A Grammar And Dependency Information Based Relation Extraction System For Streaming Data

In recent decades, the rapid growth and globally spread of electronic information phenomenon lead Internet to an information redundant, structure complex and information difficult to retrieval era. Under this kind of circumstance, we confront a new challenge of extracting effective information from data which is produced and spread in high speed. However, with the size of e-data expanding, traditional Information Extraction system needs to face big data or streaming data problem. This topic arises people’s attention and becomes a hot research field.In this paper, we mainly demonstrated our research work on Relation Extraction, the most important sub-class of Information Extraction. We proposed a pattern matching and neural network mixed extraction method, creatively adapted word vector technique into Relation Extraction and designed a Relation Extraction framework for streaming data. The main content of this thesis is as follows:Firstly, considering the weakness of Bootstrapping, we design a new pattern structure and mix Bootstrapping method and MV-RNN method into a blended relation extraction method, which combines both pros together;Secondly, taking account of applying relation extraction for streaming data, we propose an efficient Relation Extraction method based on word2vec and analyze practicality;At last, we design a Relation Extraction framework targeted at streaming data with methods we have proposed.