project-presentation
0 likes144 views
The document summarizes a research project that analyzed the relationship between tweet length and sentiment. It clustered tweets captured over 4 weeks based on their length and sentiment score. The results showed shorter tweets had sentiment scores more tightly centered around their assigned cluster centroid, while longer tweets were less centered as the increased number of terms provides more opportunities to influence the sentiment score.
project-presentation
- 1. A CLUSTERING ANALYSIS OF TWEET LENGTH AND ITS RELATION TO SENTIMENT
- 2. Research Project Matthew Mayo School of Computer Science Columbus State University Columbus, GA CPSC 6185 Intelligent Agents Dr. Rania Hodhod
- 3. Twitter • Popular microblogging web service • 140 character per message (tweet) limit • Started in 2006, over 645 million users today* • 58 million tweets per day* • 9,000 tweets per minute* * Source: www.statisticbrain.com/twitter-statistics
- 4. Sentiment Analysis • Identifying, extracting & processing subjective information from source material • Subjective information includes attitudes, emotions & opinions • Appropriate for binary classification (positive vs. negative, good vs. bad, etc.) • Useful for movie reviews, political election opinions, etc.
- 5. Project Aim Interested in exploring the relationship between: • Length of tweet (number of characters) AND • Sentiment score of tweet
- 6. Problem Description The research project tasks: 1. Capture Twitter data 2. Build custom sentiment dictionary 3. Process tweets 4. Create dataset 5. Cluster tweet data
- 7. Methodology ● Custom Python scripts to capture and process live tweets over 4 week schedule ● Use k-means clustering in Weka to look for natural sentiment patterns ● Any correlation between length of tweet and its sentiment (positive/negative/neutral)?
- 8. Results Sentiment scores of shorter tweets appear more tightly-centered around their cluster’s centroid Longer tweets become less-centered on the applicable centroid As the number of characters in a tweet would lead to a greater number of terms, which would increase the chances of terms being assigned a score, this seems intuitive
- 9. Results