Enterprise relationship extraction method, device and storage medium

Abstract

The invention discloses an enterprise relationship extraction method, a device and a storage medium. The method includes: extracting enterprise entity pair sentences, of which relationships exist, touse the same as training sample sentences to establish a sample library; extracting all training sample sentences, which contain one enterprise entity pair, from the sample library, carrying out wordsegmentation, mapping each word to a word vector x, and mapping each training sample sentence to a sentence vector S; using LSTM (Long Short-Term Memory) to calculate a first hidden-layer statevector h and a second hidden-layer state vector h' of the word vector x, obtaining a comprehensive hidden-layer state vector by splicing, and then obtaining a feature vector T; substituting the feature vector T into an average-vector expression to calculate an average vector S; substituting the average vector S and relationship types of the enterprise entity pair into a softmax classification function to calculate a weight a of each training sample sentence; and extracting sentences containing two enterprise entities, obtaining a feature vector T through bi-LSTM (Bidirectional Long Short-term Memory), and inputting the same into a trained RNN model to predict a relationship of the two enterprises. Labor costs are reduced, and the relationship between the two enterpriseentities is more accurately predicted.

Description

technical field [0001] The present invention relates to the technical field of data information processing, in particular to a method, device and computer-readable storage medium for extracting enterprise relations. Background technique [0002] Identifying the relationship between different companies in the news, such as capital transactions, supply chains, cooperation, etc., is of great significance to early warning of corporate risks. However, the common entity relationship extraction methods now require manual labeling of a large amount of training data, and corpus labeling is generally time-consuming and labor-intensive. Contents of the invention [0003] In view of the above, the present invention provides an enterprise relationship extraction method, device and computer-readable storage medium, which can extend the convolutional neural network-based relationship extraction model to remote supervision data, effectively reducing the model's dependence on manually labe...

AI Technical Summary

Problems solved by technology

However, the common entity relationship extraction methods now require manual annotation of a large amount of training data, and the corpus annotation work is generally time-consuming and labor-intensive.

Images