• Shubham Das

Multimodal sentiment analysis is a relatively new field of research with the aim of enabling machines to perceive, analyse, and express emotion. We can learn more in-depth information about the speaker's emotional qualities through the cross-modal engagement. Bidirectional Encoder Representations from Transformers is a powerful pre-trained language representation model (BERT). Fine-tuning has provided novel, state-of-the-art results on ten natural language processing tasks, that included question-answering and natural language inference. Although the majority of earlier studies that improved BERT only used text data, it is still worthwhile to investigate how to learn a better representation by incorporating multimodal data. The Cross-Modal BERT (CM BERT), which we suggest in this research, uses the communication between text and audio modality to hone the pre-trained BERT model. Masked multimodal attention, the core aspect of CM-BERT, combines the knowledge retrieved from text and audio modalities to dynamically modify the weight of words. On the open multimodal sentiment analysis datasets CMU-MOSI and CMU-MOSEI, we test our methodology. The findings of the experiment reveal that it has greatly outperformed prior baselines and text-only finetuning of BERT in terms of performance on all criteria. In addition, by using audio modality information, we demonstrate the masked multimodal attention and demonstrate that it can appropriately modify the weight of words.

Natural Language Processing, BERT, Cross-Modal BERT (CM-BERT)

"Text-Audio Sentiment Analysis Using Cross-Modal BERT", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.9, Issue 8, page no.b163-b168, August-2022, Available :http://www.jetir.org/papers/JETIR2208124.pdf

"Text-Audio Sentiment Analysis Using Cross-Modal BERT", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.9, Issue 8, page no. ppb163-b168, August-2022, Available at : http://www.jetir.org/papers/JETIR2208124.pdf