Entity relationship joint extraction model construction method based on reinforcement learning algorithm

Abstract

The invention discloses an entity relationship joint extraction model construction method based on a reinforcement learning algorithm, and the method comprises the steps: inputting a text, and carrying out the entity relationship marking of words of each statement in an original corpus through a joint extraction marking strategy; converting the text into a word2vec vector; pre-training an LSTM joint extractor; initializing a trainer network and disturbing a bag sequence; calculating a reward value of the current statement; calculating a total expected reward value; updating model parameters byusing the optimization function; if the model is converged, performing hyper-parameter tuning training on the model; and if the hyper-parameter is the optimal solution, generating a final entity relationship joint extraction scheme. According to the method, the reinforcement learning model is constructed based on the strategy gradient optimization algorithm, the statement entity relationship joint extraction problem of the complex natural language type original corpus can be effectively solved, and the entity relationship extraction accuracy and the F1 value can be effectively improved.

Description

Technical field [0001] The invention relates to the technical field of knowledge graph construction, in particular to a method for constructing an entity relationship joint extraction model based on a reinforcement learning algorithm. Background technique [0002] The extraction of entities and relationships is the key part and main link of knowledge extraction in natural language processing. The traditional pipeline method first extracts entities, and then identifies the relationship between entity pairs. This way of separation makes the two tasks easier to handle and more flexible. But in fact, the two tasks of common entity extraction and relationship extraction are closely related. The pipeline method usually leads to some wrong extraction, because the entity information obtained in the entity extraction can further help the relationship extraction, and the quality and accuracy of the entity extraction module will also affect the relationship extraction module. If the extr...

AI Technical Summary

Problems solved by technology

Based on this, this patent proposes a joint entity-relationship extraction model using a policy gradient reinforcement learning algorithm, and according to the existing policy gradient reinforcement learning algorithm, the action output by the network can be a continuous value, and the algorithm can be in a Select actions on a continuous distribution, which can avoid the traditional pipeline method and the previous joint extraction method from being unable to closely use the information between entities and relationships, as well as generating redundant other information

Images