Research On Entity Relationship Extraction

In recent years, the Internet information technology continues to penetrate traditional industries. The age of the rapid growth of Internet user scale has arrived and the number of online and offline text exponentially increases. With contradictions scarcity of resources and unlimited time network information resources between the increasingly prominent, people come to rely on the new tools to access network information. Based on the trend of information needs to realize the automatic extraction of information has a strong practical significance. It usually takes the result of information retrieval as input, using language processing technology for semantic text analyses, accurately grasp the critical information throughout the document and then store in a structured form to the database for user queries and further analysis.Based on this background, the paper studies relevant principles and techniques on entity relation extraction. We proposed and implemented a bootstrapping-based relation extractor and a forum-oriented entity clustering system, mainly to complete the work in the following areas:1. By parsing, we extract the shortest dependency path as the templatebetween entities, thereby performing pattern matching.2. Based on bootstrapping, we realize a semi-supervised system ofrelation extraction and improve the performance by solving theproblem of semantic drift caused by weak constraints.3. For the characteristics of Chinese colloquial campus forum, weintroduce new word detection. We take use of the lexical law of thecorpus itself to help position the subject of an article.4. We apply word activation force to achieve an open field extraction ofentity relationship and construct the network of entities with weights.