Comparing values and sentiment using Mechanical Turk

Human values can help to explain people’s sentiment toward current events. In this experiment, we compare people’s values with their agreement or disagreement with paragraphs that were classified as either supporting or opposing a specific topic. We found that five value types have statistically significant agreement (p<0.001) for both the supporting and opposing paragraphs, in opposite directions. We hope to use these paragraph ratings to train an automatic text classifier to agree or disagree with paragraphs based on a specific value profile. Categories and Subject Descriptors I.2.7 [Artificial Intelligence]: Natural Language Processing – discourse, text analysis. General Terms Experimentation, Human Factors. Keywords Human values, sentiment analysis, text classification.

• Type: Journal
• Year: 2011
• Where: ICONFERENCE
• Authors: Thomas Clay Templeton, Kenneth R. Fleischmann, Jordan L. Boyd-Graber
• Keywords: Human Values, ICONFERENCE 2011, Information Technology, Keywords Human Values, Paragraphs