Multi-hop question answering (MHQA) enables accurate answers to complex queries by retrieving and reasoning over evidence dispersed across multiple documents. Existing MHQA approaches mainly rely on iterative retrieval-augmented generation, which suffer from the following two major issues. On one hand, existing methods prematurely commit to surface-level entities rather than underlying reasoning structures, making question decomposition highly vulnerable to lexical ambiguity. On the other hand, existing methods overlook the logical dependencies among reasoning steps, resulting in uncoordinated execution. To address these issues, we propose STRIDE, a framework that separates strategic planning, dynamic control, and grounded execution. At its core, a Meta-Planner first constructs an entity-agnostic reasoning skeleton to capture the abstract logic of the query, thereby deferring entity grounding until after the reasoning structure is established, which mitigates disambiguation errors caused by premature lexical commitment. A Supervisor then orchestrates sub-question execution in a dependency-aware manner, enabling efficient parallelization where possible and sequential coordination when necessary. By dynamically deciding whether to retrieve new evidence or infer from existing facts, it avoids redundant queries and error propagation, while fusing cross-branch information and reformulating failed queries to enhance robustness. Grounded fact extraction and logical inference are delegated to specialized execution modules, ensuring faithfulness through explicit separation of retrieval and reasoning. While STRIDE is compatible with any large language models (LLMs), off-the-shelf open-source LLMs underperform closed-source counterparts in its structured reasoning pipeline. To close this gap, we further propose STRIDE-FT, a modular fine-tuning framework that uses self-generated execution trajectories from STRIDE, requiring neither human annotations nor stronger teacher models. Experiments show that STRIDE achieves robust and accurate reasoning on MHQA benchmarks, while STRIDE-FT effectively enhances open-source LLMs.

Large Language Models (LLMs) have demonstrated strong potential for generative recommendation by leveraging rich semantic knowledge. However, existing LLM-based recommender systems struggle to effectively incorporate collaborative filtering (CF) signals, due to a fundamental mismatch between item-level preference modeling in CF and token-level next-token prediction (NTP) optimization in LLMs. Prior approaches typically treat CF as contextual hints or representation bias, and resort to multi-stage training to reduce behavioral–semantic space discrepancies, leaving CF unable to explicitly regulate LLM generation. In this work, we propose Token-level Collaborative Alignment for Recommendation (TCA4Rec), a model-agnostic and plug-and-play framework that establishes an explicit optimization-level interface between CF supervision and LLM generation. TCA4Rec consists of (i) Collaborative Tokenizer, which projects raw item-level CF logits into token-level distributions aligned with the LLM token space, and (ii) Soft Label Alignment, which integrates these CF-informed distributions with one-hot supervision to optimize a soft NTP objective. This design preserves the generative nature of LLM training while enabling collaborative alignment with essential user preference of CF models. We highlight TCA4Rec is compatible with arbitrary traditional CF models and generalizes across a wide range of decoder-based LLM recommender architectures. Moreover, it provides an explicit mechanism to balance behavioral alignment and semantic fluency, yielding generative recommendations that are both accurate and controllable. Extensive experiments demonstrate that TCA4Rec consistently improves recommendation performance across a broad spectrum of CF models and LLM-based recommender systems. Our code is available at https://github.com/critical88/TCA4Rec

Recent years have witnessed the prosperity of Knowledge Graph-Based Recommendation System (KGRS), which enriches the representation of users, items, and entities by structural knowledge with striking improvement. Nevertheless, its unaffordable computational cost still limits researchers from exploring more sophisticated models. We observe that the bottleneck for training efficiency arises from the knowledge graph, which is plagued by the well-known issue of knowledge explosion. Recently, some works have attempted to slim the inflated KG via summarization techniques, which summarize multiple real nodes into the single virtual one. However, these summarized virtual nodes may ignore collaborative signals and thus fail to figure out the redundant nodes related to recommendation task. To this end, in this article, we propose a novel approach called KGTrimmer for knowledge graph pruning tailored for recommendation, to remove the unessential nodes while minimizing performance degradation. Specifically, we design an importance evaluator from a dual-view perspective. For the collective view, we embrace the idea of collective intelligence by extracting community consensus based on abundant collaborative signals, i.e., nodes are considered important if they attract attention of numerous users. For the holistic view, we learn a global mask to identify the valueless nodes from their inherent properties or overall popularity. With the collective and holistic importance scores, we build an end-to-end importance-aware graph neural network, which injects filtered knowledge to enhance the distillation of valuable user-item collaborative signals. Ultimately, we generate a pruned knowledge graph with lightweight, stable, and robust properties to facilitate the following-up recommendation task. Extensive experiments are conducted on three publicly available datasets to prove the effectiveness and generalizability of KGTrimmer, where it can reduce the number of triplets in KG by up to 90% without compromising performance.

Recommendation systems have become ubiquitous tools in online platforms, providing personalized suggestions based on user–item interactions. To capture the dynamic higher-order connections between users and items, recommendation approaches based on dynamic graphs have garnered significant attention from researchers. However, existing recommendation methods based on dynamic graphs are often limited by data sparsity, which prevents them from achieving satisfactory performance. Fortunately, the rapid development of large language models (LLMs) with powerful text generation capabilities and extensive domain knowledge has offered new possibilities for addressing this challenge. However, how to effectively integrate LLMs with dynamic graphs remains unexplored. To bridge this gap, in this article, we propose a novel framework, that is, DynLLM, for applying LLMs to dynamic graph-based recommendation methods. Specifically, DynLLM harnesses the power of LLMs to generate multi-faceted user profiles based on the rich textual features of historical purchase records, which in turn supplement and enrich the underlying relationships between users and items. Along this line, to fuse the multi-faceted profiles with temporal graph embedding, we engage LLMs to derive corresponding profile embeddings and further employ a distilled attention mechanism to refine the LLM-generated profile embeddings for alleviating noisy signals, while also assessing and adjusting the relevance of each distilled facet embedding for seamless integration with temporal graph embedding from continuous time dynamic graphs (CTDGs). Extensive experiments on three real datasets have validated the superior improvements of DynLLM over a wide range of state-of-the-art baseline methods. The implementation code is available online at https://github.com/meteor-gif/DynLLM .

We focus on unlearning unwanted knowledge in autoregressive large language models (LLMs) through pruning. Our goal is to selectively remove undesirable information (e.g., harmful responses, privacy-sensitive data) while ensuring the preservation of desirable knowledge (e.g., positive responses and objective facts). Previous approaches use gradient ascent (GA) over undesired knowledge to inversely optimize LLMs, which compromises the model's performance on desired knowledge. To address this limitation, we introduce a novel two-stage approach, named LLM-Eraser, for selectively identifying and editing parameters specifically associated with undesirable knowledge. LLM-Eraser operates in two stages: localization and unlearning. During the localization stage, we utilize neuron scores and trainable soft masks to identify parameters crucial to the undesired knowledge. In the unlearning stage, we prune these identified parameters and apply a selective post-training process to enhance the model's selectiveness. Our experiments, conducted across five task datasets, demonstrate that LLM-Eraser effectively unlearns undesirable knowledge-evidenced by the model's near-random performance on multiple-choice questions related to the erased knowledge-while maintaining high proficiency in desirable knowledge, with an average performance deficit of only 2.5%.

Recommender systems are crucial for personalizing user experiences but often depend on implicit feedback data, which can be noisy and misleading. Existing denoising studies involve incorporating auxiliary information or learning strategies from interaction data. However, they struggle with the inherent limitations of external knowledge and interaction data, as well as the non-universality of certain predefined assumptions, hindering accurate noise identification. Recently, large language models (LLMs) have gained attention for their extensive world knowledge and reasoning abilities, yet their potential in enhancing denoising in recommendations remains underexplored. In this paper, we introduce LLaRD, a framework leveraging LLMs to improve denoising in recommender systems, thereby boosting overall recommendation performance. Specifically, LLaRD generates denoising-related knowledge by first enriching semantic insights from observational data via LLMs and inferring user-item preference knowledge. It then employs a novel Chain-of-Thought (CoT) technique over user-item interaction graphs to reveal relation knowledge for denoising. Finally, it applies the Information Bottleneck (IB) principle to align LLM-generated denoising knowledge with recommendation targets, filtering out noise and irrelevant LLM knowledge. Empirical results demonstrate LLaRD's effectiveness in enhancing denoising and recommendation accuracy.

Multimodal Entity Linking (MEL) is a crucial task that aims at linking ambiguous mentions within multimodal contexts to the referent entities in a multimodal knowledge base, such as Wikipedia. Existing methods focus heavily on using complex mechanisms and extensive model tuning methods to model the multimodal interaction on specific datasets. However, these methods overcomplicate the MEL task and overlook the visual semantic information, which makes them costly and hard to scale. Moreover, these methods cannot solve the issues like textual ambiguity, redundancy, and noisy images, which severely degrade their performance. Fortunately, the advent of Large Language Models (LLMs) with robust capabilities in text understanding and reasoning, particularly Multimodal Large Language Models (MLLMs) that can process multimodal inputs, provides new insights into addressing this challenge. However, how to design a universally applicable LLMs-based MEL approach remains a pressing challenge. To this end, we propose UniMEL, a unified framework which establishes a new paradigm to process multimodal entity linking tasks using LLMs. In this framework, we employ LLMs to augment the representation of mentions and entities individually by integrating textual and visual information and refining textual information. Subsequently, we employ the embedding-based method for retrieving and re-ranking candidate entities. Then, with only ~0.26% of the model parameters fine-tuned, LLMs can make the final selection from the candidate entities. Extensive experiments on three public benchmark datasets demonstrate that our solution achieves state-of-the-art performance, and ablation studies verify the effectiveness of all modules. Our code is available at https://github.com/Javkonline/UniMEL.

Reciprocal Recommender Systems (RRSs) which aim to satisfy the preferences of both service providers and seekers simultaneously has attracted significant research interest in recent years. Existing studies on RRSs mainly focus on modeling the bilateral interactions between the users on both sides to capture the user preferences. However, due to the presence of exposure bias, modeling user preferences solely based on bilateral interactions often lacks precision. Additionally, in RRSs, users may exhibit varying preferences when acting in different roles, and how to effectively model users from multiple perspectives remains a substantial problem. To solve the above challenges, in this paper, we propose a novel MultI-view Reciprocal Recommender system for Online Recruitment (MIRROR). Specifically, we first propose to model the users from three different views, respectively search, active, and passive views, and we further design several Transformer-based sequential models to capture the user representation corresponding to each view. Then, we propose to divide the bilateral matching process into three stages, respectively apply, reply, and match, and a multi-stage output layer is designed based on the above multi-view modeling results. To train our MIRROR model, we first design a multi-task learning loss based on the multi-stage output results. Moreover, to bridge the semantic gap between search queries and user behaviors, we additionally design a supplementary task for next-query prediction. Finally, we conduct both offline experiments on five real-world datasets and online A/B tests, and the experiment results clearly validate the effectiveness of our MIRROR model compared with several state-of-the-art baseline methods.

In the realm of deep learning-based recommendation systems, the increasing computational demands, driven by the growing number of users and items, pose a significant challenge to practical deployment. This challenge is primarily twofold: reducing the model size while effectively learning user and item representations for efficient recommendations. Despite considerable advancements in model compression and architecture search, prevalent approaches face notable constraints. These include substantial additional computational costs from pre-training/re-training in model compression and an extensive search space in architecture design. Additionally, managing complexity and adhering to memory constraints is problematic, especially in scenarios with strict time or space limitations. Addressing these issues, this paper introduces a novel learning paradigm, Dynamic Sparse Learning (DSL), tailored for recommendation models. DSL innovatively trains a lightweight sparse model from scratch, periodically evaluating and dynamically adjusting each weight's significance and the model's sparsity distribution during the training. This approach ensures a consistent and minimal parameter budget throughout the full learning lifecycle, paving the way for "end-to-end" efficiency from training to inference. Our extensive experimental results underline DSL's effectiveness, significantly reducing training and inference costs while delivering comparable recommendation performance. We give an code link of our work: https://github.com/shuyao-wang/DSL.

Recent advances in Large Language Models (LLMs) have been changing the paradigm of Recommender Systems (RS). However, when items in the recommendation scenarios contain rich textual information, such as product descriptions in online shopping or news headlines on social media, LLMs require longer texts to comprehensively depict the historical user behavior sequence. This poses significant challenges to LLM-based recommenders, such as over-length limitations, extensive time and space overheads, and suboptimal model performance. To this end, in this paper, we design a novel framework for harnessing Large Language Models for Text-Rich Sequential Recommendation (LLM-TRSR). Specifically, we first propose to segment the user historical behaviors and subsequently employ an LLM-based summarizer for summarizing these user behavior blocks. Particularly, drawing inspiration from the successful application of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) models in user modeling, we introduce two unique summarization techniques in this paper, respectively hierarchical summarization and recurrent summarization. Then, we construct a prompt text encompassing the user preference summary, recent user interactions, and candidate item information into an LLM-based recommender, which is subsequently fine-tuned using Supervised Fine-Tuning (SFT) techniques to yield our final recommendation model. We also use Low-Rank Adaptation (LoRA) for Parameter-Efficient Fine-Tuning (PEFT). We conduct experiments on two public datasets, and the results clearly demonstrate the effectiveness of our approach.

Recent years have witnessed the rapid development of online recruitment platforms, which provide a convenient way for matching job seekers and recruiters by leveraging recommendation systems. Indeed, this is areciprocal recommendationproblem which needs to consider the preferences of both job seekers and recruiters simultaneously, making it different from traditional uni-directional user-item recommendation problems. Existing studies mainly focus on building recommendation models based on the matched person-job pairs via text matching or collaborative filtering methods. However, we propose that these methods are limited and insufficient for user modeling in recruitment platforms, since the abundant multi-typed bilateral behaviors (e.g.,apply for conversationandneglect the candidates) among users have been largely ignored. Therefore, in this paper, we propose a novel BilAteral Multi-BehaviOr mOdeling (BAMBOO) method for reciprocal recommendation in online recruitment, which can model the multi-typed interactions between job seekers and recruiters from two different perspectives, respectivelyexpectation perspectiveandcompetitiveness perspective. Specifically, for the expectation perspective, we propose to format the historical behaviors of different users as bilateral multi-behavior sequences, and we utilize a transformer-based model to learn the representations of what the userswantto obtain. For the competitiveness perspective, we propose to construct a bilateral interaction heterogeneous graph to describe the entire recruitment market, and further utilize a heterogeneous graph transformer-based model to learn the representations of what the userscanobtain. Moreover, we utilize contrastive learning methods to enhance these two modules. Furthermore, we propose to decompose the matching probability between job seekers and recruiters into the product of two parts, respectively the probability of the active party initiating the conversation and the probability of the passive party accepting it, and we train our model based on a multi-task learning strategy. Finally, we conduct both offline experiments on real-world datasets and online A/B test, and the experiment results validate the effectiveness of our BAMBOO model compared with several state-of-the-art baseline methods.

With the rapid development of online recruitment platforms, a variety of emerging recommendation services have been witnessed for benefiting both job seekers and recruiters. While many researchers have studied the problem of reciprocal recommendation in two- sided markets (e.g., marriage market and real estate market), there is still a lack of in-depth understanding of the bilateral occupational preferences of different participants in the online recruitment market. To this end, in this paper, we propose a Bilateral Occupational-Suitability-aware recommender System (BOSS) for online recruitment, in consideration of the reciprocal, bilateral, and sequential properties of realistic recruitment scenarios simultaneously. To be specific, in BOSS, we first propose a multi-group-based mixture-of-experts (MoE) module to independently learn the preference representations of job seekers and recruiters. Then, with a specially-designed multi-task learning module, BOSS can progressively model the action sequence of recruitment process through a bilateral probabilistic manner. As a result, the reciprocal recommendations can be efficiently implemented by leveraging the product of different action probabilities of job seekers and recruiters. Finally, we have conducted extensive experiments on 5 real-world large-scale datasets as well as the online environment. Both online A/B test and offline experimental results clearly validate that our recommender system BOSS can outperform other state-of-the-art baselines with a significant margin.

With the rapid growth of online medical platforms, more and more doctors are willing to manage and communicate with patients via online services. Considering the large volume and various patient conditions, identifying and classifying patients' medical records has become a crucial problem. To efficiently index these records, a common practice is to annotate them with semantically meaningful tags. However, manual labeling tags by doctors is impractical due to the possibility of thousands of tag candidates, which necessitates a tag recommender system. Due to the long tail distribution of tags and the dominance of low-activity doctors, as well as the unique uploaded medical records, this task is rather challenging. This paper proposes an efficient doctor specific tag recommendation framework for improved medical record management without side information. Specifically, we first utilize effective language models to learn the text representation. Then, we construct a doctor embedding learning module to enhance the recommendation quality by integrating implicit information within text representations and considering latent tag correlations to make more accurate predictions. Extensive experiment results demonstrate the effectiveness of our framework from the viewpoints of all doctors (20% improvement) or low-activity doctors (10% improvement).

With the rapid development of enterprise Learning Management Systems (LMS), more and more companies are trying to build enterprise training and course learning platforms for promoting the career development of employees. Indeed, through course learning, many employees have the opportunity to improve their knowledge and skills. For these systems, a major issue is how to recommend learning plans, i.e., a set of courses arranged in the order they should be learned, that can help employees improve their work performance. Existing studies mainly focus on recommending courses that users are most likely to click on by capturing their learning preferences. However, the learning preference of employees may not be the right fit for their career development, and thus it may not necessarily mean their work performance can be improved accordingly. Furthermore, how to capture the mutual correlation and sequential effects between courses, and ensure the rationality of the generated results, is also a major challenge. To this end, in this paper, we propose the Generative Learning plAn recommenDation (GLAD) framework, which can generate personalized learning plans for employees to help them improve their work performance. Specifically, we first design a performance predictor and a rationality discriminator, which have the same transformer-based model architecture, but with totally different parameters and functionalities. In particular, the performance predictor is trained for predicting the work performance of employees based on their work profiles and historical learning records, while the rationality discriminator aims to evaluate the rationality of the generated results. Then, we design a learning plan generator based on the gated transformer and the cross-attention mechanism for learning plan generation. We calculate the weighted sum of the output from the performance predictor and the rationality discriminator as the reward, and we use Self-Critical Sequence Training (SCST) based policy gradient methods to train the generator following the Generative Adversarial Network (GAN) paradigm. Finally, extensive experiments on real-world data clearly validate the effectiveness of our GLAD framework compared with state-of-the-art baseline methods and reveal some interesting findings for talent management.

Recent years have witnessed the rapid accumulation of massive electronic medical records, which highly support intelligent medical services such as drug recommendation. However, although there are multiple interaction types between drugs, e.g., synergism and antagonism, which can influence the effect of a drug package significantly, prior arts generally neglect the interaction between drugs or consider only a single type of interaction. Moreover, most existing studies generally formulate the problem of package recommendation as getting a personalized scoring function for users, despite the limits of discriminative models to achieve satisfactory performance in practical applications. To this end, in this article, we propose a novel end-to-end Drug Package Generation (DPG) framework, which develops a new generative model for drug package recommendation that considers the interaction effects between drugs that are affected by patient conditions. Specifically, we propose to formulate the drug package generation as a sequence generation process. Along this line, we first initialize the drug interaction graph based on medical records and domain knowledge. Then, we design a novel message-passing neural network to capture the drug interaction, as well as a drug package generator based on a recurrent neural network. In detail, a mask layer is utilized to capture the impact of patient condition, and the deep reinforcement learning technique is leveraged to reduce the dependence on the drug order. Finally, extensive experiments on a real-world dataset from a first-rate hospital demonstrate the effectiveness of our DPG framework compared with several competitive baseline methods.

Online medical consultation, which enables patients to remotely inquire doctors in the form of web chatting, has become an indispensable part of the social health care system. Intuitively, it is a crucial step to recommend suitable doctor candidates for patients, especially with suffering the severe cold-start challenge of patients due to the limited historical records and insufficient description of patient condition. Along this line, in this paper, we propose a novel Dialogue based Doctor Recommendation (DDR) model, which comprehensively integrates three types of information in modeling, including the profile and chief complaint from patients, the historical records of doctors and the patient-doctor dialogue. Accordingly, we propose 1) a patient encoder which represents the patient's condition and medical requirements; 2) a doctor encoder which distills the doctor's expertise and communication skills; 3) a dialogue encoder which extracts textual features from doctor-patient conversation. Specifically, since the patient-doctor dialogue is not available in the testing stage, we propose to simulate the dialogue embedding with patient embedding via a contrastive learning based module. Experimental results on a real-world data set show that the proposed DDR model can outperform state-of-the-art recommendation-based methods. Moreover, considering the accessibility variance of online medical consultation services between the youth and the elderly, we also conduct a fairness study on the proposed DDR model.

With the prosperity of recommender systems, the biases existing in user behaviors, which may lead to inconsistency between user preference and behavior records, have attracted wide attention. Though large efforts have been made to infer user preference from biased data with learning to debias, unfortunately, they mainly focus on the effect of one specific item attribute, e.g., position or modality which may affect users’ click probability on items. However, the comprehensive description for potential interactions between multiple items with various attributes, namely the context bias between items, may not be fully summarized. To that end, in this paper, we design a novel Context Bias aware Recommendation (CBR) model for describing and debiasing the context bias caused by comprehensive interactions between multiple items. Specifically, we first propose a content encoder and a bias encoder based on multi-head self-attention to embed the latent interactions between items. Then, we calculate the biased representation for users based on an attention network, which will be further utilized to infer the negative preference, i.e., the dislikes of users based on the items the user never clicked. Finally, the real user preference will be captured based on the negative preference to estimate the click prediction score. Extensive experiments on a real-world dataset demonstrate the competitiveness of our CBR framework compared with state-of-the-art baseline methods.

Recent years have witnessed the rapid accumulation of massive electronic medical records (EMRs), which highly support the intelligent medical services such as drug recommendation. However, prior arts mainly follow the traditional recommendation strategies like collaborative filtering, which usually treat individual drugs as mutually independent, while the latent interactions among drugs, e.g., synergistic or antagonistic effect, have been largely ignored. To that end, in this paper, we target at developing a new paradigm for drug package recommendation with considering the interaction effect within drugs, in which the interaction effects could be affected by patient conditions. Specifically, we first design a pre-training method based on neural collaborative filtering to get the initial embedding of patients and drugs. Then, the drug interaction graph will be initialized based on medical records and domain knowledge. Along this line, we propose a new Drug Package Recommendation (DPR) framework with two variants, respectively DPR on Weighted Graph (DPR-WG) and DPR on Attributed Graph (DPR-AG) to solve the problem, in which each the interactions will be described as signed weights or attribute vectors. In detail, a mask layer is utilized to capture the impact of patient condition, and graph neural networks (GNNs) are leveraged for the final graph induction task to embed the package. Extensive experiments on a real-world data set from a first-rate hospital demonstrate the effectiveness of our DPR framework compared with several competitive baseline methods, and further support the heuristic study for the drug package generation task with adequate performance.