Relation extraction system and method adapted to financial entities and fused with prior knowledge

A relation extraction system adapted to financial entities and fused with prior knowledge, the system at least comprising:

• a deep pretraining module, for training a deep neural network based on finance-related corpora and generating a deep pretrained model capable of recognizing attributes of the financial entities,
• a keyword analyzing module, for extracting with respect to Chinese finance-related prior knowledge and outputting positional information and importance vectors of keywords in a Chinese finance-related text;
• an attention mechanism module, for encoding the positional information of the keywords in the Chinese finance-related text so as to obtain attention masks, and inputting the attention masks and entity information of the Chinese finance-related text to the deep pretrained model so as to acquire text feature vectors, wherein the attention mechanism module is configured to:
• extract structure representation vectors of the financial entities;
• perform vector optimization on the financial entity that has a specified name based on self-attention features in sentences so that certain discrimination exists between/among the structure representation vectors of the same entity in the different sentences;
• and
• encode the positional information of the keywords so as to obtain the attention masks;
• and
• an optimal margin distribution model module, for predicting financial-entity relations based on the text feature vectors and the importance vectors received as inputs,
• wherein the attention mechanism module is in connection with the deep pretraining module, the keyword analyzing module, and the optimal margin distribution model module, respectively.