一种基于用户兴趣的微博实体链接方法  

User interest topic model for microblog entity linking

作　　者：宋俊[1,2,3] 李禹恒[1,2,3] 黄宇[1,2] 陈昊[1,2,4] 付琨[1,2] 

机构地区：[1]中科院空间信息处理与应用系统技术重点实验室,北京100190 [2]中国科学院电子学研究所,北京100190 [3]中国科学院大学,北京100190 [4]北京空间信息中继传输研究中心,北京100094

出　　处：《计算机应用研究》2016年第7期2079-2082,共4页Application Research of Computers

基　　金：国家"863"计划资助项目(2012AA011005)

摘　　要：针对微博内容较短、歧义较大的问题,利用概率主题模型对用户的兴趣进行建模,提出了一种基于用户兴趣的微博实体链接方法。具体地,利用现有的主题模型从知识库的大量数据中训练实体与上下文词汇的语义关联,提出用户兴趣主题模型来建模用户对实体的兴趣以及微博的语义,并完成实体链接的任务。此外,在真实数据集上进行了大量实验和分析,取得了87.6%的实体链接准确率,实验结果表明,与现有方法相比,该方法通过用户兴趣的建模更好地刻画了微博的语义,因而也取得了更高的实体链接准确率。The objective of this paper was to propose a novel entity linking method for microblog based on user＇ s interest. The proposed method models user＇ s interest to enrich the semantics of microblog. To achieve this ,this paper firstly utilized an existing topic model to train the relatedness between entities and context words from Wikipedia. Then it proposed a user interest top- ic model to capture user＇ s interest on entities and represent microblog＇ S semantics. And it also obtained the linking entity from this topic model. Moreover, an efficient Gibbs sampler was developed. The experimental results on real-world dataset verify that the proposed method can better reflect the semantics of microblog and also outperforms the baseline method in terms of linking accuracy.

关 键 词：自然语言理解 实体链接 实体消歧 概率主题模型 用户兴趣建模 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]