviors Lend a Helping Hand: Learning Paraphrase Query Patterns from Search Log Sessions Shiqi Zhao1,2 Hai f eng Wang1 Ting Liu2∗ (1) Baidu, Beijing, China (2) School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China zhaoshiqi@baidu.com, wanghaifeng@baidu.com, tliu@ir.hit.edu.cn ABSTRACT Search log sessions contain a large number of paraphrases contributed by users during query rewriting. However, it is a big challenge to distinguish paraphrases from the simply related queries in the sessions. This paper addresses this problem by making innovative use of user behavior information embodied in query sessions. Specifically, we learn paraphrase patterns from the search log sessions with a classification framework, in which three types of user behavior features are exploited besides the conventional features. We evaluate the method using a query log of a commercial search engine. Experimental results demonstrate the effectiveness of our method, especially the significant contribution of the user behavior features. We extract over 250,000 pairs of paraphrase patterns from the used search log, with a precision over 76%. TITLE AND ABSTRACT IN ANOTHER LANGUAGE (CHINESE) Ä Ä Äu u u^ ^ ^r r r1 1 1• • •A A A� � �l l l| | |¢ ¢ ¢¬ ¬ ¬{ { {¥ ¥ ¥� � �÷ ÷ ÷E E Eã ã ã� � �Î Î Î� � �† † † |¢Ú™^r3|¢ž²~¬é�Î?1ÓÂU�§±¼�•Ð�|¢(J§Ïd |¢F“��ά{¥•¹Œþ�Eã]�",�˜‡(J�¯K´XÛ«©�ά{¥ =��δEã'X§=�K==´ŠÂƒ'�®"�©ÏL•Ð/|^�ά{¥Û¹ �^r1•A�5)ûù˜¯K"äN/§·‚Äu©a�•{l�ά{]�¥�÷E ã�Î�†"3©a�.¥§·‚Ø=¦^�DÚ�Eã£OA�§„}Á�n«^r1 •A�"·‚¦^˜‡û’|¢Ú™�^r|¢F“5µ��©JÑ�•{"µ�(Jy ²��©•{�k�5§cÙ´^r1•A�å��²wŠ^"·‚l¤¦^�^rF“ ¥�Ä�Ñ�‡L250,000é�Eã�†§ÙO(LJL76%" KEYWORDS: paraphrase, pattern, query, search log. KEYWORDS IN CHINESE: Eã, �†, �Î, |¢F“. ∗ Corresponding author 3137 1 Introduction Paraphrases have been shown to be useful in plenty of areas, such as machine translation (MT) (Callison-Burch et al., 2006; Madnani et al., 2007; Marton et al., 2009), question answering (QA) (Lin and Pantel, 2001; Ravichandran and Hovy, 2002; Duboue and Chu-Carroll, 2006; Riezler et al., 2007), and web search (Zukerman and Raskutti, 2002). In particular, the capability of paraphrasing is essential in web search, since in many cases the user queries need to be paraphrased so as to improve the quality of the search results. This paper focuses on learning paraphrase query patterns from search log sessions, which could be useful in various applications, especially in query paraphrasing. Although search log sessions have been extensively exploited for mining related queries, this is the first work, as far as we know, to learn paraphrases from this data source. Mining paraphrase query patterns and related queries are both useful for search engines, but in different aspects. On the one hand, related queries can be used for query suggestion and recommendation, using which the users can extend their search interest and get some related information. However, the related queries are not suitable for direct query rewriting with the purpose of retrieving more and better results exactly reflecting the user’s requirement, since the related queries often have different meanings from the original user query. On the other hand, the paraphrase query patterns are mined for query paraphrasing, which is to directly rewrite user …

Content truncated to 500 words