Aixin Sun

The Second International Workshop on Data Quality-Aware Multimodal Recommendation (DaQuaMRec) addresses the increasingly important challenge of understanding how multimodal data quality affects the robustness, fairness, explainability, and practical utility of modern recommender systems. As multimodal recommendation evolves through deep representation learning, graph-based architectures, self-supervision, and foundation-model-based pipelines, the focus cannot remain solely on building stronger models. Increasing attention must also be devoted to the quality, completeness, alignment, and reliability of the multimodal data on which these systems depend. Recent studies have highlighted the impact of noisy signals, missing modalities, semantic misalignment, low-quality features, and multimodal bias, showing that these issues directly influence both system performance and trustworthiness. Building on the success of the first DaQuaMRec workshop at ACM RecSys 2025, this second edition aims to consolidate a research community around these emerging challenges, stimulate work on their assessment and mitigation, and promote reproducible evaluation practices. The workshop will combine invited talks, paper presentations, and interactive discussion sessions to foster collaboration across academia and industry and to articulate concrete future directions for data-centric multimodal recommendation.

Conversational Recommender Systems (CRSs) aim to extract user preferences from natural language interactions and deliver personalized recommendations through conversations. In multi-turn conversations, these systems often generate low-confidence responses when user intent is unclear, leading to inappropriate replies and a degraded user experience. However, existing CRSs rarely assess the confidence of generated responses or detect their semantic uncertainty . Additionally, limited modeling of item-based semantic relevance hinders the system’s ability to produce contextually coherent and preference-aligned recommendations. To this end, we propose a Semantic-enhanced CRS (SCRS) that explicitly models response-based semantic uncertainty and item-based semantic relevance to enhance both response and recommendation quality. Specifically, in conversation tasks, SCRS samples multiple candidate responses and estimates their semantic uncertainty by combining generation probabilities with semantic analysis, serving as a constraint to guide fine-tuning and improve response quality. In recommendation tasks, SCRS enriches item representations by integrating textual knowledge and calculates semantic relevance between items, then applies a dynamic optimization strategy to refine the recommendation process, generating items better aligned with user preferences. Extensive experiments on two public datasets demonstrate that SCRS achieves significant improvements in both recommendation accuracy and response quality over state-of-the-art baselines, validating the effectiveness of our proposed method.

Many studies in recommender systems (RecSys) adopt a general problem definition, that is , to recommend preferred items to users based on past interactions. Such abstraction often lacks the domain-specific nuances necessary for practical deployment. However, models are frequently evaluated using datasets collected from online recommender platforms, which inherently reflect domain or task specificities. In this article, we analyze RecSys task formulations, emphasizing key components such as input-output structures, temporal dynamics, and candidate item selection. All these factors directly impact offline evaluation. We further examine the complexities of user-item interactions, including decision-making costs, multi-step engagements, and unobservable interactions, which may influence model design. Additionally, we explore the balance between task specificity and model generalizability, highlighting how well-defined task formulations serve as the foundation for robust evaluation and effective solution development. By clarifying task definitions and their implications, this work provides a structured perspective on RecSys research. The goal is to help researchers better navigate the field, particularly in understanding specificities of the RecSys tasks and ensuring fair and meaningful evaluations.

Existing recommendation methods struggle to model users’ multifaceted preferences due to the diversity and volatility of user behavior, as well as the inherent uncertainty and ambiguity of item themes in practical scenarios. Multi-interest recommendation addresses this challenge by explicitly extracting multiple interest representations from users’ historical interactions, enabling fine-grained preference modeling and more accurate recommendations. It has attracted considerable attention in recommendation research. However, current recommendation surveys have either delved into specific recommendation tasks and downstream applications or focused on approaches that model users and items as single representations with cutting-edge techniques, overlooking users’ diverse preferences and the multifaceted aspects of items. In this work, we systematically review the progress, solutions, challenges, and future directions of multi-interest recommendation by answering the following three questions: (1) Why is multi-interest modeling significantly important for recommendation? (2) What aspects are focused on by multi-interest modeling in recommendation? and (3) How can multi-interest modeling be applied, along with the technical details of the representative modules? We hope that this survey establishes a fundamental framework and delivers a preliminary overview for researchers interested in this field and committed to further exploration. The implementation of multi-interest recommendation summarized in this survey is maintained at https://github.com/WHUIR/Multi-Interest-Recommendation-A-Survey .

Diffusion models offer appealing properties for language generation, such as parallel decoding and iterative refinement, but the discrete and highly structured nature of text challenges the direct application of diffusion principles. In this paper, we revisit diffusion language modeling from the view of diffusion process and language modeling, and outline five properties that separate diffusion mechanics from language-specific requirements. We first categorize existing approaches into continuous diffusion in embedding space and discrete diffusion over tokens. We then show that each satisfies only part of the five essential properties and therefore reflects a structural trade-off. Through analyses of recent large diffusion language models, we identify two central issues: (i) uniform corruption does not respect how information is distributed across positions, and (ii) token-wise marginal training cannot capture multi-token dependencies during parallel decoding. These observations motivate diffusion processes that align more closely with the structure of text, and encourage future work toward more coherent diffusion language models.