Self Learning Method For Social Relation Extraction

As rapid development of internet information and social networks, we need tools which could search and extract data from the large dataset urgently. Search engine technology could solve the problem of text retrieval, but unable to search for social relations. How to extract social relations from the text content in internet which is an problem as well as entity relationship extraction task.According to this problem, this paper proposed a system of social relation types. The relation extraction task is defined as a pattern classification problem. And a supervised method based on machine learning algorithm was used to solve the problem. In the building process of corpus, the ACE evaluation corpus become an important reference. We defined six social relation types and collected the format corpus from the Internet. Based on the characteristics of social relationship, we made the appropriate rules for the work of feature extraction. Based on the result of feature extraction we proposed two methods of support vector machine(SVM) and the maximumentropy model for the relation extraction experiment. The results show that support vector machine algorithm is better than the maximumentropy model. Meanwhile for the undefined relationship to and unlabeled corpus, this paper present an algorithm of unsupervised relation extraction algorithm, the algorithm does not require a lot of labeled training data as the initial corpus. It utilized the search engine ability of indexing and processing vast amounts of data and extracted representative social relationship network. The experiment was taken according to the defined social relationship and obtained a good extraction results.Finally, a social relation extraction platform is designed and implemented which included the corpus preprocessing, feature extraction and algorithm module. The platform provided the common module of relation extraction and researchers could focus on the study of extraction algorithm based on the platform.