The Design And Implementation Of Information Extraction In VOCA

With the development of society, enterprises are facing increasingly fierce competition. The quality of service is an important factor that affects the firms’ survival and development. Many companies attach great importance on their quality of service. They provide a variety of ways to get users’ feedback on their service, including questionnaires, online message, and call center services. For many large service companies, the call center is an important way. However, since the big data collected by the call center is unstructured, it is difficult to use these data to evaluate their quality of service as well as to identify the key factors that affect quality of service.The main difficulty of service quality evaluation lies in subtasks including subject extraction, emotional information extraction, evaluation unit combination and emotional tendency analysis. Subject extraction extracts the subject described in the text. Subject is object of evaluation, and it’s generally a product or service, or a property of that. Emotional information is words expressing feelings used by users to describe and appraise the product. The evaluation unit is combination of the evaluation object and evaluation of the phrase; it is the basis of sentiment analysis.This thesis provides a method that uses statement language AQL (Annotation Query Language) to create rules for unstructured data extraction based on SystemT platform. SystemT is a rule-based information extraction system. It has obvious advantages in the expressive power and performance, compared with existing stacked syntax-based information extraction system. SystemT uses the AQL to create set of rules to identify evaluation unit from free text. AQL rules can match the critical information in unstructured text, including theme-related words, evaluation words. In the subject information extraction, this paper’s method enables BA (Business Analyzer) to define subject based on the practical business scenario. In order to facilitate the use of the user, the system is designed to achieve a variety of topics defined, contain the phrase, any word with the form. In the analysis of emotional tendencies, the system provides a tabular form of the intermediate results for the user to find the detailed analysis of the results. Statistical results are given for each topic and topic keyword selected by users based on the intermediate results.In this thesis, the emotional tendency analysis system uses information extraction marking method based on SystemT. Tests on text data from Citibank credit card service have shown that the system has good recognition effect, strong pertinence and performance, and it will have broad application and prospect in customer service quality evaluation in the call center.