Multi-label classification method and system based on semantic-label multi-granularity attention

Abstract

The invention provides a multi-label classification method and system based on semantic-label multi-granularity attention. The method comprises the following steps: constructing a semantic-label multi-granularity attention model: jointly connecting outputs of a stacked expansion convolutional coding module and a label map attention module to a multi-granularity attention mechanism network; taking a weighted label output by the multi-granularity attention mechanism network as an input of a full connection layer; and inputting an output value which is obtained by the full connection layer and is used for mapping the prediction labels into a Sigmoid layer to obtain a prediction probability of each label; model training: adopting a multi-label data set to train the constructed semantic-label multi-granularity attention model, adjusting parameters until the semantic-label multi-granularity attention model converges, and obtaining a trained semantic-label multi-granularity attention model; and taking a to-be-classified multi-label data set as input, and outputting a classification result by the trained semantic-label multi-granularity attention model.

Description

technical field [0001] The invention belongs to the field of multi-label classification, in particular to a multi-label classification method and system based on semantic-label multi-granularity attention. Background technique [0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. [0003] Multi-label classification (MLC) is the task of assigning one or more labels to a given input sample. It has broad application scenarios in the real world, such as document annotation, tag recommendation, information retrieval, and dialogue systems. Since tags usually have complex dependencies, this makes this task extremely challenging in the field of natural language processing. [0004] Some early research works include: Binary Relevance (BR), Classifier Chains (CC) and Label Powerset (LP), all of which have achieved good results. With the great progress of deep learning based on arti...

AI Technical Summary

Problems solved by technology

This makes sequence-to-sequence architectures limited in time efficiency

In addition, for specific MLC tasks, ideally, the order factor cannot be considered between the output labels

However, the label sorting of the above model is fixed (usually in descending order) during the training process, which leads to the model often producing unstable predictions during testing, which reduces the performance and interpretability of the model.

For the graph neural network model, only the relationship between labels and samples is measured, without considering the impact of different content in the text on the prediction results, so it is difficult to apply it on a large scale to other classic MLC tasks

Images