Research Article
Emotion sentiment analysis from text and image using machine learning models
@INPROCEEDINGS{10.4108/eai.19-12-2025.2362581, author={Kanishk Kanishk and Akanksha Puri and Divya Rani and Akshi Pathania and Anamika Rangra}, title={Emotion sentiment analysis from text and image using machine learning models}, proceedings={Proceedings of the 1st International Conference on Computational Techniques in Data Science, IICTDS 2025, 19-20 December 2025, Chandigarh, India}, publisher={EAI}, proceedings_a={IICTDS}, year={2026}, month={11}, keywords={Sentiment Analysis MELD (Multimodal Emotion Lines Dataset) FER2013 SBERT (Sentence Bidirectional Encoder Representation from Transformers) CNN (Convolutional Neural Network)}, doi={10.4108/eai.19-12-2025.2362581} }- Kanishk Kanishk
Akanksha Puri
Divya Rani
Akshi Pathania
Anamika Rangra
Year: 2026
Emotion sentiment analysis from text and image using machine learning models
IICTDS
EAI
DOI: 10.4108/eai.19-12-2025.2362581
Abstract
Artificial intelligence is ability of machines to think and act like humans. This research uses AI,NLP to extract emotion and sentiments of humans. Emotion and Sentiment analysis have vital role in development of intelligent systems that can understand human behaviour. Analysis of human behaviour using either text or image, unimodal approach often fails to capture full human emotions. Hence, in this paper we identify the emotion and sentiments of humans by text and image by using multimodal. Multimodal means using more than one type of data-like combining text, image, audio or video, to understand and solve problems. For text-based analysis, use the MELD dataset, and for image- based analysis, use the FER2013 dataset. Using the SBERT, for textual embedding, enabling deeper contextual understanding and Convolution Neural Network for image classification. Experimental result shows that the text-based modal achieves the accuracy of 95% and image- based achieves the 72.41%. Merged the extracted features from both modalities using Random Forest algorithm and by using this accuracy achieved is 79.77%. This study demonstrates the effectiveness of a multimodal approach to emotion and sentiment analysis by combining both text and image information. This significantly improve the classification performance compared to unimodal. Future work may involve integration with speech, emotion tracking over time, language independent emotion classification.