Full-parameter fine-tuning of large language models is constrained by substantial GPU memory demands. Low-rank adaptation methods mitigate this challenge by updating only a subset of parameters. However, these approaches often limit model expressiveness and yield lower performance than full-parameter fine-tuning. Layer-wise fine-tuning methods have emerged as an alternative, enabling memory-efficient training through static layer importance sampling strategies. However, these methods overlook variations in layer importance across tasks and training stages, resulting in suboptimal performance on downstream tasks. To address these limitations, we propose GRASS, a gradient-based adaptive layer-wise importance sampling framework. GRASS utilizes mean gradient norms as a task-aware and training-stage-aware metric for estimating layer importance. Furthermore, GRASS adaptively adjusts layer sampling probabilities through an adaptive training strategy. We also introduce a layer-wise optimizer state offloading mechanism to further reduce memory usage while maintaining comparable training throughput. Extensive experiments across multiple models and benchmarks demonstrate that GRASS consistently outperforms state-of-the-art methods, achieving an average accuracy improvement of up to 4.38 points and reducing memory usage by up to 19.97%.

Emotional support conversation (ESC) aims to alleviate users’ psychological stress. Selecting the appropriate strategy is crucial for effective emotional support. Current strategy planner-based methods prioritize immediate responses while neglecting users’ future reactions. Some studies retrieve historical examples with similar emotions to the current utterance, then anticipating future emotions based on next-turn emotions of historical examples. However, their retrievals focus on the current emotion (i.e. a single-turn emotion state), while they ignore the evolution of user’s emotion before the current state. We argue that retrievals considering the whole emotional trajectories enables models to capture the dynamic emotional needs, thereby enhancing the anticipation of future emotions. To this end, we propose Markov-driven emotion anticipation framework with emotion trajectory-aware retrieval for LLM-based ESC, which anticipates future emotion states to guide strategy planning and achieve sustained emotional support. First, we construct a dynamic emotion memory and perform hierarchical retrieval that combines semantic matching and emotion trajectory alignment. Then, we model emotional transitions as Markov chains, leveraging trajectory-aware retrieval to estimate future emotion. Finally, we use the anticipated emotion to steer LLMs in generating candidate strategies and introduce active online learning to optimize the planner, boosting its robustness on diverse users. Experiments on two datasets with two models shows that our method excels all baselines.

Mixture-of-Experts (MoE) has become a dominant architecture for scaling large language models due to their sparse activation mechanism. However, the substantial number of expert activations creates a critical latency bottleneck during inference, especially in resource-constrained deployment scenarios. Existing approaches that reduce expert activations potentially lead to severe model performance degradation. In this work, we introduce the concept of activation budget as a constraint on the number of expert activations and propose Alloc-MoE, a unified framework that optimizes budget allocation coordinately at both the layer and token levels to minimize performance degradation. At the layer level, we introduce Alloc-L, which leverages sensitivity profiling and dynamic programming to determine the optimal allocation of expert activations across layers. At the token level, we propose Alloc-T, which dynamically redistributes activations based on routing scores, optimizing budget allocation without increasing latency. Extensive experiments across multiple MoE models demonstrate that Alloc-MoE maintains model performance under a constrained activation budget. Especially, Alloc-MoE achieves 1.15× prefill and 1.34× decode speedups on DeepSeek-V2-Lite at half of the original budget.

Rumor detection on social media has become an emerging topic. Traditional deep learning-based methods model rumors based on content, propagation structure, or user behavior, but these approaches are constrained by limited modeling capacity and insufficient training corpora. Recent studies have explored using LLMs for rumor detection through supervised fine-tuning (SFT), but face two issues: 1) unreliable samples sometimes mislead the model learning; 2) the model only learns the most salient input-output mapping and skips in-depth analyses of the rumored content for convenience. To address these issues, we propose an SFT-based LLM rumor detection model with Influence guided Sample selection and Game-based multi-perspective Analysis (ISGA). Specifically, we first introduce the Influence Score (IS) to assess the impact of samples on model predictions and select samples for SFT. We also approximate IS via Taylor expansion to reduce computational complexity. Next, we use LLMs to generate in-depth analyses of news content from multiple perspectives and model their collaborative process for prediction as a cooperative game. Then we utilize the Shapley value to quantify the contribution of each perspective for selecting informative perspective analyses. Experiments show that ISGA excels existing SOTA on three datasets.

Event Extraction (EE) is a challenging task that aims to extract structural event-related information from unstructured text. Traditional methods for EE depend on manual annotations, which are both expensive and scarce. Furthermore, the existing datasets mostly follow the long-tail distribution, severely hindering the previous methods of modeling tail types. Two techniques can address this issue: transfer learning and data generation. However, the existing methods based on transfer learning still rely on pre-training with a large amount of labeled data in the source domain. Additionally, the quality of data generated by previous data generation methods is difficult to control. In this paper, leveraging Large Language Models (LLMs), we propose novel methods for event extraction and generation based on dialogues, overcoming the problems of relying on source domain data and maintaining data quality. Specifically, this paper innovatively transforms the EE task into multi-turn dialogues, guiding LLMs to learn event schemas from historical dialogue information and output structural events. Furthermore, we introduce a novel LLM-based method for generating high-quality data, significantly improving traditional models’ performance with various paradigms and structures, especially on tail types. Adequate experiments on real-world datasets demonstrate the effectiveness of the proposed event extraction and data generation methods.

Temporal knowledge graph question answering (TKGQA) poses a significant challenge task, due to the temporal constraints hidden in questions and the answers sought from dynamic structured knowledge. Although large language models (LLMs) have made considerable progress in their reasoning ability over structured data, their application to the TKGQA task is a relatively unexplored area. This paper first proposes a novel generative temporal knowledge graph question answering framework, GenTKGQA, which guides LLMs to answer temporal questions through two phases: Subgraph Retrieval and Answer Generation. First, we exploit LLM’s intrinsic knowledge to mine temporal constraints and structural links in the questions without extra training, thus narrowing down the subgraph search space in both temporal and structural dimensions. Next, we design virtual knowledge indicators to fuse the graph neural network signals of the subgraph and the text representations of the LLM in a non-shallow way, which helps the open-source LLM deeply understand the temporal order and structural dependencies among the retrieved facts through instruction tuning. Experimental results on two widely used datasets demonstrate the superiority of our model.

Temporal knowledge graph completion that predicts missing links for incomplete temporal knowledge graphs (TKG) is gaining increasing attention. Most existing works have achieved good results by incorporating time information into static knowledge graph embedding methods. However, they ignore the contextual nature of the TKG structure, i.e., query-specific subgraph contains both structural and temporal neighboring facts. This paper presents the SToKE, a novel method that employs the pre-trained language model (PLM) to learn joint Structural and Temporal Contextualized Knowledge Embeddings.Specifically, we first construct an event evolution tree (EET) for each query to enable PLMs to handle the TKG, which can be seen as a structured event sequence recording query-relevant structural and temporal contexts. We then propose a novel temporal embedding and structural matrix to learn the time information and structural dependencies of facts in EET.Finally, we formulate TKG completion as a mask prediction problem by masking the missing entity of the query to fine-tune pre-trained language models. Experimental results on three widely used datasets show the superiority of our model.

Traditional approaches to the task of ACE event extraction usually depend on manually annotated data, which is often laborious to create and limited in size. Therefore, in addition to the difficulty of event extraction itself, insufficient training data hinders the learning process as well. To promote event extraction, we first propose an event extraction model to overcome the roles overlap problem by separating the argument prediction in terms of roles. Moreover, to address the problem of insufficient training data, we propose a method to automatically generate labeled data by editing prototypes and screen out generated samples by ranking the quality. Experiments on the ACE2005 dataset demonstrate that our extraction model can surpass most existing extraction methods. Besides, incorporating our generation method exhibits further significant improvement. It obtains new state-of-the-art results on the event extraction task, including pushing the F1 score of trigger classification to 81.1%, and the F1 score of argument classification to 58.9%.