当前位置: X-MOL 学术J. Intell. Inf. Syst. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
An implicit aspect modelling framework for diversity focused query expansion
Journal of Intelligent Information Systems ( IF 2.3 ) Pub Date : 2019-11-29 , DOI: 10.1007/s10844-019-00581-w
Rahul E. Dev , Vidhya Balasubramanian

Diversified Query Expansion aims to present the user with a diverse list of query expansions so as to better communicate their intent to the retrieval system. Current diversified expansion techniques either make use of external knowledge sources to explicitly model the various aspects and their relationships underlying the user query or implicitly model query aspects. However these techniques assume query aspects to be independent of each other. We propose a unified framework that produces diversified query expansions in a completely implicit manner while also considering the relationships between query aspects. In particular, the framework identifies query aspects and their relationships by making use of the semantic properties of context phrases that occur within the top-ranked retrieved documents for the supplied user query, and maps them onto a Mutating Markov Chain model to generate a diverse ordering of query aspects. We test our framework against a set of ambiguous and faceted queries used in the NTCIR-12 IMine-2 Task and through an extensive empirical analysis, we show that our framework consistently outperforms existing implicit diversified query expansion algorithms. The utility of our algorithm truly comes up in the second set of experiments where we generate diversified query expansions for a retrieval engine indexing documents from specific scientific domains. Even in such a niche scenario our algorithm consistently provides robust results and performs better than other implicit approaches.

中文翻译:

一种以多样性为中心的查询扩展的隐式方面建模框架

Diversified Query Expansion 旨在向用户呈现多样化的查询扩展列表,以便更好地将他们的意图传达给检索系统。当前多样化的扩展技术要么利用外部知识源显式地对用户查询背后的各个方面及其关系进行建模,要么隐式地对查询方面进行建模。然而,这些技术假定查询方面彼此独立。我们提出了一个统一的框架,该框架以完全隐式的方式产生多样化的查询扩展,同时还考虑了查询方面之间的关系。特别是,该框架通过利用出现在提供的用户查询的排名靠前的检索文档中的上下文短语的语义属性来识别查询方面及其关系,并将它们映射到变异马尔可夫链模型以生成查询方面的不同排序。我们针对 NTCIR-12 IMine-2 任务中使用的一组模糊和分面查询测试我们的框架,并通过广泛的实证分析表明,我们的框架始终优于现有的隐式多元化查询扩展算法。我们算法的实用性真正体现在第二组实验中,我们为特定科学领域的检索引擎索引文档生成多样化的查询扩展。即使在这样的利基场景中,我们的算法也始终如一地提供稳健的结果并且比其他隐式方法表现得更好。我们针对 NTCIR-12 IMine-2 任务中使用的一组模糊和分面查询测试我们的框架,并通过广泛的实证分析表明,我们的框架始终优于现有的隐式多元化查询扩展算法。我们算法的实用性真正体现在第二组实验中,我们为特定科学领域的检索引擎索引文档生成多样化的查询扩展。即使在这样的利基场景中,我们的算法也始终如一地提供稳健的结果并且比其他隐式方法表现得更好。我们针对 NTCIR-12 IMine-2 任务中使用的一组模糊和分面查询测试我们的框架,并通过广泛的实证分析表明,我们的框架始终优于现有的隐式多元化查询扩展算法。我们算法的实用性真正体现在第二组实验中,我们为特定科学领域的检索引擎索引文档生成多样化的查询扩展。即使在这样的利基场景中,我们的算法也始终如一地提供稳健的结果并且比其他隐式方法表现得更好。我们算法的实用性真正体现在第二组实验中,我们为特定科学领域的检索引擎索引文档生成多样化的查询扩展。即使在这样的利基场景中,我们的算法也始终如一地提供稳健的结果并且比其他隐式方法表现得更好。我们算法的实用性真正体现在第二组实验中,我们为特定科学领域的检索引擎索引文档生成多样化的查询扩展。即使在这样的利基场景中,我们的算法也始终如一地提供稳健的结果并且比其他隐式方法表现得更好。
更新日期:2019-11-29
down
wechat
bug