Multi-label character relationship automatic labeling method based on event remote supervision

Abstract

The invention discloses a multi-label character relationship automatic labeling method based on event remote supervision. The method comprises the following steps: collecting events influencing character relationships; making a corresponding event labeling template; constructing an event template knowledge base; carrying out data preprocessing; performing event labeling on the preprocessed sentences by utilizing an event labeling template; carrying out character relationship labeling; performing character relationship labeling to obtain a result; calculating the credibility between the 'event 'and the 'sentence'; calculating sentence scores; setting a threshold value, and discarding the sentences with the sentence scores lower than the threshold value; and obtaining a final character relationship label. According to the method, the events influencing the character relationship are obtained through event template knowledge base matching, then the character multi-label relationship is automatically reasoned according to the multiple events, the character relationship multi-label problem can be solved, the accuracy of multi-label character relationship labeling can be remarkably improved, and more excellent mobility is achieved.

Description

technical field [0001] The invention relates to the technical field of natural language relationship extraction, in particular to a method for automatically labeling multi-label character relationships based on event remote supervision. Background technique [0002] Character relationship extraction based on deep learning methods is widely used in natural language processing tasks. Deep learning methods are data-driven and require more labeled data. However, manual labeling methods require a lot of manpower and time, and the data scale cannot meet the needs of deep learning methods. Distant supervision is one of the common methods to automatically label person relationship datasets. It relies on a knowledge base containing a large number of triples, matching person-entity pairs in sentences with entity pairs in the knowledge base. However, the construction of the triple knowledge base is time-consuming and laborious, and the relationship of all entities needs to be exhaus...

AI Technical Summary

Problems solved by technology

However, the existing ESDS method only focuses on the labeling of a single person relationship label, ignoring the fact that some person relationships may have multiple labels.

Images