Research And Implementation Of Talent Evaluation System With Questions Of Complicated Features

Nowdays, with the improvement of the talent evaluation theory and the development of the information technology, large numbers of talent evaluation systems have sprung up. However most of these systems mainly aim at the enterpeises and education evaluation, but rarely take the administrative institutions in our country into consideration. In fact, most of the talent evaluation institutions which serve these administrative institutions in each places still take manual operation method to organize those assessments. In order to help these talent evaluation institutions to reduce the evaluation cost and improve the evaluation quality, we develop a talent evaluation system with questions of complicated features.The system we research is different from current talent evaluation institutions in its questions which have complicated features including lots of domain knowledge covered, various question types, complex structure and rich contents. Based on that unique character and the actual demands of the users, this paper researches the ways of question management, the algorithm of similarity detection between questions, the method of test paper generation including an intelligent way with improved genetic algorithm and a flexible manual way, and the implementation of the actual work flow of the talent evaluation institutions.The contribution of this paper mainly lies in the following three points:the first is using the semi-structured way to organize the question content and designing a method to dynamically update the difficulty and discrimination of a question; the second is improving the Simhash algorithm from three key points to overcome its poor performance in short text. Finally one solution stands out from several plans we made with the use of keywords filtering based on information entrophy and the weight of feature vector setted as 1 and the Cityhash method, which can well detect the similarity of questions. Then EW1CS-SimDetect Algorithm based on that solution is put forward and applied; the third is supporting users to generate test papers from questions of at least 12 first level subjects and 110 second level subjects.