基于语义特征扩展的知识库增量引文推荐算法

将知识库增量引文推荐(cumulative citation recommendation, CCR)任务分解为3个基本的关键问题:针对知识库某一实体名的查询扩展;针对文档和实体的特征提取;基于线性和非线性相结合的分类模型。提出了基于语义词典(DBpedia)与词向量(word embedding)相结合的方法进行查询扩展,以及利用LDA和ESA两种算法对文档进行特征提取,最终通过线性逻辑回归与非线性随机森林相融合的分类算法实现CCR算法。与基线系统相比,该方法在TREC KBA2014评测数据上的试验结果的F1平均提升了14.7%,表明本文设计的方法能够较好地解决引文推荐问题。

Algorithm of knowledge base cumulative citation recommendation based on semantic features expansion

The task of knowledge base cumulative citation recommendation was mainly decomposed into three basic key problems: query expansion based on an entity name in knowledge base, feature extraction for documents and entities.We proposed a method that using the combination of the semantic dictionary(DBpedia)and the word vector(word embedding)for query expansion, and using LDA and ESA algorithms for feature extraction. Finally classify documents based on linear Logistic Regresion combined with unlinear random forest. The F1 value of this system operated on TREC KBA2014 promoted 14.7% compared to the baseline, which indicated that the method raised by the study is good at dealing with question of citation recommendation.