Large Language Models (LLMs) have achieved rapid progress and are increasingly deployed in real-world applications such as digital assistants, education, healthcare, and recommendation. This deployment has driven growing interest in LLM personalization, which seeks to align model behavior with individual preferences and evolving contexts. Despite the rapid development of this area, existing research remains scattered, and a systematic tutorial dedicated to LLM personalization is still lacking. This tutorial presents a unified technical framework for LLM personalization, organized around five core dimensions: user memory, personalization architecture, alignment and post-training, inference-time adaptation, and deployment. We show how these components transform LLMs from generic response generators into user-adaptive systems, enabling structured user representation, memory integration, personalized optimization objectives, and context-aware reasoning throughout the model lifecycle. We further discuss key challenges, including lifelong learning, preference drift, privacy-preserving adaptation, trustworthiness, and evaluation under dynamic user distributions. By consolidating recent advances, this tutorial aims to equip participants with a comprehensive and principled understanding of LLM personalization and to inspire continued innovation in this rapidly evolving field.

Large language models (LLMs) are increasingly integrated into users' daily lives, leading to a growing demand for personalized outputs.Previous work focuses on leveraging a user's own history, overlooking inter-user differences that are crucial for effective personalization.While recent work has attempted to model such differences, the reliance on language-based prompts often hampers the effective extraction of meaningful distinctions.To address these issues, we propose Difference-aware Embeddingbased Personalization (DEP), a framework that models inter-user differences in the latent space instead of relying on language prompts.DEP constructs soft prompts by contrasting a user's embedding with those of peers who engaged with similar content, highlighting relative behavioral signals.A sparse autoencoder then filters and compresses both user-specific and difference-aware embeddings, preserving only task-relevant features before injecting them into a frozen LLM.Experiments on personalized review generation show that DEP consistently outperforms baseline methods across multiple metrics.

The in-context learning (ICL) capability of large language models (LLMs) enables them to perform challenging tasks using provided demonstrations. However, ICL is highly sensitive to the ordering of demonstrations, leading to instability in predictions. This paper shows that this vulnerability can be exploited to design a natural attack—difficult for model providers to detect—that achieves nearly 80% success rate on LLaMA-3 by simply permuting the demonstrations. Existing mitigation methods primarily rely on post-processing and fail to enhance the model's inherent robustness to input permutations, raising concerns about safety and reliability of LLMs. To address this issue, we propose Permutation-resilient learning (PEARL), a novel framework based on distributionally robust optimization (DRO), which optimizes model performance against the worst-case input permutation. Specifically, PEARL consists of a permutation-proposal network (P-Net) and the LLM. The P-Net generates the most challenging permutations by treating it as an optimal transport problem, which is solved using an entropy-constrained Sinkhorn algorithm. Through minimax optimization, the P-Net and the LLM iteratively optimize against each other, progressively improving the LLM's robustness. Experiments on synthetic pre-training and real-world instruction tuning tasks demonstrate that PEARL effectively mitigates permutation attacks and enhances performance. Notably, despite being trained on fewer shots and shorter contexts, PEARL achieves performance gains of up to 40% when scaled to many-shot and long-context scenarios, highlighting its efficiency and generalization capabilities.

Large Language Models (LLMs) have shown strong potential for recommendation by framing item prediction as a token-by-token language generation task. However, existing methods treat all item tokens equally, simply pursuing likelihood maximization during both optimization and decoding. This overlooks crucial token-level differences in decisiveness—many tokens contribute little to item discrimination yet can dominate optimization or decoding. To quantify token decisiveness, we propose a novel perspective that models item generation as a decision process, measuring token decisiveness by the Information Gain (IG) each token provides in reducing uncertainty about the generated item. Our empirical analysis reveals that most tokens have low IG but often correspond to high logits, disproportionately influencing training loss and decoding, which may impair model performance. Building on these insights, we introduce an Information Gain-based Decisiveness-aware Token handling (IGD) strategy that integrates token decisiveness into both tuning and decoding. Specifically, IGD downweights low-IG tokens during tuning and rebalances decoding to emphasize tokens with high IG. In this way, IGD moves beyond pure likelihood maximization, effectively prioritizing high-decisiveness tokens. Extensive experiments on four benchmark datasets with two LLM backbones demonstrate that IGD consistently improves recommendation accuracy, achieving significant gains on widely used ranking metrics compared to strong baselines. Our codes are available at \url{https://github.com/ZJLin2oo1/IGD}.

Relation extraction (RE) tends to struggle when the supervised training data is few and difficult to be collected. In this article, we elicit relational and factual knowledge from large pretrained language models (PLMs) for few-shot RE (FSRE) with prompting techniques. Concretely, we automatically generate a diverse set of natural language templates and modulate PLM's behavior through these prompts for FSRE. To mitigate the template bias which leads to unstableness of few-shot learning, we propose a simple yet effective template regularization network (TRN) to prevent deep networks from over-fitting uncertain templates and thus stabilize the FSRE models. TRN alleviates the template bias with three mechanisms: 1) an attention mechanism over mini-batch to weight each template; 2) a ranking regularization mechanism to regularize the attention weights and constrain the importance of uncertain templates; and 3) a template calibration module with two calibrating techniques to modify the uncertain templates in the lowest-ranked group. Experimental results on two benchmark datasets (i.e., FewRel and NYT) show that our model has robust superiority over strong competitors. For reproducibility, we will release our code and data upon the publication of this article.

Yang Sun, Muyi Wang, Jianzhu Bao, Bin Liang, Xiaoyan Zhao, Caihua Yang, Min Yang, Ruifeng Xu. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

In an era characterized by the rapid proliferation of information, the pervasive issues of misinformation and disinformation have significantly impacted numerous individuals. Consequently, the evaluation of information’s truthfulness and accuracy has garnered substantial attention among researchers. In this work, we present a novel fact-checking framework called PACAR, fact-checking based on planning and customized action reasoning using LLMs. It comprises four modules: a claim decomposer with self-reflection, an LLM-centric planner module, an executor for carrying out planned actions, and a verifier module that assesses veracity and generates explanations based on the overall reasoning process. Unlike previous work that employs single-path decision-making and single-step verdict prediction, PACAR focuses on the use of LLMs in dynamic planning and execution of actions. Furthermore, in contrast to previous work that relied primarily on general reasoning, we introduce tailored actions such as numerical reasoning and entity disambiguation to effectively address potential challenges in fact-checking. Our PACAR framework, incorporating LLM-centric planning along with customized action reasoning, significantly outperforms baseline methods across three datasets from different domains and with varying complexity levels. Additional experiments, including multidimensional and sliced observations, demonstrate the effectiveness of PACAR and offer valuable insights for the advancement of automated fact-checking.

The emergence of large language models (LLMs) has revolutionized the capabilities of text comprehension and generation. Multi-modal generation attracts great attention from both the industry and academia, but there is little work on personalized generation, which has important applications such as recommender systems. This paper proposes the first method for personalized multimodal generation using LLMs, showcases its applications and validates its performance via an extensive experimental study on two datasets. The proposed method, Personalized Multimodal Generation (PMG for short) first converts user behaviors (e.g., clicks in recommender systems or conversations with a virtual assistant) into natural language to facilitate LLM understanding and extract user preference descriptions. Such user preferences are then fed into a generator, such as a multimodal LLM or diffusion model, to produce personalized content. To capture user preferences comprehensively and accurately, we propose to let the LLM output a combination of explicit keywords and implicit embeddings to represent user preferences. Then the combination of keywords and embeddings are used as prompts to condition the generator. We optimize a weighted sum of the accuracy and preference scores so that the generated content has a good balance between them. Compared to a baseline method without personalization, PMG has a significant improvement on personalization for up to 8% in terms of LPIPS while retaining the accuracy of generation.

The task of entity alignment between knowledge graphs (KGs) aims to identify every pair of entities from two different KGs that represent the same entity. Many machine learning-based methods have been proposed for this task. However, to our best knowledge, existing methods all requiremanually craftedseed alignments, which are expensive to obtain. In this paper, we propose the first fully automatic alignment method named AutoAlign, which does not require any manually crafted seed alignments. Specifically, for predicate embeddings, AutoAlign constructs a predicate-proximity-graph with the help of large language models to automatically capture the similarity between predicates across two KGs. For entity embeddings, AutoAlign first computes the entity embeddings of each KG independently using TransE, and then shifts the two KGs' entity embeddings into the same vector space by computing the similarity between entities based on their attributes. Thus, both predicate alignment and entity alignment can be done without manually crafted seed alignments. AutoAlign is not only fully automatic, but also highly effective. Experiments using real-world KGs show that AutoAlign improves the performance of entity alignment significantly compared to state-of-the-art methods. Our source code is available at ruizhang-ai/AutoAlign.

The goal of semi-supervised text classification (SSTC) is to train a model by exploring both a small number of labeled data and a large number of unlabeled data, such that the learned semi-supervised classifier performs better than the supervised classifier trained on solely the labeled samples. Pseudo-labeling is one of the most widely used SSTC techniques, which trains a teacher classifier with a small number of labeled examples to predict pseudo labels for the unlabeled data. The generated pseudo-labeled examples are then utilized to train a student classifier, such that the learned student classifier can outperform the teacher classifier. Nevertheless, the predicted pseudo labels may be inaccurate, making the performance of the student classifier degraded. The student classifier may perform even worse than the teacher classifier. To alleviate this issue, in this paper, we introduce a dual meta-learning ( DML ) technique for semi-supervised text classification, which improves the teacher and student classifiers simultaneously in an iterative manner. Specifically, we propose a meta-noise correction method to improve the student classifier by proposing a Noise Transition Matrix (NTM) with meta-learning to rectify the noisy pseudo labels. In addition, we devise a meta pseudo supervision method to improve the teacher classifier. Concretely, we exploit the feedback performance from the student classifier to further guide the teacher classifier to produce more accurate pseudo labels for the unlabeled data. In this way, both teacher and student classifiers can co-evolve in the iterative training process. Extensive experiments on four benchmark datasets highlight the effectiveness of our DML method against existing state-of-the-art methods for semi-supervised text classification. We release our code and data of this paper publicly at https://github.com/GRIT621/DML.

Significant progress has been made by joint entity and relation extraction methods, which directly generate the relation triplets and mitigate the issue of overlapping relations. However, previous models generate the entity-relation triplets solely from input sentences. Such information is insufficient to support the modeling of interactive information between entities and relations. In this paper, we define the features that provide mutual supports for entity and relation detection but can only be accessed at training time as privileged features for relation extraction, and devise two teacher models to exploit privileged entity and relation features, respectively. Meanwhile, we propose a novel contrastive student-teacher learning framework for joint extraction of entities and relations (STER), where a student network is encouraged to amalgamate privileged knowledge from two expert teacher networks that additionally utilize the privileged features, based on contrastive learning. Experiment results on three benchmark datasets (i.e., ADE, SciERC and CoNLL04) demonstrate that STER has robust superiority over competitors and sets state-of-the-art. For reproducibility, we will release the data and source code once the paper is accepted.

Few-shot relation classification aims to classify the relation type between two given entities in a sentence by training with a few labeled instances for each relation. However, most of existing models fail to distinguish multiple relations that co-exist in one sentence. This paper presents a novel dependency-aware prototype learning (DAPL) method for few-shot relation classification. Concretely, we utilize dependency trees and shortest dependency paths (SDP) as structural information to complement the contextualized representations of input sentences by using the dependency-aware embedding as attention inputs to learn attentive sentence representations. In addition, we introduce a gate controlled update mechanism to update the dependency-aware representations according to the output of each network layer. Extensive experiments on the FewRel dataset show that DAPL achieves substantially better performance than strong baselines. For reproducibility, we will release our code and data upon the publication of this paper at https://github.com/publicstaticvo/DAPL.

Deep neural networks (DNN) based recommender models often require numerous parameters to achieve remarkable performance. However, this inevitably brings redundant neurons, a phenomenon referred to as over-parameterization. In this paper, we plan to exploit such redundancy phenomena for recommender systems (RS), and propose a top-N item recommendation framework called PCRec that leverages collaborative training of two recommender models of the same network structure, termed peer collaboration. We first introduce two criteria to identify the importance of parameters of a given recommender model. Then, we rejuvenate the unimportant parameters by copying parameters from its peer network. After such an operation and retraining, the original recommender model is endowed with more representation capacity by possessing more functional model parameters. To show its generality, we instantiate PCRec by using three well-known recommender models. We conduct extensive experiments on two real-world datasets, and show that PCRec yields significantly better performance than its counterpart with the same model (parameter) size.

Chinese word segmentation (CWS) is an important research topic in information retrieval (IR) and natural language processing (NLP). Significant progresses have been made by deep neural networks with context features. However, these deep models may fail to deal with rare or ambiguous words, thus limit the overall CWS performance. In this paper, we propose a lexicon-enhanced adaptive attention network (LAAN), which takes full advantage of external lexicons to deal with the rare or ambiguous words. Specifically, we devise an adaptive attention mechanism to learn the lexicon-aware representation. In addition, we propose a fusion gate to effectively integrate the additional word information with context information to improve the performance of CWS. LAAN is evaluated on four benchmark datasets, and the experimental results demonstrate that LAAN has robust superiority over the compared methods.