Research On The Method Of Evaluating Experts Based On Topic Analysis

In the process of project reviews, it often needs to select experts in related fields based on the project content and features to review and check the projects. The traditional artificial selection method for evaluation experts has the following defects, such as time-consuming, laborious and affecting fairness. Therefore, it is highly necessary to recommend appropriate evaluation experts for project application automatically. According to the rich expert database information, evaluation experts recommendation is the process of recommending appropriate evaluation experts for project application automatically, which could find projects that match the experts based on the technology of data mining and pattern recognition. Project topic model characterizes the content information of projects, and it plays a supporting role in the recommendation for the follow-up evaluation experts; It is easy to characterize content information of project documents, which is used to impove the recommendation effect of project review experts effectively; Additionally, mining the historical project information of the reviewed projects, which plays an important role in the project evaluation expert recommendation. This paper does research on the following key problems around project topic model construction, the review expert recommendation based on content, review expert recommendation based on collaborative filtering. Mainly completed the following research work:(1) A method to construct project topic model based on semi-supervised graph clustering is proposed. We first analyze structural characteristics of project documents to extract project name, project keywords and other structural information that could indicate project topics. Combined with expert evidence documents, expert relationship networks and other external resources which could indicate expert topics, we define and extract the association relationship features among project document fragments. Then, using expectation maximization algorithm to determine weights of different types of association relationships, calculate correlation among project document fragments and build undirected graph model for project document fragments. Finally, taking the marked association relationship features as supervised information for clustering, we use semi-supervised graph clustering algorithm to cluster for project document fragments to obtain project topic words, which lays the foundation for the subsequent evaluation expert recommendation.(2) A method for evaluation expert recommendation based on Markov network is put forward. Firstly, this model uses the relevance among topics and the relationships among experts to respectively construct the Markov network of topics and the Markov network of experts, using them to extract the maximum topic clique and maximum expert clique. Secondly, combine the information of the two maximum clique to calculate the relevance between experts and projects, and realize evaluation expert recommendation. Finally, we do experiments on five domain data set and the result shows that our model could improve the effect of evaluation expert recommendation.(3) A method of review experts collaborative recommendation based on topic relationship is proposed. First of all, we obtain the topic of projects and experts, and build topic relationship network of projects (experts). Then, through the topic relationship among the projects (experts), we find the largest similarity collection of neighbors with the current projects(experts), which were integrated into the recommendation algorithm based on matrix factorization. Finally, by learning the rating matrix to get feature vectors of the projects and experts, we can predict review experts’ rating, and achieve the evaluation expert recommendation. An experiment on real dataset showed that the proposed method could predict the review expert’ rating more effectively, and improved the recommendation effect of evaluation experts.(4) Design and implement a prototype system to realize evaluation expert recommendation, which provides convenience for further study on the method of evaluation expert recommendation.