融合加权动态权威度和兴趣度的专家推荐方法

Weighted Dynamic Degree of Authority and Degree of Interest for Expert Recommendation

问答系统是目前热门的知识库构建方式之一.然而,当前的问答系统普遍采用专家自主回答或分类随机推荐方式,问题回答的准确率、及时性均较低,导致知识库中噪音知识泛滥.针对以上现象,提出一种基于加权动态权威度的专家推荐方法.该方法首先通过分析专家历史回答内容,并将专家加权动态权威度与LDA模型相结合,构建专家偏好档案;然后及时、精准地将新问题推荐给潜在最适宜专家,从而达到提高问答系统知识库准确性的效果.为了验证本文方法的可行性和有效性,我们使用新浪爱问真实数据集进行分析实验,实验结果表明该方法能够有效地提高专家推荐的准确率.

Question answering system has become a very popular knowledge base. However,in current QA system, experts answer the question independently or the question is recommended to answerers randomly, which results in the low accuracy and promptness of question. This procedure leads to the flooding of noise knowledge in knowledge base. To solve this problem, in this paper, an expert recommendation method based on Weighted Dynamic Degree of Authority is presented. By analyzing answerers＇ answering history , interest profile of answerers are modeled with the mixture weighted dynamic degree of authority and the Latent Dirichlet Allocation model. Finally, the new question will be recommended to the most appropriate experts so as to improve the accuracy of the question- answer system. And we evaluate the effectiveness and feasibility of our method using real dataset from Sinalask. Experiment results show that this method can improve the performance of expert recommendation comparing to the baseline method.