Research And Application Of Expert Comprehensive Evaluation Based On Unstructured Data

Expert resource is one of the most valuable resources in various organizations, important knowledge stored in the minds of the experts. Therefore, finding the right experts is the key to solving important problems. Mining and analysis based on expert research papers and other unstructured data, evaluation and retrieval expert in the research field has become an important tool to solve difficult problems.The expert finding purpose is to find experts in a particular field to help organizations solve the problems, expert need to have the appropriate expertise in the field, have a certain influence. So this paper not only research on the expert modeling method based on unstructured document content, but also consider the expert citation which characterizes the authority. Combining expert content of the documents and the citation relationship evaluate expert more accurately and more comprehensively.On the basis of analyzing the corresponding demand and technique, this paper establishes a three layers architecture including information collection module, feature construction module and expert retrieval module, which focus on the characteristics construction module and expert retrieval module. Information collection module is mainly based on the strategy to collect expert information automatically and related technology expert knowledge resources from expert library, academic database, patent information database and other data sources, and then cleaning, regulating, and storing. Feature construction module use the chinese word segmentation based on statistical technique segment documents to words, statistical analysis of the frequency of each word appears in the documents, and then use candidate model to calculate relevance between expert and word, finally build content feature library. Simultaneous analysis and extract the citation information, calculate the weights to build citation relationship network. Then, based on the network use relevance propagation model to update the content feature, when it reaches steady state or after a finite iteration, combine expert document content and citation relationship to generate feature and build feature library. Expert search module, the user input a query, system use the chinese word segmentation to segment query into multiple query terms, for each query term search and match expert feature in the database, read the value as the expert evaluation score on the query word. Assuming each query term mutually independent, merge all query terms scores based on the principle of multiplication, finally evaluate expert on the query comprehensively, and return the experts sort list. In order to improve user’s retrieval performance, take information collection, feature construction as offline process, when a user sends a query, system evaluate the experts directly.Finally, this paper implements an expert information retrieval system and applies in a large pharmaceutical company, this system can help user to find out experts in related fields, improve the efficiency of company.