Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{207372,
        author = {B. Nirmala and V.V.N.Bharadwaj and P. Pramod and V. Tej Shyam Sundar and M. Krishna Veni},
        title = {Multimodal Emotion Recognition with Personalized Guidance Using Cross Attention Networks},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {3},
        pages = {714-722},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=207372},
        abstract = {Though feelings deeply influence how people think, interact, or react, many current tools that detect emotions rely on just one source - like words, voice, or face movements - without adapting to individual users. These methods struggle when emotions are faint, layered, or masked. Because they ignore personal patterns, their accuracy drops precisely when nuance matters most. Human affect rarely fits into isolated channels; meaning slips through rigid frameworks. A new approach to recognizing emotions uses personal cues alongside Cross-Attention Networks. Instead of treating each input type separately, visuals, sound, and words move through shared processing layers. Because it links modalities dynamically, meaning transfers more naturally across forms. With scaled dot-product attention, connections between modes grow stronger where needed most. A closer look at how users respond over time shapes the profile module, which tracks mood trends, choices, and habits. Because insights build on past interactions, suggestions become more relevant. When combined with a language model trained on diverse inputs, advice adjusts to fit real-life situations. So instead of just labeling emotions, the system supports decisions tied to personal contexts. 
Through a series of broad tests across standard collections like IEMOCAP, FER-2013, RAVDESS, and Go Emotions, results show stronger outcomes in weighted accuracy, F1-score, and processing speed when using the method described - especially next to single-mode approaches and current merging strategies. While built in separate functional parts, it holds up even if some data channels are absent, proving fit for actual use cases: think emotion-aware tools in therapy aids, responsive digital helpers, or education platforms that adapt as you go.},
        keywords = {Multimodal Emotion Recognition, Cross Attention Network, Affective Computing, Personalized Guidance, Deep Learning, BERT, Bi-LSTM, CNN, IEMOCAP},
        month = {August},
        }