A Method for Sentiment Polarity Analysis of Heterogeneous Migration Images Based on Multimodal Deep Latent Correlation

Abstract

The present invention proposes a method for analyzing the emotional polarity of heterogeneous migration images based on multimodal deep potential correlation, which includes the following steps: 1) Constructing an initial emotional image data set; using the emotional polarity corresponding to the emotional vocabulary as the emotional polarity of the image label; 2) Clear the noise data in the initial emotional image dataset; use the emotion consistency discriminant method and the probabilistic sampling model based on the multimodal deep convolutional neural network to remove the noise; 3) Construct the heterogeneous transfer model; then train source domain text and target domain image; 4) construct multimodal embedding space; embed semantic information of source domain text into target domain image; 5) train image sentiment polarity classifier for image sentiment polarity gender analysis. This method can obtain large data scale, low labor cost, low data noise, high prediction accuracy, strong model interpretability and strong classification ability, and can achieve better image emotional polarity analysis effect.

Description

technical field [0001] The invention relates to the technical field of image content understanding and data analysis, in particular to a method for analyzing the emotional polarity of heterogeneous migration images based on multimodal deep latent correlation. Background technique [0002] With the popularity of social media, social networks have an irreplaceable position in people's daily life. More and more social media users are more inclined to use visual content to express their opinions and share their experiences, and a large number of user-generated images are also generated. Facing the huge number of user-generated images, how to mine user-generated image data? The academic and commercial value of the image has become an urgent problem to be solved in academia and industry, especially mining and analyzing user opinions and emotions in the data. Therefore, using user-generated image data for opinion mining and sentiment analysis has become a research hotspot. [0003...

AI Technical Summary

Problems solved by technology

[0005] Recently, some researchers have begun to study the problem of multimodal social media sentiment analysis by combining visual content and text information. Although these methods achieve better results than those that only consider visual features, they ignore the visual and textual modalities. Structural correlation between modalities, exploiting the correlation between visual and textual features associated with images has improved the performance of some cross-modal retrieval and image annotation tasks, but the effectiveness in image sentiment analysis has not been fully investigated. prove

Images