A Short Text Classification Method Based on Multiple Weakly Supervised Ensemble

Abstract

The invention discloses a short text classification method based on multiple weak supervision integration, which includes: obtaining the original data set and knowledge base, and performing data preprocessing; performing knowledge extraction on the preprocessed data; expressing the extracted knowledge as a label Functions, and used for data labeling; label integration through conditionally independent models; training classification models based on fully connected neural networks; evaluating and optimizing the classification models to obtain the optimal model; using the optimal model to classify short text. The short text classification method based on the integration of multiple weak supervision in the present invention: uses the combination of keyword matching, regular expressions and distant supervision clustering to fully express the explicit knowledge and implicit knowledge; the probability label generated by the label integration mechanism , which not only realizes the automatic labeling of unlabeled data, alleviates the data sparsity problem of short texts, but also effectively solves the imbalanced classification problem of short texts.

Description

technical field [0001] The invention belongs to the field of natural language processing, and in particular relates to a short text classification method based on multiple weak supervision integration. Background technique [0002] Under the background of mobile Internet, the development of instant messaging not only promotes the surge of short text, but also makes the research and application of short text classification more and more important. [0003] Supervised machine learning mainly relies on manually labeled data and good feature representation. Good feature expression can be learned automatically with deep learning. However, due to the thousands of parameters that need to be learned, supervised deep learning is still inseparable from a large amount of labeled data. In fact, the training data for supervised learning is still dominated by manual annotation. Manual labeling is very expensive and time-consuming. Furthermore, as real-world applications continue to ch...

AI Technical Summary

Problems solved by technology

[0009] In view of this, the present invention is committed to providing a short text classification method based on multiple weak supervision integration, which can solve the problems of label bottleneck, data sparseness and unbalanced classification in short text classification as a whole.

The method of the present invention not only innovatively introduces three sources of weak supervision: keyword matching, regular expressions, and far-supervised clustering into short text annotations for the particularity of short texts; it also proposes a multiple Weakly supervised integration method, which integrates the discrete labels directly output by multiple weak supervisions into probability labels, in order to solve the imbalanced classification problem

Images