Research On The Extraction Of Entity Relationship In Chinese Field Based On Multi - Core Fusion

Chinese entity relation extraction is an important subtasks in information extraction, and is also one of the hot issues in the field of natural language processing now. Chinese field entity relation extraction is to find the relations between entities at sentence-level in the document of Chinese specific domain. This paper focuses on the issues of domain information extraction, finding kernel function in Chinese field entity relation extraction, and has done in-depth studies and discussion in the following several aspects:(1)Building domain knowledge tree. This domain knowledge tree uses the Chinese domestic tourism information of BaiduTravel which is Web resource site. Place was served as a node. It includes some attributes like food, lodging and specialty. Domain knowledge tree will provide domain information when selecting domain feature vector and constructing structure tree which integrates domain knowledge.(2) Proposing a Chinese field entity relation extraction method based on convex combination kernel function. For the problem that single plane of kernel function is not the optimal kernel function, this method finds the optimal convex combination kernel function on multiple planes, and constructs the Chinese domain entity relation extraction classification model.(3) Proposing an entity relation extraction based on multiple kernels fusion method which fuse domain knowledge tree kernel function and plane kernel function. For the problem that the traditional convolution tree kernel function cannot express domain structure information, this method makes domain knowledge subtree which contains entities hang on the root of the shortest path tree, then integrates this kernel function and the optimal plane convex combination kernel function into kernel function that multiple kernels fusion to construct Chinese domain entity relation extraction classification model.(4) Designing and implementing the prototype system of the Chinese filed relation extraction based on multiple kernels fusion with the above research results.