[Paper Review] Using Twitter to Predict Sales: A Case Study

Remco Dijkman, Panagiotis G. Ipeirotis+2 more|arXiv (Cornell University)|Mar 16, 2015

Digital Marketing and Social Media45 references20 citations

TL;DR

This study demonstrates that classifying Twitter sentiment—particularly positive tweets from individual users—enables accurate prediction of sales for low-attention products. Using a four-country case study, the authors show that spikes in personal, positive tweets strongly correlate with subsequent sales increases, outperforming raw tweet volume and highlighting the importance of sentiment classification and advanced statistical modeling over simple correlation.

ABSTRACT

This paper studies the relation between activity on Twitter and sales. While research exists into the relation between Tweets and movie and book sales, this paper shows that the same relations do not hold for products that receive less attention on social media. For such products, classification of Tweets is far more important to determine a relation. Also, for such products advanced statistical relations, in addition to correlation, are required to relate Twitter activity and sales. In a case study that involves Tweets and sales from a company in four countries, the paper shows how, by classifying Tweets, such relations can be identified. In particular, the paper shows evidence that positive Tweets by persons (as opposed to companies) can be used to forecast sales and that peaks in positive Tweets by persons are strongly related to an increase in sales. These results can be used to improve sales forecasts and to increase sales in marketing campaigns.

Motivation & Objective

• To investigate whether Twitter activity can predict real-world sales, especially for products receiving limited social media attention.
• To address the limitations of using raw tweet volume by exploring sentiment classification as a more effective predictor.
• To develop and validate a method that combines sentiment classification with statistical modeling to improve sales forecasting accuracy.
• To examine the differential impact of tweets from individuals versus organizations on sales outcomes.

Proposed method

• The study collects Twitter data and sales data from a company operating in four countries over a defined period.
• Tweets are classified into categories, with a focus on sentiment (positive, negative, neutral) and source (individuals vs. organizations).
• Statistical models are applied to analyze the temporal relationship between tweet volume—especially positive individual tweets—and sales trends.
• Advanced statistical techniques, including time-series analysis and correlation modeling, are used to identify predictive patterns beyond simple correlation.
• The model is validated across multiple countries to assess generalizability and robustness of findings.
• Sentiment classification is performed using natural language processing techniques to distinguish between personal and corporate tweets.

Experimental results

Research questions

• RQ1 Can positive sentiment in individual-user tweets predict future sales increases for low-attention products?
• RQ2 How does the source of a tweet (individual vs. organization) affect its predictive power for sales?
• RQ3 Does raw tweet volume outperform sentiment-classified tweet data in forecasting sales?
• RQ4 To what extent do peaks in positive individual tweets precede and predict sales spikes?
• RQ5 Can statistical models incorporating sentiment and temporal dynamics improve sales forecasting accuracy compared to correlation-based approaches?

Key findings

• Positive tweets from individual users show a strong, statistically significant correlation with subsequent increases in sales, especially when spikes occur.
• Tweets from organizations do not exhibit the same predictive power, indicating that user-generated content is more informative for sales forecasting.
• Raw tweet volume alone fails to predict sales for low-attention products, underscoring the need for sentiment classification.
• Peaks in positive individual tweets consistently precede and predict sales surges, suggesting a causal or leading relationship.
• Statistical modeling that incorporates sentiment and temporal dynamics significantly improves forecasting accuracy compared to simple correlation.
• The predictive relationship holds across multiple countries, indicating cross-cultural validity of the approach.