A Text Intent Recognition Method and System Based on Projected Gradient Descent and Label Smoothing

Abstract

The invention discloses a text intent recognition method and system based on projection gradient descent and label smoothing, and relates to the field of natural language processing question answering systems. Including (1) obtaining the initial vector encoding through the embedding layer; (2) using the projected gradient descent algorithm to add disturbances that satisfy the L2 constraint in the embedding layer to form an adversarial sample; (3) using the Transformer network to encode contextual semantic information; (4) using labels Smoothly scale the real intent category; (5) Input the encoder output features into the classifier, and calculate the cross entropy between the smoothed label; (6) Optimize the objective function; (7) After the model is trained, predict the intent category and output. In the classification task, the model of the present invention can fully encode the semantic vector of the input intention; at the same time, add disturbance to the text embedding layer to form an adversarial sample, and perform label sliding on the final classification target, which can significantly improve the robustness and generalization of the model ability.

Description

technical field [0001] The invention relates to the field of natural language processing question answering systems, in particular to a text intent classification method and system based on projection gradient descent and label smoothing. Background technique [0002] With a large number of publicly available online question answering corpora, question answering systems have received attention from researchers in both industry and academia. Q&A systems are usually based on intelligent products that meet the needs of B-end companies, which can significantly improve work efficiency and relieve customer service personnel. Its greatest hidden value is to automatically accumulate standardized data in actual scenarios, reduce costs and improve efficiency in mining customer service value information, and can also be used for future precision marketing and product upgrades. A typical application of a question answering system is to question and answer knowledge in a certain field, ...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem in the field of intent recognition, the existing text classification model lacks a good trade-off between model complexity and model generalization performance. Too few parameters tend to make the semantic coding ability weak, and the accuracy rate is low in the case of many classification categories. ; Too many parameters make the model too complex and difficult to adapt to the situation of limited training samples

Images