[HCI Perspectives] Beyond the Algorithm: Bridging Emotional Computing and Discourse Analysis in the Brexit Debate

TL;DR

This research addresses the inherent limitations of automated sentiment analysis in capturing the complexity of political discourse. By applying a dual-method approach—SentiStrength for quantitative emotional tagging and Discourse Analysis for qualitative inquiry—the study reveals that what algorithms label as "neutral" often contains sophisticated rhetorical strategies such as fear-mongering and xenophobia.

Background & Motivation: The "Neutrality" Trap

In the era of viral social media, governments and companies rely on Emotional Computing to gauge public pulse. However, the author argues that current SOTAs are often blinded by the "packaging" of language. Social media users rarely express dissent in flat, easily identifiable terms; instead, they use irony, metaphors, and cultural stereotypes.

The problem is twofold:

1. Loss of Context: Tools like SentiStrength often ignore stop-words or connectors that fundamentally shift a sentence's meaning.
2. False Neutrality: As seen in the Brexit study, many tweets are scored as "not negative" (-1) or "not positive" (1) because they lack explicit emotional keywords, even when their underlying message is highly provocative.

Methodology: The Two-Stage Synergy

The author proposes a pipeline where Emotional Computing serves as the "filter" and Discourse Analysis serves as the "microscope."

1. Quantitative Sentiment Tagging

Using SentiStrength, a corpus of 200 #Brexit tweets was analyzed. The tool provides a dual-score system, acknowledging that a single human sentence can contain both positive and negative valences simultaneously.

2. Qualitative Discourse Refinement

This stage looks at what the numbers missed. The researcher analyzed the specific communicative strategies used in the tweets, such as:

• Generation of Fear: Tweets predicting the loss of free health cover without factual backing.
• Media Distrust: Claims of media outlets being "forced" to retract Brexit benefits.
• Social Polarization: Identifying Brexit as a "Tory proxy war," a nuance an algorithm would likely miss.

Experimental Insights: The Invisible Conflict

The experiment highlights a massive gap between automated scoring and human interpretation. While the tables below show a overwhelming cluster around "neutral" values (-1 and 1), the qualitative analysis found the dataset to be highly charged.

Key findings from the discourse phase:

• Xenophobia Detection: A tweet asking, "Do you want more Mosques in your country?" might not trigger a "negative sentiment" score in many basic lexicons but represents a clear exclusionary discourse.
• Assumption vs. Fact: Users frequently used assumptions to emphasize negative implementation effects, aiming to trigger emotional responses (fear) rather than logical debate.

Critical Analysis & Conclusion

Takeaway

The paper confirms that while Emotional Computing is a "great help" for systematic study, the ambiguity of language remains a formidable barrier. The core contribution is the validation of a hybrid model where qualitative experts "audit" the findings of the automated tools.

Limitations

• Sample Size: The study uses a relatively small corpus of 200 tweets.
• Manual Effort: The reliance on manual discourse analysis limits the scalability of the "refinement" stage for Big Data applications.

Future Outlook

The author suggests that future work should expand this integrated approach to larger datasets. Within the field of AI, this points toward a need for Neuro-Symbolic approaches or LLMs that can better simulate the "interpretive" nature of Discourse Analysis, moving beyond simple keyword-based sentiment.