Video bullet screen emotion analysis method based on multi-scale attention convolutional coding network

Abstract

The invention discloses a video bullet screen sentiment analysis method based on a multi-scale attention convolutional coding network, and belongs to the field of natural language processing and sentiment analysis. An attention mechanism of a multi-level structure is combined with the convolutional neural network, and the defect that the convolutional neural network is difficult to extract long-dependence information of a text is overcome. The method comprises the steps of respectively performing multichannel feature learning of a convolutional neural network on an input sample and a target word for encoding; extracting parts, related to the target words, in the sentences by utilizing an attention mechanism, splicing the feature vectors coded by the final parts to obtain a multi-scale comprehensive feature vector, and inputting the multi-scale comprehensive feature vector into a classifier as a final vector to perform sentiment classification.

Description

technical field [0001] The invention belongs to the field of deep learning and sentiment analysis, and proposes a Hierarchical Attention Convolution Neural Network (HACNN) model using a multi-scale attention mechanism, which can effectively analyze the sentiment information corresponding to target words in complex video barrage, In this way, sentiment analysis can be performed on video barrage samples. Background technique [0002] With the rapid development of the Internet, video barrage has emerged as a new way of commenting and communicating. Viewers can express their views synchronously while watching videos on the Internet. Since bullet chatting is a new way of commenting, so far there are few studies on this aspect, but bullet chatting contains a lot of real-time emotional information about the video, and video feedback users' attitudes and emotions towards each part of the video. Emotional labeling of video key frames according to the emotion of the barrage can facil...

AI Technical Summary

Problems solved by technology

[0005] However, if the sample contains multiple target words and these target words have a variety of different emotional polarities, using CNN alone to process the sample is far from achieving the desired effect

Although CNN can extract the sample information and associated information in the window, it cannot process the entire sample, which leads to the fact that CNN may not be able to fully extract the emotional polarity of the sample or even generate errors.

Moreover, compared with the traditional RNN model, the existing CNN model is still slightly inferior in extracting sample features.

Images