Detecting Priming News Events

Di Wu, Yiping Ke, Jeffrey Xu Yu, Zheng LiuarXiv 2012 · OpenReview

Abstract

We study a problem of detecting priming events based on a time series index and an evolving document stream. We define a priming event as an event which triggers abnormal movements of the time series index, i.e., the Iraq war with respect to the president approval index of President Bush. Existing solutions either focus on organizing coherent keywords from a document stream into events or identifying correlated movements between keyword frequency trajectories and the time series index. In this paper, we tackle the problem in two major steps. (1) We identify the elements that form a priming event. The element identified is called influential topic which consists of a set of coherent keywords. And we extract them by looking at the correlation between keyword trajectories and the interested time series index at a global level. (2) We extract priming events by detecting and organizing the bursty influential topics at a micro level. We evaluate our algorithms on a real-world dataset and the result confirms that our method is able to discover the priming events effectively.

Introduction

In an era of 24-hour news cycles and social media deluges, thousands of events occur daily. However, only a handful actually change public perception or market behavior. Researchers have developed a framework to distinguish the signal from the noise by identifying priming events: news stories that trigger significant, measurable shifts in public indices like presidential approval ratings or the stock market. Traditional Topic Detection and Tracking (TDT) often fails to identify which topics truly matter because it ignores external "ground truth" indices. This paper introduces a two-step framework to automatically detect these events by correlating evolving text streams with time-series volatility.

Methodology

The proposed approach moves from a "global" understanding of topics to a "micro" detection of specific events through two primary phases:

1. Global Influential Topic Detection

Instead of grouping words based solely on co-occurrence, the algorithm groups keywords that correlate with the volatility (standard deviation of returns) of an external index. This ensures the system captures "impact" regardless of whether the news is perceived as positive or negative. For example, a topic is only considered "influential" if its historical appearance coincides with index turbulence.

2. Micro Priming Event Detection

Within specific time windows, the system identifies local topic clusters to represent snapshots of an event. It then uses "composite path detection" to connect these snapshots over time, tracking an event from its initial burst to its eventual fade-out. Events are ranked using a Priming Score that combines keyword intensity, index volatility, and the linear correlation between news intensity and index movement.

Experimental Results

The framework was evaluated on three distinct real-world datasets: the Bush era (8 years of news), the Obama era (13 months of news), and the "Tsunami" financial dataset (3 years of news paired with the S&P 500).

Figure 1: Approval rating of President Obama annotated with key news events, illustrating the correlation between specific political actions and public sentiment shifts.

The system successfully identified major historical priming moments with high accuracy. In the Bush dataset, the 9/11 Terrorist Attack was the top-ranked event. In the Obama dataset, as shown in Figure 1, the system correctly identified the "Prof. Gates police arrest" controversy in July 2009 as a major driver of a sudden approval drop. In the financial dataset, the system pinpointed the Lehman Brothers bankruptcy and the subprime mortgage crisis as the primary drivers of market volatility. When compared to baseline "bursty event" detectors that ignore index data, this method significantly outperformed them by filtering out news events that had zero correlation with actual market or social movements.

Conclusion and Future Work

By focusing on index volatility and contextual grouping, this framework avoids flagging "trivial" topics that do not impact public sentiment. While the system is dependent on the frequency of the index updates and the breadth of the news database, it provides a robust method for identifying influential narratives. Future applications could include tracking how celebrity news affects brand sentiment or how product launches influence consumer electronics indices.

Summary updated

Abstract

We study a problem of detecting priming events based on a time series index and an evolving document stream. We define a priming event as an event which triggers abnormal movements of the time series index, i.e., the Iraq war with respect to the president approval index of President Bush. Existing solutions either focus on organizing coherent keywords from a document stream into events or identifying correlated movements between keyword frequency trajectories and the time series index. In this paper, we tackle the problem in two major steps. (1) We identify the elements that form a priming event. The element identified is called influential topic which consists of a set of coherent keywords. And we extract them by looking at the correlation between keyword trajectories and the interested time series index at a global level. (2) We extract priming events by detecting and organizing the bursty influential topics at a micro level. We evaluate our algorithms on a real-world dataset and the result confirms that our method is able to discover the priming events effectively.
BibTeX
@misc{wu2012detecting,
  author = {Di Wu and Yiping Ke and Jeffrey Xu Yu and Zheng Liu},
  title = {{Detecting Priming News Events}},
  year = {2012},
  eprint = {1201.3458},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1201.3458},
}
Figure