Leveraging auxiliary textual data can help with user profiling and item characterization in recommender systems (RSs). However, incomplete user and item descriptions limit the potential of textual information. To this end, we propose a graph-aware convolutional large language model (LLM) method, eliciting LLMs to summarize from a high-order interaction graph to generate fine-grained descriptions for users and items. We focus on two challenges in this paper: 1) the incompatibility between structural graph and text-aware LLMs . Although some work converts graph data into textual form using templates and sampling strategies, they limit the LLMs’ ability to maintain a global perspective on graphs for LLM reasoning. 2) the limitation of LLMs’ capability for long context. Simply describing a large-scale graph in a textual format makes LLM struggle to robustly comprehend information from lengthy contextual inputs. To bridge the gap between graph structures and LLMs, we employ the LLM as an aggregator for the graph convolution process, eliciting it to aggregate and refine the graph-based information iteratively. To mitigate the information overload associated with large-scale graphs, we segment the graph processing into manageable steps, progressively incorporating multi-hop information in a least-to-most manner. Experiments on three real-world datasets demonstrate that our method consistently outperforms state-of-the-art approaches.
Recommender systems are among the most impactful applications of artificial intelligence, serving as critical infrastructure connecting users, merchants, and platforms. However, most industrial systems remain heavily reliant on historical co-occurrence patterns and log-fitting objectives, i.e. , optimizing for past user interactions without explicitly modeling user intent. This log-fitting approach often leads to overfitting to narrow historical preferences, failing to capture users’ potential interests. As a result, it reinforces filter bubbles and long-tail phenomena, ultimately harming user experience and threatening the sustainability of the recommendation ecosystem. To address these challenges, we rethink the overall design paradigm of recommender systems and propose RecGPT, a fully integrated, production-ready framework that places user intent at the center of the recommendation pipeline. By integrating large language models (LLMs) into key stages of user interest mining, item retrieval, and explanation generation, RecGPT transforms log-fitting recommendation into an intent-centric process. To effectively align general-purpose LLMs to the above domain-specific recommendation tasks at scale, RecGPT incorporates a multi-stage training paradigm, which integrates reasoning-enhanced pre-alignment and self-training evolution, guided by a Human-LLM cooperative judge system. Currently, RecGPT has been fully deployed on the Taobao App. Online experiments demonstrate that RecGPT achieves consistent performance gains across stakeholders: users benefit from increased content diversity and satisfaction ( e.g. , CICD +5.53%, DT +4.35%), merchants and the platform gain greater exposure and conversions ( e.g. , CTR +5.65%, IPV +7.55%, DCAU +2.46%), and sustained long-term retention improvements (LT-30 +1.63%). These comprehensive improvement results across all stakeholders validate that our intent-centric design can foster a more sustainable and mutually beneficial ecosystem.
Cross-Domain Sequential Recommendation (CDSR) seeks to improve next-item prediction based on the historical behaviors of users in multiple domains. However, existing methods face two critical limitations: (1) static time-decay assumptions fail to capture short-term interest shifts, and (2) negative sampling through random perturbation undermines cross-domain consistency. To overcome these limitations, we propose DiffCDSR, a novel framework that combines diffusion-guided contrastive learning with a Position Re-weighting (PR) module. The PR module dynamically adjusts position importance, enabling fine-grained modeling of user interest drift. Meanwhile, a diffusion mechanism generates semantically perturbed, yet structure-preserving negative samples, enhancing the contrastive learning process. Experiments on three CDSR benchmarks demonstrate that DiffCDSR outperforms most state-of-the-art methods in NDCG@10 and HR@10, with ablations confirming component contributions. We also provide insights into hyperparameter impacts and discuss future directions for multimodal integration and efficient diffusion design. Project website: https://github.com/ShengGit/DiffCDSR.
Heterogeneous information networks (HINs) play an indispensable role in a wide range of domain-specific applications, from recommender systems to conversational platforms. Textual HINs are graphs with abundant textual information. Currently, most advanced approaches to mine textual features from HINs follow a “pre-training and fine-tuning” schema, which may cause a “negative transfer” problem since there is a gap between pre-training tasks and downstream tasks. We propose a prompt-learning framework P-HIN that provides a new angle to align textual information and graph information, while narrowing down the gap between the pre-trained models and various downstream tasks. To the best of our knowledge, we are among the first to introduce and exploit the idea of prompt learning to align HIN features and textual features. The proposed framework P-HIN is composed of a text encoder and a graph encoder and uses contrastive learning to align and fuse the graph–text pair. This pre-training operation naturally fits the few-shot learning setting. For the graph encoder, we introduce two graph pre-training tasks, masked node modeling and edge reconstruction, to exploit self-supervised information. During optimization, instead of handcrafted prompts, we use a learnable continuous text that enables more efficient and task-relevant transfer to downstream datasets. We consider a residual connection to use the context from the graph to prompt the text encoder. In experiments, P-HIN consistently and significantly outperforms state-of-the-art alternatives on all real-life datasets.
Large language models have facilitated the development of sophisticated dialogue planner policies. However, their reliance on LLM-based simulation and feedback for policy optimization may introduce preference bias. This article presents the first comprehensive analysis of preference bias in LLM-based dialogue planners. It systematically evaluates four state-of-the-art planning methods across three dialogue domains, utilizing multiple LLM families at varying scales. Our investigation reveals that all tested planners exhibit significant preference bias, favoring narrow strategy sets over maintaining balanced distributions. Notably, larger models paradoxically amplify bias. User simulation emerges as the primary bias driver, while diverse persona simulation proves ineffective as a mitigation strategy. Most concerning, preference bias leads planners toward ethically problematic strategies that achieve short-term success while compromising real-world effectiveness and ethical standards. These findings establish fundamental challenges for responsible deployment of LLM-based dialogue systems and provide crucial insights for developing more reliable and ethically aligned planning approaches.
