Cross-Modal Sentiment Analysis of Text and Video Based on Bi-GRU Cyclic Network and Correlation Enhancement

Cross-modal sentiment analysis is an emerging research area in natural language processing. The core task of cross-modal fusion lies relationship extraction and joint feature learning. existing methods focus on static text, video, audio, other modality data but ignore the fact that different are often unaligned practical applications. There a long-term time dependence among sequences, it difficult to explore interaction between modalities. paper proposes model (UA-BFET) based enhancement technology scenarios, which can perform text video social media. Firstly, adds cyclic memory network across steps. Then, obtained features with applied unimodal process next step Bi-directional Gated Recurrent Unit (Bi-GRU) so progressively enhanced continuously complement each other. Secondly, extracted taken jointly from subjected canonical correlation (CCA) input into fully connected layer Softmax function for analysis. Through experiments executed public datasets MOSI MOSEI, UA-BFET has achieved or even exceeded effect audio outstanding advantages solving scenarios.