Transformers are the cornerstone of modern large language models, but their quadratic computational complexity limits efficiency in long-sequence processing. Recent advancements in Mamba, a state space model (SSM) with linear complexity, offer promising efficiency gains but suffer from unstable contextual learning and multitask generalization. Some works conduct layer-level hybrid structures that combine Transformer and Mamba layers, aiming to make full use of both advantages. This paper proposes TransMamba, a novel sequence-level hybrid framework that unifies Transformer and Mamba through shared parameter matrices (QKV and CBx), and thus could dynamically switch between attention and SSM mechanisms at different token lengths and layers. We design the Memory Converter to bridge Transformer and Mamba by converting attention outputs into SSM-compatible states, ensuring seamless information flow at TransPoints where the transformation happens. The TransPoint scheduling is also thoroughly explored for balancing effectiveness and efficiency. We conducted extensive experiments demonstrating that TransMamba achieves superior training efficiency and performance compared to single and hybrid baselines, and validated the deeper consistency between Transformer and Mamba paradigms at sequence level, offering a scalable solution for next-generation language modeling.

Mingze Xu, Zijing Zhao, Qiming Peng, Houwen Peng, Han Hu, Zhanhui Kang, Yuxing Han. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recommendations usually focus on immediate accuracy metrics like Click-Through Rate (CTR), ignoring user long-term metrics. User retention, which reflects the percentage of today’s users who will return to the system in the next few days, should be paid more attention to. However, most existing methods did not focus on user retention, since their complexity and uncertainty make it extremely hard to discover why a user will or will not return to a system. Recently, a few pioneers have optimized user retention, focusing solely on accuracy without delving into its underlying rationale. This is primarily due to the absence of explicit supervised signals. In this work, we design a Behavior-wise Contrastive Multi-Instance Learning (BCMIL) module, which jointly models clicked and impressed items to capture interpretable user retention. Specifically, we conduct in-depth analyses in real-world scenarios to discover implicit retention-related supervised signals. To model these signals, we design a Forward Supervised Signals Extractor (FSSE) that utilizes a heterogeneous graph, enhancing the reliability of user retention. To mitigate randomness and uncertainty, we propose a Backward Supervised Signals Stabilizer (BSSS) that utilizes overlooked label-part behaviors within each training window to retrospectively guide the training process. Offline and online evaluations of an industrial system verify the effectiveness of our methods.

Recommender system (RS) aims to capture personalized preferences from massive user behaviors, making them pivotal in the era of information explosion. However, the presence of “information cocoons,” interaction sparsity, cold-start problem, and feedback loops inherent in RS make users interact with a limited number of items. Conventional recommendation algorithms typically focus on the positive historical behaviors, while neglecting the essential role of negative feedback in user preference understanding. As a promising but easy-to-ignored area, negative sampling is proficient in revealing the genuine negative aspect inherent in user behaviors, emerging as an inescapable procedure in RS. In this survey, we first discuss existing user feedback, the critical role of negative sampling and the optimization objectives in RS, and thoroughly analyze challenges that consistently impede its progress. Then, we conduct an extensive literature review on the existing negative sampling strategies in RS and classify them into five categories with their discrepant techniques. Finally, we detail the insights of the tailored negative sampling strategies in diverse RS scenarios and outline an overview of the prospective research directions toward which the community may engage and benefit.

Contrastive learning is a prevalent technique in self-supervised vision representation learning, typically generating positive pairs by applying two data augmentations to the same image. Designing effective data augmentation strategies is crucial for the success of contrastive learning. Inspired by the story of the blind men and the elephant, we introduce JointCrop and JointBlur. These methods generate more challenging positive pairs by leveraging the joint distribution of the two augmentation parameters, thereby enabling contrastive learning to acquire more effective feature representations. To the best of our knowledge, this is the first effort to explicitly incorporate the joint distribution of two data augmentation parameters into contrastive learning. As a plug-and-play framework without additional computational overhead, JointCrop and JointBlur enhance the performance of SimCLR, BYOL, MoCo v1, MoCo v2, MoCo v3, SimSiam, and Dino baselines with notable improvements.

Large language models (LLMs) have demonstrated remarkable performance in multimodal tasks even with frozen LLM Block and only a few trainable parameters. However, the underlying mechanisms of how LLMs enhance multimodal performance remains unclear. In this work, we focus on the phenomenon that ``Merely concatenating a frozen LLM block to the Vision Transformer (ViT) encoder can yield significant performance enhancements. Moreover, the choice of LLM block and insertion position can have a substantial impact, leading to varying degrees of improvement''. We analyze the optimization of the training process from the perspective of gradient dynamics and find that frozen LLM blocks act as gradient coherence rectifiers, aligning the gradients of different samples more closely during training. Furthermore, we demonstrate that the representation similarity between the inserted LLM block and the adjacent ViT block influences performance, with greater similarity tending to yield larger positive gains. Through these findings, we can justify the selection of suitable LLM blocks to be inserted at appropriate positions, and introduce additional gradient backpropagation paths by incorporating LLM blocks, could improve the performance of vanilla ViT through the rectification effect of gradient consistency during the training process, without the need to add LLM blocks during inference. Our experiments demonstrate the effectiveness of this strategy, making the practical application of the gradient rectification effect feasible.

Catastrophic forgetting remains a formidable obstacle to building an omniscient model in large language models (LLMs).Despite the pioneering research on task-level forgetting in LLM fine-tuning, there is scant focus on forgetting during pre-training.We systematically explored the existence and measurement of forgetting in pre-training, questioning traditional metrics such as perplexity (PPL) and introducing new metrics to better detect entity memory retention.Based on our revised assessment of forgetting metrics, we explored low-cost, straightforward methods to mitigate forgetting during the pre-training phase.In addition, we carefully analyzed the learning curves, offering insights into the dynamics of forgetting.Extensive evaluations and analyses on forgetting of pre-training could facilitate future research on LLMs.

Multimodal Large Language Models (MLLMs) hallucinate, resulting in an emerging topic of visual hallucination evaluation (VHE). This paper contributes a ChatGPT-Prompted visual hallucination evaluation Dataset (PhD) for objective VHE at a large scale. The essence of VHE is to ask an MLLM questions about specific images to assess its susceptibility to hallucination. Depending on what to ask (objects, attributes, sentiment, etc.) and how the questions are asked, we structure PhD along two dimensions, i.e. task and mode. Five visual recognition tasks, ranging from low-level (object/attribute recognition) to middle-level (sentiment/position recognition and counting), are considered. Besides a normal visual QA mode, which we term PhD-base, PhD also asks questions with specious context (PhD-sec) or with incorrect context (PhD-icc), or with AI-generated counter common sense images (PhD-ccs). We construct PhD by a ChatGPT-assisted semi-automated pipeline, encompassing four pivotal modules: task-specific hallucinatory item (hitem) selection, hitem-embedded question generation, specious/incorrect context generation, and counter-common-sense (CCS) image generation. With over 14k daily images, 750 CCS images and 102k VQA triplets in total, PhD reveals considerable variability in MLLMs’ performance across various modes and tasks, offering valuable insights into the nature of hallucination. As such, PhD stands as a potent tool not only for VHE but may also play a significant role in the refinement of MLLMs.

An Wang, Xingwu Sun, Ruobing Xie, Shuaipeng Li, Jiaqi Zhu, Zhen Yang, Pinxue Zhao, Weidong Han, Zhanhui Kang, Di Wang, Naoaki Okazaki, Cheng-zhong Xu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The goal of transferable sequential recommendation (TSR) is to improve the performance of sequential recommenders in multiple target domains leveraging knowledge transferred from source domains. Most existing transferable sequential recommenders rely on item modality information but pay insufficient attention to the significant differences across domains (i.e., distinctive user behavior characteristics) in recommender systems. Besides, they are typically built on specific pre-trained model structures, limiting their model generalization to various different downstream domains and sequential recommenders in practical applications. This paper aims to address these issues by exploring the potential of mixture-of-LoRAs based domain adaption and an ensemble of sequential recommenders for learning generalizable and transferable domain-adaptive item representations. We propose MoL-TSR, a pre-trained plug-in mixture of low-rank adaption model for TSR. Specifically, we first devise a mixture-ofLoRAs based domain projector to produce domain-adaptive item representations based on both item modality and user behavior information for recommendation, enhancing the item discrimination and domain transfer ability. Then, we pre-train an ensemble of sequential recommendation models with contrastive learning to enhance the model-aspect generalization. As a general pretrained model, the mixture-of-LoRAs based domain projector in our MoL-TSR is a plug-in item semantic representation learning module, which is convenient to migrate into other transferable sequential recommenders to enhance item semantics better suited for recommendation. Extensive experiments demonstrate the effectiveness and model generalization of MoL-TSR, promising a practical solution for learning generalizable and transferable item semantic representations in recommendation. Our code is available at https://github.com/RUCAIBox/MoL-TSR.

The Text-to-Video Retrieval (T2VR) task aims to retrieve unlabeled videos by textual queries with the same semantic meanings. Recent CLIP-based approaches have explored two frameworks: Two-Tower versus Single-Tower framework, yet the former suffers from low effectiveness, while the latter suffers from low efficiency. In this study, we explore a new Hybrid-Tower framework that can hybridize the advantages of the Two-Tower and Single-Tower framework, achieving high effectiveness and efficiency simultaneously. We propose a novel hybrid method, Fine-grained Pseudo-query Interaction and Generation for T2VR, ie, PIG, which includes a new pseudo-query generator designed to generate a pseudo-query for each video. This enables the video feature and the textual features of pseudo-query to interact in a fine-grained manner, similar to the Single-Tower approaches to hold high effectiveness, even before the real textual query is received. Simultaneously, our method introduces no additional storage or computational overhead compared to the Two-Tower framework during the inference stage, thus maintaining high efficiency. Extensive experiments on five commonly used text-video retrieval benchmarks demonstrate that our method achieves a significant improvement over the baseline, with an increase of $1.6\% \sim 3.9\%$ in R@1. Furthermore, our method matches the efficiency of Two-Tower models while achieving near state-of-the-art performance, highlighting the advantages of the Hybrid-Tower framework.

Mixture-of-Experts (MoE) models mostly use a router to assign tokens to specific expert modules, activating only partial parameters and often outperforming dense models. We argue that the separation between the router’s decision-making and the experts’ execution is a critical yet overlooked issue, leading to suboptimal expert selection and learning. To address this, we propose Autonomy-of-Expert (AoE), a novel MoE paradigm in which experts autonomously select themselves to process inputs. AoE is based on the insight that an expert is aware of its own capacity to effectively process a token, an awareness reflected in the scale of its internal activations. In AoE, routers are removed; instead, experts pre-compute internal activations for inputs and are ranked based on their activation norms. Only the top-ranking experts proceed with the forward pass, while the others abort. The overhead of pre-computing activations is reduced through a low-rank weight factorization. This self-evaluating-then-partner-comparing approach ensures improved expert selection and effective learning. We pre-train language models having 700M up to 4B parameters, demonstrating that AoE outperforms traditional MoE models with comparable efficiency.

Low-precision training is considered an effective strategy for reducing both training and downstream inference costs. Previous scaling laws for precision mainly focus on integer quantization, which pay less attention to the constituents in floating-point (FP) quantization, and thus cannot well fit the LLM losses in this scenario. In contrast, while FP quantization training is more commonly implemented in production, it's research has been relatively superficial. In this paper, we thoroughly explore the effects of FP quantization targets, exponent bits, mantissa bits, and the calculation granularity of the scaling factor in FP quantization training performance of LLM models. In addition to an accurate FP quantization unified scaling law, we also provide valuable suggestions for the community: (1) Exponent bits contribute slightly more to the model performance than mantissa bits. We provide the optimal exponent-mantissa bit ratio for different bit numbers, which is available for future reference by hardware manufacturers; (2) We discover the formation of the critical data size in low-precision LLM training. Too much training data exceeding the critical data size will inversely bring in degradation of LLM performance; (3) The optimal FP quantization precision is directly proportional to the computational power, but within a wide computational power range. We estimate that the best cost-performance precision should lie between 4-8 bits.

In retrieval-augmented generation (RAG) systems, Large Language Models (LLMs) have been shown to be effective for re-ranking. However, existing research often prioritizes passage relevance over reliability, which can result in the incorporation of conflicting information and the generation of ambiguous responses. This issue becomes particularly pronounced when addressing inter-context knowledge conflicts, where candidate documents present contradictory information that may mislead the model. To mitigate this problem, we propose a novel cross-validation re-ranking technique designed specifically to resolve inter-context knowledge conflicts during the retrieval process. We also develop a new dataset, ContraPRT, to evaluate the ability of models to rank passages containing conflicting knowledge. Experimental results using GPT-4 and LlaMA3-70B demonstrate that our approach not only effectively filters out conflicting information but also ensures accurate passage rankings, thereby providing reliable supplementary knowledge for the generation module.

Recent advances in large vision-language models (LVLMs) have showcased their remarkable capabilities across a wide range of multimodal vision-language tasks. However, these models remain vulnerable to visual adversarial attacks, which can substantially compromise their performance. In this paper, we introduce F3, a novel adversarial purification framework that employs a counterintuitive ''fighting fire with fire'' strategy: intentionally introducing simple perturbations to adversarial examples to mitigate their harmful effects. Specifically, F3 leverages cross-modal attentions derived from randomly perturbed adversary examples as reference targets. By injecting noise into these adversarial examples, F3 effectively refines their attention, resulting in cleaner and more reliable model outputs. Remarkably, this seemingly paradoxical approach of employing noise to counteract adversarial attacks yields impressive purification results. Furthermore, F3 offers several distinct advantages: it is training-free and straightforward to implement, and exhibits significant computational efficiency improvements compared to existing purification methods. These attributes render F3 particularly suitable for large-scale industrial applications where both robust performance and operational efficiency are critical priorities. The code is available at https://github.com/btzyd/F3.

Large vision-language models (LVLMs) have demonstrated exceptional performance on complex multimodal tasks. However, they continue to suffer from significant hallucination issues, including object, attribute, and relational hallucinations. To accurately detect these hallucinations, we investigated the variations in cross-modal attention patterns between hallucination and non-hallucination states. Leveraging these distinctions, we developed a lightweight detector capable of identifying hallucinations. Our proposed method, Detecting Hallucinations by Cross-modal Attention Patterns (DHCP), is straightforward and does not require additional LVLM training or extra LVLM inference steps. Experimental results show that DHCP achieves remarkable performance in hallucination detection. By offering novel insights into the identification and analysis of hallucinations in LVLMs, DHCP contributes to advancing the reliability and trustworthiness of these models. The code is available at https://github.com/btzyd/DHCP.

Yudong Zhang, Ruobing Xie, Jiansheng Chen, Xingwu Sun, Zhanhui Kang, Yu Wang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Large Language Models (LLMs) have emerged as a new paradigm for recommendation by converting interacted item history into language modeling. However, constrained by the limited context length of LLMs, existing approaches have to truncate item history in the prompt, focusing only on recent interactions and sacrificing the ability to model long-term history. To enable LLMs to model long histories, we pursue a concise embedding representation for items and sessions. In the LLM embedding space, we construct an item's embedding by aggregating its textual token embeddings; similarly, we construct a session's embedding by aggregating its item embeddings. While efficient, this way poses two challenges since it ignores the temporal significance of user interactions and LLMs do not natively interpret our custom embeddings. To overcome these, we propose PatchRec, a multi-grained patch training method consisting of two stages: (1) Patch Pre-training, which familiarizes LLMs with aggregated embeddings -- patches, and (2) Patch Fine-tuning, which enables LLMs to capture time-aware significance in interaction history. Extensive experiments show that PatchRec effectively models longer behavior histories with improved efficiency. This work facilitates the practical use of LLMs for modeling long behavior histories.

Recently, many efforts have been devoted to building effective sequential recommenders.Despite their effectiveness, these methods typically develop a single model to serve all users.However, our empirical studies reveal that different sequential encoders have intrinsic architectural biases and tend to focus on specific behavioral patterns, i.e., particular frequency range of user behavior sequences.For example, the Self-Attention module is essentially a low-pass filter, focusing on low-frequency information while neglecting the high-frequency details.This evidently limits their ability to capture diverse user patterns, leading to suboptimal recommendations.To tackle this problem, we present FamouSRec, a Frequency-Augmented Mixture-of-Heterogeneous-Experts Framework for personalized Recommendations.Our approach builds an MoEbased recommender system, integrating the strengths of various experts to achieve diversified user modeling.For developing the MoE framework, as the key to our approach, we instantiate experts with various model architectures, aiming to leverage their inherent architectural biases and capture diverse behavioral patterns.For selecting appropriate experts to serve individuals, we introduce a frequency-augmented router.It first identifies frequency components in user behavior sequences that are suited for expert encoding, and then conducts customized routing based on the informativeness of these components.Building on this framework, we further propose two novel contrastive tasks to enhance expert

With the thriving of the pre-trained language model (PLM) widely verified in various NLP tasks, pioneer efforts attempt to explore the possible cooperation of the general textual information in PLM with the personalized behavioral information in user historical behavior sequences to enhance sequential recommendation (SR). However, despite the commonalities of input format and task goal, there are huge gaps between the behavioral and textual information, which obstruct thoroughly modeling SR as language modeling via PLM. To bridge the gap, we propose a novel unified pre-trained language model enhanced sequential recommendation (UPSR) that thoroughly transfers the next item prediction task to a text generation task, aiming to build a unified pre-trained recommendation model for multi-domain recommendation tasks. We formally design five key indicators, namely naturalness, domain consistency, informativeness, noise and ambiguity, and text length, to guide the text \(\rightarrow\) item adaptation (selecting appropriate text to form the item textual representation) and behavior sequence \(\rightarrow\) text sequence adaptation (transferring the sequence of item textual representations into a text sequence) differently for pre-training and fine-tuning stages, which are essential but under-explored by previous works. In experiments, we conduct extensive evaluations on seven datasets with both supervised and zero-shot settings and achieve the overall best performance. Comprehensive model analyses also provide valuable insights for behavior modeling via PLM, shedding light on large pre-trained recommendation models. The source codes will be released in the future.

Classical sequential recommendation models generally adopt ID embeddings to store knowledge learned from user historical behaviors and represent items. However, these unique IDs are challenging to be transferred to new domains. With the thriving of pre-trained language model (PLM), some pioneer works adopt PLM for pre-trained recommendation, where modality information is considered universal across domains via PLM. Unfortunately, the behavioral information in ID embeddings is verified to currently dominate in recommendation compared to modality information and thus limits these models’ performance. In this work, we propose a novel ID-centric recommendation pre-training paradigm (IDP), which directly transfers informative ID embeddings learned in pre-training domains to item representations in new domains. Specifically, in pre-training stage, besides the ID-based sequential recommendation model, we also build a Cross-domain ID-matcher (CDIM) learned by both behavioral and modality information. In the tuning stage, modality information of new domain items is regarded as a cross-domain bridge built by CDIM. They first adopted to retrieve behaviorally and semantically similar items from pre-training domains using CDIM. Next, these retrieved items’ pre-trained ID embeddings are directly adopted to generate downstream new items’ embeddings. Through extensive experiments on real-world datasets, we demonstrate that our proposed model significantly outperforms all baselines.

Despite the great success of large language models (LLMs) in various tasks, they suffer from generating hallucinations. We introduce Truth Forest, a method that enhances truthfulness in LLMs by uncovering hidden truth representations using multi-dimensional orthogonal probes. Specifically, it creates multiple orthogonal bases for modeling truth by incorporating orthogonal constraints into the probes. Moreover, we introduce Random Peek, a systematic technique considering an extended range of positions within the sequence, reducing the gap between discerning and generating truth features in LLMs. By employing this approach, we improved the truthfulness of Llama-2-7B from 40.8% to 74.5% on TruthfulQA. Likewise, significant improvements are observed in fine-tuned models. We conducted a thorough analysis of truth features using probes. Our visualization results show that orthogonal probes capture complementary truth-related features, forming well-defined clusters that reveal the inherent structure of the dataset.

Instruction-following is particularly crucial for large language models (LLMs) to support diverse user requests. While existing work has made progress in aligning LLMs with human preferences, evaluating their capabilities on instruction-following remains a challenge due to complexity and diversity of real-world user instructions. While existing evaluation methods focus on general skills, they suffer from two main shortcomings, i.e., lack of fine-grained task-level evaluation and reliance on singular instruction expression. To address these problems, this paper introduces DINGO, a fine-grained and diverse instruction-following evaluation dataset that has two main advantages: (1) DINGO is based on a manual annotated, fine-grained and multi-level category tree with 130 nodes derived from real-world user requests; (2) DINGO includes diverse instructions, generated by both GPT-4 and human experts. Through extensive experiments, we demonstrate that DINGO can not only provide more challenging and comprehensive evaluation for LLMs, but also provide task-level fine-grained directions to further improve LLMs.

Pioneering efforts have verified the effectiveness of the diffusion models in exploring the informative uncertainty for recommendation. Considering the difference between recommendation and image synthesis tasks, existing methods have undertaken tailored refinements to the diffusion and reverse process. However, these approaches typically use the highest-score item in corpus for user interest prediction, leading to the ignorance of the user's generalized preference contained within other items, thereby remaining constrained by the data sparsity issue. To address this issue, this paper presents a novel Plug-in Diffusion Model for Recommendation (PDRec) framework, which employs the diffusion model as a flexible plugin to jointly take full advantage of the diffusion-generating user preferences on all items. Specifically, PDRec first infers the users' dynamic preferences on all items via a time-interval diffusion model and proposes a Historical Behavior Reweighting (HBR) mechanism to identify the high-quality behaviors and suppress noisy behaviors. In addition to the observed items, PDRec proposes a Diffusion-based Positive Augmentation (DPA) strategy to leverage the top-ranked unobserved items as the potential positive samples, bringing in informative and diverse soft signals to alleviate data sparsity. To alleviate the false negative sampling issue, PDRec employs Noise-free Negative Sampling (NNS) to select stable negative samples for ensuring effective model optimization. Extensive experiments and analyses on four datasets have verified the superiority of the proposed PDRec over the state-of-the-art baselines and showcased the universality of PDRec as a flexible plugin for commonly-used sequential encoders in different recommendation scenarios. The code is available in https://github.com/hulkima/PDRec.

This paper studies vision-language (V&L) pre-training for deep cross-modal representations. Recently, pre-trained V&L models have shown great success in V&L tasks. However, most existing models apply multi-modal encoders to encode the image and text, at the cost of high training complexity because of the input sequence length. In addition, they suffer from noisy training corpora caused by V&L mismatching. In this work, we propose a lightweight vision-language pre-training (LightVLP) for efficient and effective V&L pre-training. First, we design a new V&L framework with two autoencoders. Each autoencoder involves an encoder, which only takes in unmasked tokens (removes masked ones), as well as a lightweight decoder that reconstructs the masked tokens. Besides, we mask and remove large portions of input tokens to accelerate the training. Moreover, we propose a gated interaction mechanism to cope with noise in aligned image-text pairs. As for a matched image-text pair, the model tends to apply cross-modal representations for reconstructions. By contrast, for an unmatched pair, the model conducts reconstructions mainly using uni-modal representations. Benefiting from the above-mentioned designs, our base model shows competitive results compared to ALBEF while saving 44% FLOPs. Further, we compare our large model with ALBEF under the setting of similar FLOPs on six datasets and show the superiority of LightVLP. In particular, our model achieves 2.2% R@1 gains on COCO Text Retrieval and 1.1% on refCOCO+.

The graph-based recommendation has achieved great success in recent years. However, most existing graph-based recommendations focus on capturing user preference based on positive edges/feedback, while ignoring negative edges/feedback (e.g., dislike, low rating) that widely exist in real-world recommender systems. How to utilize negative feedback in graph-based recommendations still remains underexplored. In this study, we first conducted a comprehensive experimental analysis and found that (1) existing graph neural networks are not well-suited for modeling negative feedback, which acts as a high-frequency signal in a user-item graph. (2) The graph-based recommendation suffers from the representation degeneration problem. Based on the two observations, we propose a novel model that models positive and negative feedback from a frequency filter perspective called Dual-frequency Graph Neural Network for Sign-aware Recommendation (DFGNN). Specifically, in DFGNN, the designed dual-frequency graph filter (DGF) captures both low-frequency and high-frequency signals that contain positive and negative feedback. Furthermore, the proposed signed graph regularization is applied to maintain the user/item embedding uniform in the embedding space to alleviate the representation degeneration problem. Additionally, we conduct extensive experiments on real-world datasets and demonstrate the effectiveness of the proposed model. Codes of our model will be released upon acceptance.

Multi-modal recommender systems (MRSs) are pivotal in diverse online web platforms and have garnered considerable attention in recent years. However, previous studies overlook the challenges of (1)noisy multi-modal content, (2) noisy user feedback, and (3) aligning multi-modal content and user feedback. To tackle these challenges, we propose Denoising and Aligning Multi-modal Recommender System (DA-MRS). To mitigate noise in multi-modal content, DA-MRS first constructs item-item graphs determined by consistent content similarity across modalities. To denoise user feedback, DA-MRS associates the probability of observed feedback with multi-modal content and devises a denoised BPR loss. Furthermore, DA-MRS implements Alignment guided by User preference to enhance task-specific item representation and Alignment guided by graded Item relations to provide finer-grained alignment. Extensive experiments verify that DA-MRS is a plug-and-play framework and achieves significant and consistent improvements across various datasets, backbone models, and noisy scenarios.

Sequential recommendation (SR) has seen significant advancements with the help of Pre-trained Language Models (PLMs). Some PLM-based SR models directly use PLM to encode user historical behavior’s text sequences to learn user representations, while there is seldom an in-depth exploration of the capability and suitability of PLM in behavior sequence modeling. In this work, we first conduct extensive model analyses between PLMs and PLM-based SR models, discovering great underutilization and parameter redundancy of PLMs in behavior sequence modeling. Inspired by this, we explore different lightweight usages of PLMs in SR, aiming to maximally stimulate the ability of PLMs for SR while satisfying the efficiency and usability demands of practical systems. We discover that adopting behavior-tuned PLMs for item initializations of conventional ID-based SR models is the most economical framework of PLM-based SR, which would not bring in any additional inference cost but could achieve a dramatic performance boost compared with the original version. Extensive experiments on five datasets show that our simple and universal framework leads to significant improvement compared to classical SR and SOTA PLM-based SR models without additional inference costs. Our code can be found in https://github.com/777pomingzi/Rethinking-PLM-in-RS.

Conversational recommender systems (CRSs) are able to elicit user preferences through multi-turn dialogues. They typically incorporate external knowledge and pre-trained language models to capture the dialogue context. Most CRS approaches, trained on benchmark datasets, assume that the standard items and responses in these benchmarks are optimal. However, they overlook that users may express negative emotions with the standard items and may not feel emotionally engaged by the standard responses. This issue leads to a tendency to replicate the logic of recommenders in the dataset instead of aligning with user needs. To remedy this misalignment, we introduce empathy within a CRS. With empathy we refer to a system’s ability to capture and express emotions. We propose an empathetic conversational recommender (ECR) framework.

Large language models (LLMs) have proven to be very superior to conventional methods in various tasks.However, their expensive computations and high memory requirements are prohibitive for deployment.Model quantization is an effective method for reducing this overhead.The problem is that in most previous works, the quantized model was calibrated using a few samples from the training data, which might affect the generalization of the quantized LLMs to unknown cases and tasks.Hence in this work, we explore an important question: Can we design a data-free quantization method for LLMs to guarantee its generalization performance?In this work, we propose EasyQuant, a trainingfree and data-free weight-only quantization algorithm for LLMs.Our observation indicates that two factors: outliers in the weight and quantization ranges, are essential for reducing the quantization error.Therefore, in EasyQuant, we leave the outliers (less than 1%) unchanged and optimize the quantization range to reduce the reconstruction error.With these methods, we surprisingly find that EasyQuant achieves comparable performance to the original model.Since EasyQuant does not depend on any training data, the generalization performance of quantized LLMs are safely guaranteed.Moreover, EasyQuant can be implemented in parallel so that the quantized model could be attained in a few minutes even for LLMs over 100B.To our best knowledge, we are the first work that achieves comparable performance with datadependent algorithms under a data-free setting and our algorithm runs over 10 times faster than the data-dependent methods.

The success of CutMix in image classification has sparked interest in saliency-based mix augmentation methods, which refer to detecting saliency regions to generate more valid images. However, existing mix works either require external tools to locate saliency regions, or rely on additional complex optimization policy for generating new images, which limits their application ranges. To address these deficiencies, we propose Gradient Saliency-based Mix (GradSalMix), a simple yet more general mix augmentation, whose operations are all based on the gradients of the training neural network itself. Specifically, we first locate the saliency regions of two images via their gradients of manifolds, and then directly migrate the region, sampled around the center with a large gradient response value, from one image to another. Afterwards, the labels of images are weighted by their accumulated gradient values for new soft labels, which are shown more accurate than the ones weighted by area ratio. The experimental results show that our proposed method outperforms previous works, in terms of accuracy and robustness against adversarial attacks, on four image classification benchmarks. Moreover, extensive experiments on object detection and point cloud classification also verify the superiority and generality of our method.

This paper introduces ChinaOpen, a dataset sourced from Bilibili, a popular Chinese video-sharing website, for open-world multimodal learning. While the state-of-the-art multimodal learning networks have shown impressive performance in automated video annotation and cross-modal video retrieval, their training and evaluation are primarily conducted on YouTube videos with English text. Their effectiveness on Chinese data remains to be verified. In order to support multimodal learning in the new context, we construct ChinaOpen-50k, a webly annotated training set of 50k Bilibili videos associated with user-generated titles and tags. Both text-based and content-based data cleaning are performed to remove low-quality videos in advance. For a multi-faceted evaluation, we build ChinaOpen-1k, a manually labeled test set of 1k videos. Each test video is accompanied with a manually checked user title and a manually written caption. Besides, each video is manually tagged to describe objects / actions / scenes shown in the visual content. The original user tags are also manually checked. Moreover, with all the Chinese text translated into English, ChinaOpen-1k is also suited for evaluating models trained on English data. In addition to ChinaOpen, we propose Generative Video-to-text Transformer (GVT) for Chinese video captioning. We conduct an extensive evaluation of the state-of-the-art single-task / multi-task models on the new dataset, resulting in a number of novel findings and insights.

Position Embedding (PE) is essential for transformer to capture the sequence ordering of input tokens.Despite its general effectiveness verified in Natural Language Processing (NLP) and Computer Vision (CV), its application in cross-modal tasks remains unexplored and suffers from two challenges: 1) the input text tokens and image patches are not aligned; 2) the encoding space of each modality is different, making it unavailable for feature comparison.In this paper, we propose a unified position embedding method for these problems, called AnChor-basEd Relative Position Embedding (ACE-RPE), in which we first introduce an anchor locating mechanism to bridge the semantic gap and locate anchors from different modalities.Then we conduct the distance calculation of each text token and image patch by computing their shortest paths from the located anchors.Last, we embed the anchor-based distance to guide the computation of crossattention.In this way, it calculates cross-modal relative position embeddings for cross-modal transformer.Benefiting from ACE-RPE, our method obtains new SOTA results on a wide range of benchmarks, such as Image-Text Retrieval on MS-COCO and Flickr30K, Visual Entailment on SNLI-VE, Visual Reasoning on NLVR2 and Weakly-supervised Visual Grounding on RefCOCO+.

Past years, with the growth of smart-phones and applications, APP market has become an important mobile internet portal. As an important function in application market, APP search gains lots of attentions.However, mismatch between queries and APP is the most critical problem in APP search because of less text within term matching search engine. In this talk, we describe a semantic matching architecture in APP search--which mining topics and tags in big data. It enriches query and APP representations with topics and tags to achieve semantic matching in search. Some challenge must be considered: 1) How to extract tag-APP relationship from large web text. 2) How to use machine learning technologies to process de-noising and computing confidence. 3) How to hybrid ranking apps retrieved by different matching method. These will be introduced in some of our related works and as examples to describe how semantic matching is used in Tencent MyApp, an application market which serving hundreds of millions of users.