A Study on Effectiveness of Syntactic Relationship in Dependence Retrieval Model

A Study on Effectiveness of Syntactic Relationship in Dependence Re- trieval Model Fan Ding1,2 1: Graduate University, Chinese Academy of Sciences Beijing, 100080, China dingfan@ict.ac.cn Bin Wang2 2: Institute of Computing Technology, Chinese Academy of Sciences Beijing, 100080, China wangbin@ict.ac.cn Abstract To relax the Term Independence Assump- tion, Term Dependency is introduced and it has improved retrieval precision dramati- cally. There are two kinds of term depend- encies, one is defined by term proximity, and the other is defined by linguistic de- pendencies. In this paper, we take a com- parative study to re-examine these two kinds of term dependencies in dependence language model framework. Syntactic rela- tionships, derived from a dependency parser, Minipar, are used as linguistic term dependencies. Our study shows: 1) Lin- guistic dependencies get a better result than term proximity. 2) Dependence retrieval model achieves more improvement in sen- tence-based verbose queries than keyword- based short queries. 1 Introduction For the sake of computational simplicity, Term Independence Assumption (TIA) is widely used in most retrieval models. It states that terms are statis- tically independent from each other. Though un- reasonable, TIA did not cause very bad perform- ance. However, relaxing the assumption by adding term dependencies into the retrieval model is still a basic IR problem. Relaxing TIA is not easy be- cause improperly relaxing may introduce much noisy information which will hurt the final per- formance. Defining the term dependency is the first step in dependence retrieval model. Two re- search directions are taken to define the term de- pendency. The first is to treat term dependencies as term proximity, for example, the Bi-gram Model (F. Song and W. B. Croft, 1999) and Markov Ran- dom Field Model (D. Metzler and W. B. Croft, 2005) in language model. The second direction is to derive term dependencies by using some linguis- tic structures, such as POS block (Lioma C. and Ounis I., 2007) or Noun/Verb Phrase (Mitra et al., 1997), Maximum Spanning Tree (C. J. van Rijsbergen, 1979) and Linkage Model (Gao et al., 2004) etc. Though linguistic information is intensively used in QA (Question Answering) and IE (Infor- mation Extraction) task, it is seldom used in docu- ment retrieval (T. Brants, 2004). In document re- trieval, how effective linguistic dependencies would be compared with term proximity still needs to be explored thoroughly. In this paper, we use syntactic relationships de- rived by a popular dependency parser, Minipar (D. Lin, 1998), as linguistic dependencies. Minipar is a broad-coverage parser for the English language. It represents the grammar as a network of nodes and links, where the nodes represent grammatical cate- gories and the links represent types of dependency. We extract the dependencies between content words as term dependencies. To systematically compare term proximity with syntactic dependencies, we study the dependence retrieval models in language model framework and present a smooth-based dependence language model (SDLM). It can incorporate these two kinds of term dependencies. The experiments in TREC collections show that SDLM with syntactic rela- tionships achieves …

Content truncated to 500 words