Emotion analysis method based on regional CNN-LSTM

Abstract

The invention discloses an emotion analysis method based on regional CNN-LSTM. The method comprises the following steps of building a regional CNN-LSTM model; constructing word vectors of words in a text, and representing the text by a sequence of the word vectors; performing regional module decomposition according to sentences in the text or phrases in the sentences; taking regional modules afterthe decomposition in the previous step as input primitives of a CNN, and taking results obtained after text word vector matrixes of regions are subjected to a convolution layer and a maximum poolinglayer as input vectors of an LSTM layer; inputting information of the regional modules obtained in the previous step into the LSTM layer according to an occurrence sequence of the regional modules inthe text, so as to obtain a text vector of the whole text; and inputting the text vector obtained in the previous step into a linear decoder, and carrying out prediction of an emotion value Valence and an excitation value Arousal so as to obtain a VA emotion value result.

Description

technical field [0001] The invention relates to the field of text emotion analysis, in particular to an emotion analysis method based on a region CNN-LSTM. Background technique [0002] Sentiment analysis technology is currently widely used in Internet applications such as online translation, user evaluation analysis and opinion mining. Especially for various emerging online social platforms and shopping websites, quickly obtaining the emotional tendency of user comments can provide great convenience for merchants in advertising and pushing hot topics. Judging from the current market demand and technological development level, the research and innovation of sentiment analysis technology has great value and room for improvement. [0003] At present, there are mainly two methods for the division and expression of emotions: one is categorical emotion representation, which uses clear categories to distinguish emotions, and the most basic method is binary classification represen...

AI Technical Summary

Problems solved by technology

[0014] However, most of the above techniques use simple regression models to predict the emotional value of unmarked words based on words with marked emotional values. These methods are not so ideal in effect

Moreover, these methods often fail to "capture" the rich associations between words in sentences and between sentences in text when they calibrate sentences and texts according to the emotional values ​​of words, and there are still many useful methods. feature omission

[0015] However, if it is based on the neural network method used in the processing of discrete category text sentiment classification, there will be some shortcomings

Convolutional Neural Network (CNN) can extract the feature information contained in the input word vector very well, but CNN cannot effectively "consider" the feature information contained in a sentence or the whole text

Long-Short Term Memory (LSTM) can solve this problem by sequentially modeling sentences or text, but it does not have the advantages of CNN

Images