From Noise to Signal: When Outliers Seed New Topics

TL;DR

This paper demonstrates that certain outliers in dynamic topic modeling can serve as early indicators of emerging topics, linking weak signals with temporal evolution in news data.

Contribution

It introduces a temporal taxonomy of document trajectories that distinguishes anticipatory outliers from reinforcing or isolated documents, enhancing early detection of new topics.

Findings

High consensus on anticipatory outliers among models

Retrospective evaluation on French news corpus

Qualitative case studies illustrate trajectory types

Abstract

Outliers in dynamic topic modeling are typically treated as noise, yet we show that some can serve as early signals of emerging topics. We introduce a temporal taxonomy of news-document trajectories that defines how documents relate to topic formation over time. It distinguishes anticipatory outliers, which precede the topics they later join, from documents that either reinforce existing topics or remain isolated. By capturing these trajectories, the taxonomy links weak-signal detection with temporal topic modeling and clarifies how individual articles anticipate, initiate, or drift within evolving clusters. We implement it in a cumulative clustering setting using document embeddings from eleven state-of-the-art language models and evaluate it retrospectively on HydroNewsFr, a French news corpus on the hydrogen economy. Inter-model agreement reveals a small, high-consensus subset of anticipatory outliers, increasing confidence in these labels. Qualitative case studies further illustrate these trajectories through concrete topic developments.