摘要
提出了一种基于文档内位置关系的伪相关反馈框架LRoc(location-based Rocchio framework)。该框架采用不同的核函数对候选词项在反馈文档中的位置进行建模,得到候选扩展词的位置重要度,并将其应用到经典的Rocchio模型中。该方法在选择和评估候选扩展词时,不仅考虑了词频,也考虑了词项位置的影响,有助于获取与查询更可能相关的扩展词。最后,在5种TREC数据集的实验结果表明:基于LRoc框架提出的3种模型(LRoc1、LRoc2和LRoc3)对比基线模型在MAP和P@20指标上具有显著提升。
This paper proposes a location-based Rocchio framework(LRoc),with three variants.The method uses different kernel functions to model the term location in the feedback documents,obtains the importance information from the locations of candidate expansion terms,and integrates it into the classic Rocchio model.When selecting and evaluating the candidate expansion terms,this method not only considers term frequency,but also considers the influence of term location,which helps to obtain the expansion terms that are more likely to be relevant to the query.Finally,a series of experiments are performed on five standard text REtrieval conference(TREC)datasets.The proposed three models(LRoc1,LRoc2 and LRoc3)based on the LRoc framework all have got significant improvements over the baseline model in terms of the mean average precision(MAP)and precision at position 20(P@20)indicators.
作者
王雪彦
何婷婷
黄翔
王俊美
潘敏
WANG Xue-yan;HE Ting-ting;HUANG Xiang;WANG Jun-mei;PAN Min(Hubei Provincial Key Laboratory of Artificial Intelligence and Smart Learning,Wuhan 430070,Hubei,China;School of Com-puter,Central China Normal University,Wuhan 430070,Hubei,China;National Language Resources Monitor&Research Center for Network Media,Wuhan 430070,Hubei,China;National Engineering Research Center for E-Learning,Central China Normal University,Wuhan 430070,Hubei,China;School of Mathematics and Statistics,Central China Normal University,Wuhan 430070,Hubei,China;School of Computer and Information Engineering,Hubei Normal University,Huangshi 435000,Hubei,China)
出处
《山东大学学报（理学版）》
CAS
CSCD
北大核心
2021年第5期76-84,共9页
Journal of Shandong University(Natural Science)
基金
国家自然科学基金资助项目(61532008,61932008)
武汉市科技计划项目(2019010701011392)
国家语委科研中心项目(ZDI135-135)
湖北省重点研发计划项目(2020BAB017)。
关键词
伪相关反馈
位置关系
查询扩展
pseudo-relevance feedback
locational relationship
query expansion
