Context-Dependent Term Relations for Information Retrieval

Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP 2006), pages 551–559, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Context-Dependent Term Relations for Information Retrieval Jing Bai Jian-Yun Nie Guihong Cao DIRO, University of Montreal CP. 6128, succ. Centre-ville, Montreal, Quebec H3C 3J7, Canada {baijing,nie,caogui}@iro.umontreal.ca Abstract Co-occurrence analysis has been used to determine related words or terms in many NLP-related applications such as query expansion in Information Retrieval (IR). However, related words are usually determined with respect to a single word, without relevant information for its application context. For example, the word “programming” may be considered to be strongly related to “Java”, and applied inappropriately to expand a query on “Java travel”. To solve this problem, we propose to add another context word in the relation to specify the appropriate context of the relation, leading to term relations of the form “(Java, travel) → Indonesia”. The extracted relations are used for query expansion in IR. Our experiments on several TREC collections show that this new type of context-dependent relations performs much better than the traditional co-occurrence relations. 1. Introduction A query usually is a poor expression of an information need. This is not only due to its short length (usually a few words), but also due to the inability of users to provide the best terms to describe their information need. At best, one can expect that some, but not all, relevant terms are used in the query. Query expansion thus aims to improve query expression by adding related terms to the query. However, the effect of query expansion is strongly determined by the term relations used (Peat and Willett, 1991). For example, even if “programming” is strongly related to “Java”, if this relation is used to expand a query on “Java travel”, the retrieval result will likely deteriorate because the irrelevant term “programming” is introduced, leading to the retrieval of irrelevant documents about “programming”. A number of attempts have been made to deal with the problem of selecting appropriate expansion terms. For example, Wordnet has been used in (Voorhees, 1994) to determine the expansion terms. However, the experiments did not show improvement on retrieval effectiveness. Many experiments have been carried out using associative relations extracted from term co- occurrences; but they showed variable results (Peat and Willett, 1991). In (Qiu and Frei, 1993), it is observed that one of the reasons is that one tried to determine expansion terms according to each original query term separately, which may introduce much noise. Therefore, they proposed to determine the expansion terms by summing up the relations of a candidate expansion term to each of the query terms. In so doing, a candidate expansion term is preferred if it has a strong relationship with many of the query terms. However, it is still difficult to prevent the expansion process from adding “programming” to a query on “Java travel” because of its very strong relation with “Java”. The approach used in (Qiu and Frei, 1993) indeed tries to …

Content truncated to 500 words