Scene-based text classification model and text classification method and device

Abstract

The invention relates to a scene-based text classification model and a text classification method and device, belongs to the technical field of natural language processing, and aims to realize text classification in multiple scenes and improve the accuracy of a text classification result. The model comprises an input layer used for obtaining a short text and a scene label, and the scene label is used for representing the type of a scene where the short text is located; an embedding layer used for adding a scene code and a position code to the short text to obtain a code representation of the short text in the scene; an encoding layer which comprises a multi-head self-attention layer and is used for extracting text features in the short text according to the encoding representation obtainedby the embedding layer to obtain feature vectors of the short text; a decoding layer which comprises multiple layers of sensors and is used for outputting scores of all service types according to thefeature vectors, obtained by the encoding layer, of the short text; and an output layer which is used for outputting a classification result according to the score of each service type.

Description

technical field [0001] The present invention relates to the technical field of natural language processing, and more specifically, to a scene-based text classification model, text classification method and device. Background technique [0002] At present, in the business of various social platforms (such as Weibo, Zhihu, etc.), there are a large number of similar scenarios and the same task. The tasks in different scenarios are similar but not completely consistent, and there may be differences in data sources, target categories, and discrimination definitions. The characteristics of multi-scene similar tasks are mainly 1. There are many scenes. For example, scenes can contain questions, answers, comments, articles, bullet chats, ideas, etc. 2. The tasks in each scene are similar, but not completely consistent. For example, for unfriendly business, although unfriendly related content is identified in each scenario, the definition of unfriendly in different scenarios is sl...

AI Technical Summary

Problems solved by technology

2. The tasks in each scene are similar, but not exactly the same

For example, for unfriendly business, although unfriendly related content is identified in each scenario, the definition of unfriendly in different scenarios is slightly different, the target labels are different, and the data distribution in each scenario is quite different

[0003] The existing technology trains models, goes online, and optimizes each scene separately, but the development and maintenance are relatively complicated, and the commonality of tasks in each scene is ignored. When the number of samples in a single scene is extremely insufficient, it is even more difficult to train an effective model. Make the results of text classification based on the model inaccurate

Images