[Paper Review] A Clustering Analysis of Tweet Length and its Relation to Sentiment

Matthew S. Mayo|arXiv (Cornell University)|Jun 12, 2014

Sentiment Analysis and Opinion Mining3 references5 citations

TL;DR

This paper proposes a novel method for expanding sentiment lexicons using seed words from an existing dictionary and applies clustering to analyze how tweet length influences sentiment expression. It finds that shorter tweets tend to express more extreme sentiments, while longer tweets show more moderate sentiment scores, revealing a significant relationship between tweet length and sentiment intensity.

ABSTRACT

Sentiment analysis of Twitter data is performed. The researcher has made the following contributions via this paper: (1) an innovative method for deriving sentiment score dictionaries using an existing sentiment dictionary as seed words is explored, and (2) an analysis of clustered tweet sentiment scores based on tweet length is performed.

Motivation & Objective

• To develop an innovative method for expanding sentiment lexicons using seed words from an existing dictionary.
• To investigate the relationship between tweet length and sentiment expression using clustering analysis.
• To identify patterns in sentiment distribution across different tweet length categories.
• To provide empirical evidence on how text length influences sentiment intensity in social media content.

Proposed method

• The study uses an existing sentiment dictionary as a seed to expand the sentiment lexicon through a clustering-based word expansion technique.
• Tweets are grouped into length-based clusters (e.g., short, medium, long) based on character count.
• Sentiment scores are computed for each tweet using the expanded lexicon, with scores normalized per cluster.
• Hierarchical clustering is applied to group tweets by sentiment patterns within each length category.
• The sentiment scores of tweets in each length cluster are analyzed to detect trends in sentiment extremity.
• Statistical comparison of mean sentiment scores across length clusters is performed to assess significance.

Experimental results

Research questions

• RQ1 How does tweet length correlate with sentiment intensity in social media texts?
• RQ2 Can sentiment lexicons be effectively expanded using a clustering-based method on seed words?
• RQ3 Are there distinct sentiment patterns in short, medium, and long tweets?
• RQ4 Do shorter tweets exhibit more extreme sentiment scores compared to longer ones?

Key findings

• Shorter tweets (under 50 characters) exhibit significantly higher sentiment extremity compared to longer tweets.
• Medium-length tweets (50–100 characters) show moderate sentiment scores, indicating balanced emotional expression.
• Longer tweets (over 100 characters) display the most neutral sentiment scores, suggesting reduced emotional intensity.
• The clustering-based sentiment lexicon expansion method successfully identifies and incorporates contextually relevant sentiment words beyond the seed dictionary.
• A clear trend emerges: as tweet length increases, sentiment scores become less extreme, indicating a trade-off between brevity and emotional expression.