Text sentiment analysis method based on convolutional neural network based on Chinese character component features

Abstract

The invention provides a text sentiment analysis method based on the convolutional neural network of Chinese character component features. The method first considers the emotional intensity of the sentiment word, and optimizes the weight of the sentiment dictionary in combination with Attention; The feature of granularity, because the basic morpheme of Chinese is a Chinese character component, which carries rich information such as phonetics and semantics, which is different from the 26 letters of English, and finally proposes a convolution based on a dual-channel word embedding of Chinese character components and sentiment lexicon Chinese text sentiment classification method based on neural network. Experiments on multiple public data sets prove that the model can significantly improve the sentiment classification effect of text.

Description

technical field [0001] The present invention relates to the field of natural language in the direction of computer technology artificial intelligence, and more specifically, relates to a text sentiment analysis method based on a convolutional neural network based on the features of Chinese character components. Background technique [0002] In today's society, Internet users interact with information explosion, and interactive platforms in multiple industries such as e-commerce shopping, entertainment, catering, transportation and culture have produced a large number of short texts (Short Text). The emotional orientation of these text data is important for both parties. communication bridge. For example, movie distributors and movie fans are more inclined to show and watch new movies that are well received by the public; merchants and potential customers pay close attention to buyers' satisfaction with newly purchased products. These are typical usage scenarios of short text...

AI Technical Summary

Problems solved by technology

The former generally refers to the combination of grammatical rules and sentiment lexicon to identify words with emotional polarity to calculate the emotional orientation of the text. However, due to the dependence on the size and quality of the sentiment lexicon, it not only requires a lot of manual preprocessing, but also has poor generalization, especially when used in When cross-domain text; the latter is divided into two stages, the first stage mainly uses traditional machine learning techniques, such as support vector machine (SVM), naive Bayesian (NB) and random forest (RF) classification Algorithm, the main problem is that it needs to manually construct features, and Chinese emotional features are different from English, Chinese has no spaces between words, and Chinese words are usually composed of more than one Chinese character, so word segmentation is required first based on word extraction features; so Usually extracting features is not only time-consuming and laborious, but also the classification results are too dependent on the extracted features; in the second stage, deep learning technology appeared, mainly referring to the application of various deep neural networks CNN, RNN, etc. to achieve classification. This method does not require manual labeling. Emotional dictionaries do not need to manually construct features, and rely entirely on self-learning to extract features, but the granularity of feature extraction is not fine enough, and often requires a large amount of labeled corpus, which is scarce in the field of Chinese sentiment analysis

Images