The effect of wording on message propagation: Topic- and author-controlled natural experiments on Twitter (1405.1438v1)

Abstract: Consider a person trying to spread an important message on a social network. He/she can spend hours trying to craft the message. Does it actually matter? While there has been extensive prior work looking into predicting popularity of social-media content, the effect of wording per se has rarely been studied since it is often confounded with the popularity of the author and the topic. To control for these confounding factors, we take advantage of the surprising fact that there are many pairs of tweets containing the same url and written by the same user but employing different wording. Given such pairs, we ask: which version attracts more retweets? This turns out to be a more difficult task than predicting popular topics. Still, humans can answer this question better than chance (but far from perfectly), and the computational methods we develop can do better than both an average human and a strong competing method trained on non-controlled data.

Analysis of Wording Effects on Twitter Message Propagation

The research presented in this paper investigates the nuanced effects of message wording on social media propagation, specifically on the platform Twitter. While previous studies have typically focused on predicting overall popularity of social media content, the unique contribution of this work lies in specifically isolating the influence of wording, independent of other influential factors such as author popularity and topic interest. To achieve this level of control, the authors leveraged what they describe as "natural experiments" in the form of tweet pairs posted by the same user, linking to the same URL but with differing textual content.

Research Methodology

This paper's research method involves a sophisticated data collection strategy, gathering over 1.77 million topic- and author-controlled tweet pairs from Twitter, which are paired by identical URL links despite differing in original textual content. This unusual collection technique allows for a controlled comparison sans external biases such as timing effects or fan base size that normally plague broader tweet studies. A strict filtering procedure further focused the dataset to 11,404 pairs by excluding pairs with insignificant differences in text and ambiguous retweet counts.

The paper then proceeds with a detailed exploration of multiple linguistic features that could potentially impact retweet rates. These features range from explicit sharing requests (e.g., "please retweet") to stylistic elements such as news headline resemblance, informativeness, sentiment expressions, and conformity to community language norms. Human subjects were enlisted via AMT to identify more retweet-worthy tweets among selected pairs, achieving moderate accuracy, which the authors used as a benchmark for their computational models.

Key Findings

Through a series of computational experiments, the paper reports several insights:

1. Word Choice and Order Matter: Explicit requests to share and informativeness notably improve retweet rates. Messages with richer information tend to propagate more, challenging prior research on meme brevity.
2. Conformity and Headlines: Language conformity—be it to personal norms or general community norms—augments message success, as does mimicking attention-grabbing headlines.
3. Sentiment and Readability: Positive and negative sentiments generally assist propagation; however, readability scores did not fare as well in efficacy evaluations.
4. Predictive Models: The authors developed predictive models using logistic regression and various linguistic features, demonstrating notable success in predicting which tweet among a controlled pair would be more retweeted. This model outperformed a comparison trained on non-controlled data incorporating author and timing metadata.

Implications and Speculations

The findings of this paper underscore the importance of phrasing as a strategic tool for maximizing message reach on Twitter. For practitioners—especially those engaged in social media marketing, political campaigns, or corporate communications—the work suggests strategies focusing extensively on word choice, message richness, and alignment with community norms may be more effective than attention solely on content.

On a theoretical level, this research fuels the discourse on the mechanics of information spread in networked social systems, contributing to better models of virality that consider linguistic style alongside traditional factors.

Future Directions

Moving forward, the paper hints at promising research avenues such as the adaptability of these features to longer content forms and a deeper theory into the psychological and cultural factors that underpin wording effectiveness. This could lead to broader applications in understanding online communication dynamics, potentially transcending social media platforms to encompass digital communication more generally.

This paper offers a fine-grained approach to understanding social media virality, pioneering controlled experimentation in an otherwise chaotic digital environment, setting a foundation for subsequent research in networked communication.