Khóa luận tốt nghiệp: Xâu chuỗi văn bản theo sự kiện

The continuous explosion of digital data presents significant challenges and opportunities for effective information management. In this context, the task of event-oriented document grouping emerges as a crucial area of research, essential for transforming raw information into actionable knowledge. This problem, which involves identifying and linking related events within textual data, has garnered substantial academic interest, as evidenced by initiatives like MUC and TDT. The ability to effectively group documents by events allows users to gain a comprehensive overview of online information, particularly news articles, by organizing them into coherent event sequences. This thesis addresses the pressing need for advanced methodologies to handle the ever-increasing volume of data, proposing a structured approach to enhance information retrieval and understanding.

Researchers and practitioners in natural language processing, information extraction, data mining, and artificial intelligence, particularly those interested in event detection and document organization for news and online media.

This thesis, titled "EVENT – ORIENTED DOCUMENT GROUPING," tackles the critical challenge of organizing vast amounts of textual information, especially online news documents, by identifying and chaining related events. The core problem, known as event-oriented document grouping, is conceptualized as a two-stage process: event detection and event sequence detection. The author explores existing approaches, including those utilizing TF-IDF measures for both event detection and document grouping, and methods incorporating temporal and spatial information for discerning event sequences.
The proposed methodology focuses specifically on online news documents, leveraging the informative nature of headlines to facilitate grouping. For event detection, the author develops a robust system based on building a comprehensive dictionary of noun phrases, verb phrases, and entity names, combined with a set of predefined rules. Following event identification, the system employs cosine similarity between terms/phrases and integrates time features to effectively group related events into event sequences. Experimental results demonstrate the efficacy of this approach for Vietnamese news texts, validating its practical relevance. While processing full text might yield higher precision, using titles significantly reduces processing time. The study implies that refining dictionaries and rules for specific topics can further enhance the accuracy and applicability of the proposed document grouping method. This research contributes to advancing information extraction techniques in the face of massive data growth, offering a valuable tool for organizing and understanding dynamic information flows.