Research On Feature-based Semantic Relation Extraction Between Entities

With the rapid development and wide use of the internet, large amounts of information are in front of people in the form of unstructured text. In order to use the unstructured information better, information extraction which extracts pre-specified information from natural text and gives a structured description is proposed. The basic task of semantic relation extraction which is an important issue of information extraction is to identify and determine the specific relationship between entities, the key technology for the extraction is feature-based and kernel-based machine learning algorithms. The significance of relation extraction is that it has a direct correlation with information filtering, information retrieval and question answering system and plays an important role in the research of understandable for the content, context generation, automatic summarization, machine translation, text classification and information filtering.In this paper, two major issues (feature extraction and feature selection) in the relation extraction based on feature vectors are studied, which specifically includes the following two aspects:1. Proposed a relation extraction scheme based on semantic role featureRelation extraction based on feature vectors is essentially a classification problem which classifies the relationship between the entities into predefined relationship type. Now, the features common used in relation extraction are entity and context features, verb features, distance features and entity expansion features. However, determining the relationship between entities is semantic level problem and not just relies on word level. Then a relation extraction scheme based on semantic role feature is proposed in this paper which adds semantic role feature into common features. As the semantic role annotation represents a shallow grammatical structure of the sentence, which not only implies semantic relationships between the predicate and other words in the predicate framework, but also implies semantic relationships of different words in the predicate framework, Therefore, the semantic role feature which contains a wealth of semantic information will help to distinguish between different types of relationships and improve extraction results.2. Proposed a relation extraction scheme based on feature selectionIn the text classification it is a problem that the dimension of feature space can reach tens of thousands which improves the cost time of training classification model greatly and decreases the extraction result due to the introduction of unnecessary features. In response to these problems, it has been for a long period of study, and achieved certain results. There are similar problems in the feature-based relation extraction, excessive feature space increases the time cost, and reduces the performance. Taking the similarity between text classification and relation extraction problem into account, this paper presents a relation extraction scheme which introduces feature selection used in text classification algorithms, such as information gain, expected cross entropy and x2 statistics, implements feature space dimensionality reduction in relation extraction.