摘要
文章在传统的伪相关反馈基础上引入深度强化学习的查询扩展方法来改善信息检索中由词不匹配造成的检索效果不佳问题。选择eBay于2017年发布的用户查询与商品名称作为实验数据,利用深度学习框架抽取词的抽象特征,并把召回率作为奖励,使用强化学习方法对扩展词进行选择。当使用召回率、精度和平均精度均值三个指标对模型进行评价时,文章提出的基于深度强化学习的查询扩展方法明显优于基线方法(原始查询、基于TF-IDF的查询扩展、基于余弦相似度的查询扩展和基于深度学习的查询扩展),扩展后的查询检索效果在召回率上比原始查询高1.32%。实验结果表明基于深度强化学习的查询扩展模型能够改善词不匹配带来的问题,提高系统检索效果。
In this paper,based on traditional pseudo-relevance feedback,we introduce deep reinforcement learning-based query expansion(DRLQE)to improve the poor performance of information retrieval,which is caused by word mismatch.Query phrases and product names posted on eBay in 2017 are selected as experimental data.A deep learning framework is used to extract the abstract features of terms,recall is regarded as reward,and reinforcement learning is employed to select expanding words.The proposed model has been tested on eBay dataset,and when we use Recall,Precision and Mean Average Precision(MAP)metrics to evaluate the model,the query expansion method based on deep reinforcement learning is significantly better than baseline methods(raw query,query expansion based on TF-IDF,cosine similarity and deep learning).The recall value of our method is 1.32%higher than raw query.The experiment results show that the proposed model can effectively improve the problems caused by word mismatch and improve the system retrieval performance.
出处
《情报理论与实践》
CSSCI
北大核心
2019年第9期146-153,共8页
Information Studies:Theory & Application
基金
国家自然科学基金面上项目“大数据环境下基于领域知识获取与对齐的观点检索研究”(项目编号:71373286)
教育部哲学社会科学研究重大课题攻关项目“提高反恐怖主义情报信息工作能力对策研究”(项目编号:17JZD034)的成果
关键词
深度强化学习
查询扩展
伪相关反馈
信息检索
deep reinforcement learning
query expansion
pseudo-relevance feedback
information retrieval