一种基于相关反馈的跨语言信息检索查询翻译优化技术研究

A Relevance Feedback Based Query Translation Enhancement Technique in Cross Language Information Retrieval

本文提出了一种新的基于相关反馈的跨语言信息检索查询翻译优化技术,就实现该技术的关键步骤＂估计检索词在相关文献集合中的翻译概率＂设计了4种不同的算法,并通过伪相关反馈实验比较了这4种算法,验证了查询翻译优化技术的有效性.实验结果显示,4种翻译优化算法都能够提高检索结果的精度,其中基于词对齐的翻译算法相对更优越.此外,查询式的长度和检索主题的特征对不同查询翻译优化算法产生着不同程度的影响.

This paper proposes a new cross language relevance feedback technique called query translation enhancement （TE） , and studies four algorithms for ＂estimating query terms ＇ translation probabilities based on obtained relevant document pair set＂ , which is an important step in query translation enhancement. We conduct a series of pseudo relevance feedback experiments to compare the four algorithms and validate the effectiveness of query translation enhancement. Experiment results show that all four algorithms can raise the effectiveness of retrieval results, but the algorithm based on word alignment technique has clear superiority. In addition, we find that query length and retrieval topics can greatly affect various query translation enhancement approaches.