Attribute recognition method based on knowledge distillation, terminal equipment and storage medium

Abstract

The invention relates to an attribute recognition method based on knowledge distillation, terminal equipment and a storage medium. The method comprises the steps of collecting question and answer data of different fields to form a training set; constructing a teacher network model, and training the teacher network model through the training set; according to knowledge maps of different fields, extracting all attributes corresponding to each entity; performing entity identification on the question and answer data in the training set, obtaining all attributes possibly corresponding to the question and answer data according to the identified entities and all attributes corresponding to the entities, and obtaining priori knowledge; constructing a student network model, training the student network model through the training set based on knowledge distillation and prior knowledge, and obtaining a final student network model until performance indexes meet requirements; and using the final student network model to carry out attribute identification on the question and answer data. The method overcomes the defect that the precision and speed of a traditional attribute recognition method cannot be both considered.

Description

technical field [0001] The present invention relates to the field of knowledge graphs, in particular to an attribute recognition method based on knowledge distillation, a terminal device and a storage medium. Background technique [0002] With the rapid development of information technology, people's demand for fast and accurate access to information is becoming stronger and stronger, and question answering systems have also emerged as the times require. Among them, the Knowledge Base Question Answering (KBQA) system based on knowledge graph is currently receiving extensive attention because of its fine management, efficient reusability and precise understanding. [0003] A necessary part of building a question answering system based on knowledge graphs is to analyze the key information of the question, that is, element recognition, including: entity recognition, attribute recognition, label recognition, relationship recognition, etc. Large-scale knowledge graphs often have...

AI Technical Summary

Problems solved by technology

Large-scale knowledge graphs often have tens of thousands of attributes. When using a small model with low complexity and simple structure for attribute recognition, it has disadvantages such as low accuracy and poor recognition effect; when using a large model with high complexity for attribute recognition, it can meet the effect. Requirements, but the large number of parameters of the large model requires a long time for reasoning and calculation, and the prediction speed is slow, which affects the efficiency of the question answering system and does not meet the requirements of production deployment

Images