Entity Set Expansion Based On Distributed Representation

As the amount of information in the society exponentially increases,Knowledge Base becomes more and more concerned because of its ability to solve information overload. Knowledge Base stores the structured information with the entity as the node and the relation as the edge. Entity set expansion is an important task of knowledge base completion, which focuses on the unitary information of the entity category within the node. Entity set expansion is an important part of knowledge base completion.Distributed representation and neural networks have achieved great success in the field of natural language processing, while the related research on the direction of entity set expansion is not deep. In this paper,we study entity set expansion based on distributed representation, use neural network model based on distributed representation and fine-grained entity classification technology based on weak supervision to identify the set of entities in sentences. The main research contents and achievements are as follows:1. This paper designs and implements a model of entity set expansion based on multi-layer perceptron, and compares with the model based on semantic and grammatical features. The multi-layer perceptron model is combined with weakly supervised entity distribution to avoid the artificial features and outperforms the baseline model in recall rate.2. In this paper, we study a model based on convolutional neural network, which can capture more information as well as reducing the parameter size, thus improving the recall rate and avoiding the loss of precision, and eventually improve the whole performance of the model.3. This paper designs and implements a model based on recurrent neural network to do entity set expansion. It uses Bi-LSTM network to extract the context features, which can express the long contexts better,and can obtain useful information and meanwhile avoid introducing too much noise. This model outperforms all other models in many aspects greatly, indicating that the model can better solve the problem of entity set expansion.