基于LDA的问答网站话题抽取算法  

LDA-BASED Q & A WEBSITES QUESTION LABEL EXTRACTION ALGORITHM

作　　者：战学刚[1] 王晓[1] 

机构地区：[1]辽宁科技大学软件学院,辽宁鞍山114051

出　　处：《计算机应用与软件》2016年第4期95-98,共4页Computer Applications and Software

摘　　要：为了帮助用户在使用问答网站时准确地描述所提问题的话题,对社会化问答网站问题及话题进行了建模,发现问题的潜在语义关系,提出一种基于潜在狄利克雷分布LDA(Latent Dirichlet Allocation)的话题抽取算法。该算法通过挖掘问题与问题之间的潜在语义信息,找到潜在语义相类似的问题,在语义层面上抽取出话题集合,找到最符合的话题列表。在真实网站中的数据进行试验证实,应用该算法可以有效扩大话题抽取的准确率和召回率。To help people accurately describe the topics of the question raised when using question and answer（ Q A） websites,we modelled the questions and topics in socialised QA websites,found the latent semantic relationship among questions,and proposed an LDAbased topic extraction algorithm. The algorithm finds the questions with latent semantics similarity by digging up latent semantic information between questions,extracts the topics set on semantic level,and finds the list of topics that matches the most. It has been proved by the test with the data in actual websites that the application of the algorithm can effectively improve the precision and recall rates of topic extraction.

关 键 词：LDA 问答网站 协同过滤 话题模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]