Our laboratory published an article in the international journal Information Fusion (Scopus, Q1):

Markitantov M., Ryumina E., Dvoynikova A., Karpov A. Multi-Lingual Approach for Multimodal Emotion and Sentiment Recognition Based on Triple Fusion // Information Fusion, 2026, vol. 132, 104207. (WOS IF=17.4 Q1, Scopus SJR=4.2 Q1)

Emotion and sentiment recognition based on audio, video, and textual data is one of the key tasks in multimodal human behavior analysis. Many existing approaches are designed for a single language, dataset, or task, which limits their generalization capabilities. The paper presents a multilingual, multimodal, multi-task approach for simultaneous emotion and sentiment recognition in Russian and English. The proposed architecture employs a novel Triple Fusion strategy that combines features from different modalities both at the level of unimodal representations and after multimodal representations have been formed. Experiments are conducted on the Russian-language RAMAS corpus and the English-language MELD and CMU-MOSEI corpora. Temporal encoders based on Transformer, Mamba, and xLSTM, as well as different feature-fusion strategies, are also compared. On the CMU-MOSEI Test subset, the proposed approach achieves mWF=88.6% for emotion recognition and WF=84.8% for sentiment recognition; on MELD, the corresponding WF scores are 49.6% and 60.0%, while on RAMAS they reach 71.8% and 90.0%, respectively. The results demonstrate the effectiveness of the proposed approach across different languages, corpora, and affective-state analysis tasks.