查询意图自动分类的方法改进探讨  

Discussion on the Improvement of Methods for Automatic Classification of Query Intent

作　　者：贺国秀 张晓娟[3] 

机构地区：[1]武汉大学信息检索与知识挖掘研究所,武汉430072 [2]武汉大学信息管理学院,武汉430072 [3]西南大学计算机与信息科学学院,重庆400715

出　　处：《数字图书馆论坛》2018年第1期53-60,共8页Digital Library Forum

基　　金：国家社会科学基金青年项目"融合用户个性化与实时性意图的查询推荐模型研究"(编号:15 CT Q019)资助

摘　　要：本文在降低数据标注成本的基础上,提高查询意图自动分类的准确率。首先,将ODP主题类目体系映射到Rose等意图类目体系,利用启发式和匹配的方法形成标注规则,对查询日志数据进行自动标注;其次,在提取查询的统计特征、用户行为特征和基于自然语言处理的语义特征基础上,提取查询的句法依赖关系作为分类特征;最后,使用集成学习模型GBDT作为分类器,对查询意图进行分类研究。实验表明,本文提出的标注规则可以获得大量被标注的训练数据集,新增的句法依赖关系特征可以提高查询意图的分类效果,GBDT作为集成学习模型相比线性分类模型可以明显提高查询意图分类的准确率。On the basis of reducing the cost of data marked, the research is to improve the accuracy of query intent automatic classification. Firstly, the ODP subject category system is mapped to Rose＇s intent class system, the heuristic and matching methods are used to form the annotation rules, and the query log data is automatically marked; Then, based on statistical characteristics of the extracted query, the characteristics of user＇s behavior and natural language such as word segmentation and part of speech, the syntactic dependency of the query is extracted as the classification feature; Finally, the integrated learning model GBDT is used as a classifier to classify the query intent. The experimental results show that the proposed rules can be used to obtain a large number of trained training data sets. The new syntactic and relational feature can improve the classification result of query intent. GBDT can improve the accuracy rate of query intent classification as an integrated learning model compared with linear classification model.

关 键 词：GBDT 机器学习 查询日志 查询意图 自然语言处理 

分 类 号：G353.4[文化科学—情报学]