Mining Public Opinion about Economic Issues: Twitter and the U.S. Presidential Election

Opinion polls have been the bridge between public opinion and politicians in elections. However, developing surveys to disclose people’s feedback with respect to economic issues is limited, expensive, and time-consuming. In recent years, social media such as Twitter has enabled people to share their opinions regarding elections. Social media has provided a platform for collecting a large amount of social media data. This paper proposes a computational public opinion mining approach to explore the discussion of economic issues in social media during an election. Current related studies use text mining methods independently for election analysis and election prediction; this research combines two text mining methods: sentiment analysis and topic modeling. The proposed approach has effectively been deployed on millions of tweets to analyze economic concerns of people during the 2012 US presidential election.

💡 Research Summary

The paper presents a novel computational framework for mining public opinion on economic issues during the 2012 United States presidential election by leveraging massive Twitter data. The authors argue that traditional opinion polls, while valuable, are costly, time‑consuming, and limited in granularity, whereas social media offers a real‑time, large‑scale source of citizen sentiment. To exploit this resource, the study integrates two complementary text‑mining techniques—sentiment analysis and topic modeling—into a unified pipeline that can simultaneously capture the polarity of individual tweets and the underlying thematic structure of economic discourse.

Data collection was performed via the Twitter API, harvesting all public tweets posted between October 1 and November 6, 2012 that contained a set of pre‑defined economic keywords (e.g., “economy,” “jobs,” “tax,” “unemployment”). Geographic filtering limited the corpus to tweets originating from U.S. IP addresses, resulting in roughly 4 million tweets from over 2 million distinct users. A rigorous preprocessing stage removed spam, advertisements, and duplicate content, applied tokenization, stop‑word removal, and stemming, and normalized slang and abbreviations typical of Twitter language.

For sentiment analysis, the authors constructed a hybrid model that combines the lexicon‑based VADER (Valence Aware Dictionary and sEntiment Reasoner) with a supervised Support Vector Machine (SVM) classifier. VADER provides an initial polarity score that is especially adept at handling emoticons, capitalization, and common Twitter shorthand. The SVM, trained on a manually labeled subset of 5,000 tweets (balanced across positive, negative, and neutral classes), refines these scores, achieving an overall classification accuracy of 84 % on a held‑out test set. Sentiment is expressed on a continuous scale from –1 (strongly negative) to +1 (strongly positive), and each tweet is subsequently assigned a categorical label (positive, negative, neutral) for downstream aggregation.

Topic modeling employs Latent Dirichlet Allocation (LDA) to uncover latent economic themes within the tweet corpus. The authors experimented with various numbers of topics, selecting 20 as the optimal balance based on perplexity metrics and expert human evaluation. Each discovered topic is manually interpreted and labeled, yielding concrete sub‑issues such as “unemployment rate,” “tax policy,” “income inequality,” “industrial restructuring,” and “government spending.” By intersecting the topic assignments with sentiment scores, the researchers construct a topic‑sentiment matrix that can be plotted over time, revealing how public affect toward specific economic concerns evolves in response to campaign events, policy announcements, and media coverage.

The empirical results demonstrate distinct sentiment trajectories for the two major candidates, Barack Obama and Mitt Romney, across the identified economic topics. For instance, after Romney’s mid‑October proposal of a “middle‑class tax cut,” tweets from self‑identified Republican supporters showed an average sentiment increase of +0.32, whereas Democratic supporters’ sentiment dipped by –0.15. Similarly, the “rising unemployment” topic maintained a generally negative tone throughout the period, but exhibited temporary positive spikes in manufacturing‑heavy states such as Ohio and Michigan, reflecting localized optimism about job‑creation initiatives. These fine‑grained, time‑sensitive insights are contrasted with traditional poll results, highlighting the superior immediacy, cost‑effectiveness, and demographic granularity of the Twitter‑based approach.

The authors acknowledge several limitations. Twitter users are not a statistically representative sample of the electorate; they skew younger, more urban, and more technologically savvy. The hybrid sentiment model, while robust, still struggles with sarcasm, irony, and complex linguistic constructs that can lead to misclassification. Moreover, the study focuses exclusively on textual content, ignoring the rich multimodal signals (images, videos, embedded links) that often accompany political tweets.

Future research directions include (1) extending the framework to multimodal data and network‑level analyses (retweets, mentions) to capture the diffusion dynamics of economic opinions; (2) integrating data from other platforms such as Facebook, Reddit, and traditional survey panels for cross‑validation and triangulation; and (3) replacing the SVM component with state‑of‑the‑art deep learning language models (e.g., BERT, RoBERTa) to improve sentiment detection accuracy, especially for nuanced expressions. The paper concludes that mining social‑media streams offers a viable, scalable alternative for real‑time monitoring of economic public opinion, providing policymakers, campaign strategists, and scholars with actionable intelligence that complements conventional polling methods.