查询专指度对检索效果的影响研究

The Impacts of Query Specificity on Information Retrieval

【目的】针对不同查询专指度语句的检索效果进行全面分析,为改善搜索引擎性能、提高用户检索体验提供借鉴。【方法】基于TREC Web Track查询语句,人工构建查询专指度标注集,选用语言模型狄利克雷平滑、语言模型线性插值平滑和BM25三种模型,以常用的信息检索评价指标为基准,探讨查询专指度强弱对检索效果在不同层次上的影响。【结果】在最靠前的几条检索结果中,强弱专指度查询语句的检索效果差异最大,强专指度的检索效果要明显好于弱专指度。【局限】仅在TREC数据集上进行实验测试,还需在其他数据集上进一步检验。【结论】搜索引擎在专指度这一维度下,应重点关注最靠前的几条检索结果的准确性,以此为切入点改善检索模型。

[Objective] This paper analyzes the impacts of query specificity on the effectiveness of information retrieval systems, aiming to improve the performance of search engine and user experience. [Methods] First, we manually constructed a labeling set for queries from the TREC Web Track. Second, we adopted the Dirichlet language model, linear interpolation language model and BM25 model to examine each query＇s performance. Finally, we used the average information retrieval evaluation index as the benchmark to explore the impacts of query specificity. [Results] For the highest-ranked results, the queries with narrower specificity had better retrieval performance than their boarder counterparts. [Limitations] The proposed method was only examined with data provided by TREC. More studies were needed to evaluate its performance with other data sets. [Conclusions] Search engines should focus on the precision of the highest ranked results, and then modify their retrieval model accordingly.