Author(s)

Chenyue Wang

Corresponding Author

Chenyue Wang

Abstract

With the increasing number of educational events, participants have put forward higher requirements for the real-time and accuracy of event information acquisition. To this end, this paper proposes and implements an intelligent customer service system for event scenarios, integrating core functions such as document information extraction, semantic modeling, question-answering response, and dynamic knowledge update. First, for unstructured PDF specification documents, a field extraction method based on regular expressions and keyword window recognition is designed to achieve structured expression of the core information of the event. Secondly, the system introduces the Sentence-BERT semantic embedding model, and through question classification and intent recognition strategies, a multi-type question-answering system that supports basic queries, statistical analysis, and open-ended questions and answers is constructed. Furthermore, a knowledge base update mechanism with version control capabilities is designed to support new document identification, field difference comparison, and vector index reconstruction to ensure the stable operation and data consistency of the system in a dynamic environment. Experimental results show that the accuracy of this system on basic questions is 94.2%, on statistical questions is 89.5%, and the overall question-answering accuracy reaches 87.6%. This research result has good versatility and expansion potential, and can be promoted and applied to intelligent question-answering tasks in document-intensive scenarios such as education and government affairs in the future.

Keywords

Intelligent Customer Service, Natural Language Processing, Information Extraction, Semantic Matching, Knowledge Base Update, PDF parsing