基于话题模型的专家发现方法

专家发现是实体检索的一个重要方面。经典的专家发现模型建立在专家与词项的条件独立性假设基础上。在实际应用中该假设通常不成立,使得专家发现的效果不够理想。本文提出了一种基于话题模型的专家发现方法,该方法无需依赖候选专家与词项的条件独立性假设,且其可操作性比经典模型更强。同时,使用了一种排序截断技术,该技术极大地降低了模型的计算复杂度。使用CERC(CSIRO Enterprise Research Collection)数据集对模型的性能进行评估。实验结果表明,基于话题模型的专家发现方法在各个评价指标上均优于经典的专家发现模型,能够有效地提高专家发现的效能。

An expert finding method based on topic model

Expert finding is an important part of entity retrieval. Classical expert finding models rest upon the conditional independence assumption between the candidate and term given the document. However, this assumption is usually invalid in real world applications, which makes the performances of classical expert finding models not ideal. In this paper, we proposed an expert finding method based on topic model (EFTM). This method discarded the conditional independence assumption in classical models and is more maneuverable. In addition, a ranking truncation approach which largely decreased the computational complexity of the model was used. Finally, the performances of the new model were evaluated using the CSIRO Enterprise Research Collection. The results showed that the EFTM model outperformed the classical model significantly on all the metrics and could effectively improve the performances of the expert finding system.