As one of the important research areas in human-computer interaction, conversational recommendation systems aim to provide users with personalized and efficient recommendation services. Accurately capturing and understanding user preferences during the conversational recommendation process is key to achieving effective recommendations, directly impacting system performance and user experience. However, existing methods face limitations in capturing user preferences due to their reliance on current context, and they struggle to handle redundant or low-confidence relationships in knowledge graphs, resulting in decreased accuracy and reliability of recommendations. Moreover, the lack of diversity in generated conversational responses affects user satisfaction to some extent. To address these challenges, this study proposes a novel conversational recommendation system that improves recommendation accuracy and enhances response diversity and coherence through the collaborative enhancement of long-term and short-term interests and diffusion optimization. Specifically, we construct user affinity networks from the dataset to establish users’ long-term stable interests, effectively combining them with short-term immediate interests in the current conversation to comprehensively capture changes in user preferences. Additionally, a diffusion model is introduced to optimize the knowledge graph by removing invalid or low-confidence relationships, enhancing its reliability and effectiveness for more accurate recommendations. Finally, we design a semantic-noise-based diffusion semantic BERT model to deeply model conversational data, improving response diversity and user satisfaction. Extensive experiments have demonstrated the effectiveness of our approach.

Emotion Recognition in Conversation (ERC) plays an important role in driving the development of human-machine interaction. Emotions can exist in multiple modalities, and multimodal ERC mainly faces two problems: (1) the noise problem in the cross-modal information fusion process, and (2) the prediction problem of less sample emotion labels that are semantically similar but different categories. To address these issues and fully utilize the features of each modality, we adopted the following strategies: first, deep emotion cues extraction was performed on modalities with strong representation ability, and feature filters were designed as multimodal prompt information for modalities with weak representation ability. Then, we designed a Multimodal Prompt Transformer (MPT) to perform cross-modal information fusion. MPT embeds multimodal fusion information into each attention layer of the Transformer, allowing prompt information to participate in encoding textual features and being fused with multi-level textual information to obtain better multimodal fusion features. Finally, we used the Hybrid Contrastive Learning (HCL) strategy to optimize the model's ability to handle labels with few samples. This strategy uses unsupervised contrastive learning to improve the representation ability of multimodal fusion and supervised contrastive learning to mine the information of labels with few samples. Experimental results show that our proposed model outperforms state-of-the-art models in ERC on two benchmark datasets.

Abstract In recent years, sentiment analysis based on aspects has become one of the research hotspots in the field of natural language processing. Aiming at the fact that the existing network model cannot fully obtain the interrelationship between sentences in the same comment and the long-distance dependence of specific aspects in the whole comment, a multilingual deep hierarchical model combining regional convolutional neural network and bidirectional LSTM network is proposed. The model obtains the time series relationship of different sentences in the comments through the regional CNN, and obtains the local features of the specific aspects in the sentence and the long-distance dependence in the whole comment through the hierarchical attention network. In addition, the model improves the word vector representation based on the gate mechanism to make the model completely independent of the language. Experimental results for different domain datasets in multi-language show that the proposed model achieves better classification results than the traditional deep network model, the network model combining with the attention mechanism and considering the relationship between sentences.