Research On Building Social Networks Based On Information Extraction

Nowadays, research on social networks has become one of the most popular topicsin the cross-disciplinary fields. It is of great significance to extract rich and reliablepersonal relations which are the foundation of social networks. This paper mainlyinvestigates the methods of social network construction based on information extractiontechnology. The main contributions include as follows:First, we implement a personal relation extraction prototype system for WWW. Itbegins with a handful of seeds and extracts personal family relations from large-scaleInternet resources in a bootstrapping fashion, laying the foundation for further researchon personal relation network construction.Second, we investigate the method of personal relation network construction fromlocal corpus. After the step of personal relation extraction via bootstrapping,Cross-Document Co-reference Resolution (CDCR), i.e. name disambiguation andclustering, are adopted to construct preliminary personal relation networks, pointing outthe direction for further study.Finally, we propose a method of constructing personal relation networks viaWithin-Document Coreference Resolution techniques. Pattern generalization andcentering theory-based Within-Document Coreference Resolution are adopted to extractmore personal relations for further person network construction. Meanwhile, a newevaluation metric called GROVE is devised to evaluate the performance of personalrelation networks more reasonably.The experimental results on the Gigaword corpus show that, social networks can beconstructed effectively via bootstrapping-based personal relation extraction andCDCR-based network fusion. In addition, center theory-based Within-Document Coreference Resolution can enlarge the scale of personal relations for social networkconstruction, thus enriching the social networks with reliable relations.