问答社区中基于问题粒度的用户专业性预测方法

在线问答社区中大量问题等待回答时间过长、高质量回答数极少，对社区用户在具体问题上的专业程度进行度量具有现实需求。现有的基于链接分析和基于文本分析等方法多集中在社区和话题粒度的专业性度量，并未深入到问题粒度。针对上述问题，定义了问答社区中基于问题粒度的用户专业性概念，在此基础上提出了基于问题粒度的用户专业性预测方法，包括用户专业性度量方法和用户专业性预测模型。该预测方法先利用问答社区中社区用户对回答质量的评价机制，在问题粒度上为用户建立专业性度量；再基于矩阵分解，融合用户偏差、问题偏差以及用户已回答问题集的隐含反馈等信息，构建用户在问题粒度上的专业性预测模型，进而预测用户在待回答问题上的专业程度。利用知乎问答社区互联网话题下的问答数据集，设计了与前述两种主流方法的对比实验。实验结果表明，提出的用户专业性度量方法可以有效地度量用户在具体问题上的专业程度，基于此方法构建的用户专业性预测模型具有更高的预测精度。

Method of Predicting User Professionalism Based on Question Granularity in Community Question Answering

In online community question answering (CQA), many raised questions under go long response time and lack high quality answers. There is a realistic need to measure the community users'' professionalism degree on a specific problem. To date, previous methods based on link analysis or text analysis focused on only the professional metrics of community and topic, and did not fully investigate the question granularity. To address this issue, the concept of user professionalism based on question granularity in CQA was defined, and a prediction method for user professionalism based on question granularity was proposed, including a measurement method and a prediction model. Based on the community users'' evaluation mechanism of answering qualities, the prediction method established professional metrics of users on the question granularity. Integrating together the user bias, the problem bias and the latent feedback of the question set that users answered, a model on problem granularity based on matrix factorization is constructed to predict how professional the user is in answering questions. By using the question-answer (QA) dataset under topics of Internet in Zhihu, comparative experiments with two mainstream methods were conducted. The results showed that the proposed measurement method of evaluating the degree of user professionalism is effective, and the prediction model has higher prediction accuracy.