Image sentiment analysis method based on sentiment attribute mining

Abstract

The invention belongs to the field of image content analysis, and discloses an image sentiment analysis method based on sentiment attribute mining, which comprises the following steps: mining by using user metadata information, constructing a sentiment attribute subset with large semantic coverage and strong sentiment discriminatability, and predicting a label matrix by using a fusion neural network and a matrix decomposition model. And the accuracy and stability of an image emotion analysis result are improved. According to user tag meta-information and emotion attribute characteristics, fusing four emotion attribute characteristics and optimizing a concept selection model to complete emotion attribute subset selection; utilizing a convolutional neural network to extract visual features and performing consistency optimization through visual contents; constructing an adjacent graph guided by an external knowledge base, and performing semantic relevance optimization; mapping the image visual features and the label semantic features into a potential shared space to realize reconstruction of a prediction label matrix; and finally, reconstructing a prediction label matrix by using the trained reconstruction model, and performing final emotion recognition by using a linear classifier.

Description

technical field [0001] The invention relates to the field of image content analysis, in particular to an image emotion analysis method based on emotion attribute mining. Background technique [0002] Emotion has always been a psychological tool for people to adapt to survival. With the vigorous development of information technology and digital media technology, tens of thousands of electronic information data are disseminated on the Internet every day, and more and more people will pass on the Internet through Various forms express personal opinions and emotional tendencies, and images have gradually become an important medium for Internet users to express themselves. Analyzing image content makes abstract things concrete, which can not only be used as an auxiliary power for user emotion-oriented image retrieval and recommendation systems; it can also combine big data and feedback mechanisms to provide auxiliary decision-making for product design in terms of work creation s...

AI Technical Summary

Problems solved by technology

[0004] First, most studies use empirical learning methods, which require a large amount of diverse training data, and there are problems such as low efficiency, unexplainable process, and poor effect on small-scale tasks; second, the lack of effective reduction of image features and emotional semantics The middle-level semantic representation of the semantic gap between them is difficult to deal with the complexity of the image emotion-semantic mapping relationship, and it is not superior in fine-grained emotion recognition tasks; third, most studies use supervised learning methods, which require manual Data labeling requires a huge amount of data collection; Fourth, in research using Internet user images and user tags, the tag information provided by users usually cannot meet the high-quality standards for visual content descriptions, and there are certain errors and deficiencies

Images