Research On Multimodal Feature Fusion Sentiment Analysis Algorithm Based On Contrast Learning

Sentiment analysis,which refers to determining the sentiment tendency of data,has important applications in a variety of fields such as social media analysis,customer service,and market research.In sentiment analysis tasks,multimodal data can provide more comprehensive and accurate representations than unimodal data,which is more conducive for models to learn the sentiment features in the data and can improve the performance of sentiment analysis models.In this paper,we propose a sentiment analysis algorithm based on different feature fusion strategies for multimodal sentiment analysis tasks with the goal of improving the accuracy of sentiment analysis models by addressing the difficulty of fusing the features of each modality in multimodal sentiment analysis tasks,as follows:(1)In this paper,we propose a CLLF(Contrastive Learning and Late Fusion)sentiment analysis algorithm using an attention mechanism based on a multimodal feature late fusion strategy.This algorithm combines the attention mechanism to encode each modality data individually,focus weighting on key information,remove redundant information,retain complete emotion-related features,and introduce a contrast learning method between text and images so that the model can capture the common emotional features in different modalities and the emotion-linked features between modalities.(2)To further enhance the ability of the model to capture the emotion-linked features between different modalities,a multimodal deep coding CLEF(Contrastive Learning and Early Fusion)sentiment analysis algorithm is proposed based on a multimodal feature early fusion strategy.This algorithm advances the feature fusion behavior to the encoder layer,obtains multimodal fusion features and then performs sentiment classification,makes full use of the complete sentiment features of single modality and the sentiment connection features between different modalities to jointly complete the classification task,and introduces a comparative learning method of source fusion features and data-enhanced fusion features so that the model can learn the sentiment features in different expressions of modality and enhance the model’s robustness to data The robustness of the model to the data is enhanced.(3)The experimental results show that the CLLF algorithm and CLEF algorithm proposed in this paper achieve better sentiment analysis results on the publicly available MVSA dataset.Compared with the single-modal sentiment analysis model,both CLLF and CLEF perform better.Compared with the multimodal sentiment analysis model,the CLLF model improved the F1 value by 2.07% and the ACC value by 0.43% on the MVSA-Single dataset and the MVSA-Multiple dataset,respectively,over the Co-MN-Hop6 model;the CLEF model compared with the MGNNS model on the MVSA-Single and MVSA-Multiple data sets,the F1 values were improved by0.97% and 0.43%,respectively.