跳转至

C-Unseen：利用大模型推理检测动态时序知识图谱中的弱信号

文章背景与核心概要

弱信号是指在重大变革成为主流之前，预示这些变革发生的早期、低可见度指标。传统的检测方法主要依赖关键词频率、主题建模或无类型图拓扑结构，往往无法捕捉信号展现时所依托的深层语义与关系结构。为了填补这一空白，本文提出了 C-Unseen 框架。

C-Unseen 是一个专为动态时序知识图谱（DTKGs）中的弱信号检测而设计的创新、可自我解释的框架。它包含两个核心组件：一是“稀疏子图提取器”（Rare Subgraphs Extractor），利用大语言模型（LLMs）通过思维链推理识别出内容与快照主流叙事相悖的子图；二是“弱信号警报器”（Weak Signal Alerter），用于追踪这些稀疏子图在连续时间步中的持久性，从而隔离并验证真正的弱信号。实验结果表明，C-Unseen 在性能上显著优于传统的基于关键词、主题和图的基线方法。

C-Unseen is an innovative, self-interpretable framework designed for weak signal detection within Dynamic Temporal Knowledge Graphs (DTKGs). Weak signals are subtle, low-visibility indicators that foreshadow significant upcoming changes before they become mainstream. Traditional methods—relying on keyword frequencies, topic modeling, or untyped graph topologies—often fail to capture the underlying semantic and relational structures.

The C-Unseen framework addresses this gap through two core components: 1. Rare Subgraphs Extractor: Utilizes Large Language Models (LLMs) and chain-of-thought reasoning to identify subgraphs whose contents run counter to the dominant narrative of a snapshot. 2. Weak Signal Alerter: Tracks the persistence of these rare subgraphs across consecutive time steps to isolate and verify true weak signals.

Experimental results show that C-Unseen significantly outperforms conventional keyword-, topic-, and graph-based baselines.

论文元数据

  • arXiv ID: arXiv:2608.26870 [cs.AI]
  • 作者: Yassir Lairgi, Ludovic Moncla, Khalid Benabdeslem, Rémy Cazabet, Pierre Cléau
  • 提交日期: 2026年8月27日
  • 主学科: 人工智能 (cs.AI)
  • 次学科: 计算与语言 (cs.CL)、社会与信息网络 (cs.SI)
  • 会议背景: 已被 WISE 2026 会议的 AI4SE 2026 特别分会场（Special Track）接受
  • arXiv ID: arXiv:2608.26870 [cs.AI]
  • Authors: Yassir Lairgi, Ludovic Moncla, Khalid Benabdeslem, Rémy Cazabet, Pierre Cléau
  • Submission Date: August 27, 2026
  • Primary Subject: Artificial Intelligence (cs.AI)
  • Secondary Subjects: Computation and Language (cs.CL), Social and Information Networks (cs.SI)
  • Conference Context: Accepted at the AI4SE 2026 Special Track, held within the WISE 2026 Conference

摘要

弱信号是预示重大变革发生的早期、低可见度指标，通常在变革确立之前出现。现有的检测方法基于关键词频率、主题建模或无类型图拓扑结构，无法捕捉这些信号赖以展现的语义和关系结构。在本文中，我们提出了 C-Unseen，这是一个用于动态时序知识图谱（DTKGs）中弱信号检测的可自我解释框架。我们将弱信号定义为在连续的时序知识图谱快照中不断蔓延的稀疏且语义连贯的子图。该框架通过两个模块运作：稀疏子图提取器，其中大模型通过思维链推理识别出内容与快照主流叙事存在张力的子图；弱信号警报器，通过追踪这些稀疏子图跨时间步的持久性来隔离真正的弱信号。实验结果表明，C-Unseen 的性能优于基于关键词、主题和图的基线方法。

Weak signals are early, low-visibility indicators that precede significant changes before those changes become established. Existing detection methods, based on keyword frequency, topic modeling, or untyped graph topology, fail to capture the semantic and relational structure through which such signals manifest. In this paper, we propose C-Unseen, a self-interpretable framework for weak signal detection in Dynamic Temporal Knowledge Graphs (DTKGs). We define a weak signal as a rare, semantically coherent subgraph that proliferates across consecutive TKG snapshots. The framework operates through two modules: a Rare Subgraphs Extractor, in which an LLM identifies subgraphs whose content is in tension with the dominant snapshot narrative via chain-of-thought reasoning, and a Weak Signal Alerter, in which the persistence of these rare subgraphs is tracked across time steps to isolate true weak signals. Experimental results demonstrate that C-Unseen outperforms keyword-, topic-, and graph-based baselines.

全文与资源

引用与参考文献