Safety-aligned LLMs suffer from two failure modes: jailbreak (answering harmful inputs) and over-refusal (declining benign queries).Existing vector steering methods adjust the magnitude of answer vectors, but this creates a fundamental trade-off-reducing jailbreak increases over-refusal and vice versa.We identify the root cause: LLMs encode the decision to answer (answer vector v a ) and the judgment of input safety (benign vector v b ) as nearly orthogonal directions, treating them as independent processes.We propose LLM-VA, which aligns v a with v b through closedform weight updates, making the model's willingness to answer causally dependent on its safety assessment-without fine-tuning or architectural changes.Our method identifies vectors at each layer using SVMs, selects safetyrelevant layers, and iteratively aligns vectors via minimum-norm weight modifications.Experiments on 12 LLMs demonstrate that LLM-VA achieves 11.45% higher F1 than the best baseline while preserving 95.92% utility, and automatically adapts to each model's safety bias without manual tuning.

Despite the widespread use of Transformerbased text embedding models in NLP tasks, surprising "sticky tokens" can undermine the reliability of embeddings.These tokens, when repeatedly inserted into sentences, pull sentence similarity toward a certain value, disrupting the normal distribution of embedding similarities and degrading downstream performance.In this paper, we systematically investigate such anomalous tokens, formally defining them and introducing an efficient detection method, Sticky Token Detector (STD), based on sentence and token filtering.Applying STD to 40 checkpoints across 14 model families, we discover a total of 868 sticky tokens.Our analysis reveals that these tokens often originate from special or unused entries in the vocabulary, as well as fragmented subwords from multilingual corpora.Notably, their presence does not strictly correlate with model size or vocabulary size.We further evaluate how sticky tokens affect downstream tasks like clustering and retrieval, observing substantial performance degradation that approaches 50% in certain cases.Through attention-layer analysis, we show that sticky tokens disproportionately dominate the model's internal representations, raising concerns about tokenization robustness.Our findings show the need for better tokenization strategies and model design to mitigate the impact of sticky tokens in future text embedding applications.�

Large Language Models (LLMs) have been found to show over-refusal problems—erroneously rejecting benign queries due to overly conservative safety measures—a critical functional flaw that undermines their reliability and usability. Current methods for testing this behavior are demonstrably inadequate, suffering from flawed benchmarks and limited test generation capabilities, as highlighted by our empirical user study. To the best of our knowledge, this paper introduces the first evolutionary testing framework, ORFuzz, for the systematic detection and analysis of LLM over-refusals. ORFuzz uniquely integrates three core components: (1) safety category-aware seed selection for comprehensive test coverage, (2) adaptive mutator optimization using reasoning LLMs to generate effective test cases, and (3) OR-Judge, a human-aligned judge model validated to accurately reflect user perception of toxicity and refusal. Our extensive evaluations demonstrate that ORFuzz generates diverse, validated over-refusal instances at a rate (6.98% average) more than double that of leading baselines, effectively uncovering vulnerabilities. Furthermore, ORFuzz’s outputs form the basis of ORFuzzSet, a new benchmark of 1,786 highly transferable test cases that achieves a superior 57.37% average over-refusal rate across 14 diverse LLMs, significantly outperforming existing datasets. ORFuzz and ORFuzzSet provide a robust automated testing framework and a valuable community resource, paving the way for developing more reliable and trustworthy LLM-based software systems. The code of this paper is available at: https://github.com/HotBento/ORFuzz.

Recently, Graph Neural Networks (GNNs) have become the dominant approach for Knowledge Graph-aware Recommender Systems (KGRSs) due to their proven effectiveness. Building upon GNN-based KGRSs, Self-Supervised Learning (SSL) has been incorporated to address the sparity issue, leading to longer training time. However, through extensive experiments, we reveal that: (1)compared to other KGRSs, the existing GNN-based KGRSs fail to keep their superior performance under sparse interactions even with SSL. (2) More complex models tend to perform worse in sparse interaction scenarios and complex mechanisms, like attention mechanism, can be detrimental as they often increase learning difficulty. Inspired by these findings, we propose LightKG, a simple yet powerful GNN-based KGRS to address sparsity issues. LightKG includes a simplified GNN layer that encodes directed relations as scalar pairs rather than dense embeddings and employs a linear aggregation framework, greatly reducing the complexity of GNNs. Additionally, LightKG incorporates an efficient contrastive layer to implement SSL. It directly minimizes the node similarity in original graph, avoiding the time-consuming subgraph generation and comparison required in previous SSL methods. Experiments on four benchmark datasets show that LightKG outperforms 12 competitive KGRSs in both sparse and dense scenarios while significantly reducing training time. Specifically, it surpasses the best baselines by an average of 5.8% in recommendation accuracy and saves 84.3% of training time compared to KGRSs with SSL. Our code is available at https://github.com/1371149/LightKG.

Recommender systems (RSs) are designed to provide personalized recommendations to users. Recently, knowledge graphs (KGs) have been widely introduced in RSs to improve recommendation accuracy. In this study, however, we demonstrate that RSs do not necessarily perform worse even if the KG is downgraded to the user-item interaction graph only (or removed). We propose an evaluation framework KG4RecEval to systematically evaluate how much a KG contributes to the recommendation accuracy of a KG-based RS, using our defined metric KG utilization efficiency in recommendation (KGER). We consider the scenarios where knowledge in a KG gets completely removed, randomly distorted and decreased, and also where recommendations are for cold-start users. Our extensive experiments on four commonly used datasets and a number of state-of-the-art KG-based RSs reveal that: to remove, randomly distort or decrease knowledge does not necessarily decrease recommendation accuracy, even for cold-start users. These findings inspire us to rethink how to better utilize knowledge from existing KGs, whereby we discuss and provide insights into what characteristics of datasets and KG-based RSs may help improve KG utilization efficiency. The code and supplementary material of this article are available at: https://github.com/HotBento/KG4RecEval .

Recommender systems tend to excessively expose longer-standing items, resulting in significant unfairness to new items with little interaction records, despite they may possess potential to attract considerable amount of users. The existing fairness-based solutions do not specifically consider the exposure fairness of new items, for which a systematic definition also lacks, discouraging the promotion of new items or contents. In this work, we introduce a multi-degree new-item exposure fairness definition, which considers item entry-time, and also is configurable regarding different fairness requirements. We then propose a configurable new-item fairness-aware framework named CNIF, which employs two-stage training where fairness degrees are incorporated for guidance. Extensive experiments on multiple popular datasets and backbone models demonstrate that CNIF can effectively enhance fairness of the existing models regarding the exposure resources of new items (including the brand-new items with no interaction). Specifically, CNIF demonstrates a substantial advancement with a 65.59% improvement in fairness metric and a noteworthy 9.97% improvement in recommendation accuracy compared to backbone models on the KuaiRec dataset. In comparison to various fairness-based solutions, it stands out by achieving the best trade-off between fairness and recommendation accuracy, surpassing the best baseline by 14.20%.