A new study published on arXiv investigates the phenomenon of rating-sentiment incongruence in online reviews, specifically examining tourism attraction reviews from Sri Lanka. Researchers found that in 18.6% of the analyzed 16,156 reviews, the sentiment expressed in the text did not align with the star rating provided. The study identified specific user behaviors, such as "Conservative Rater" and "Obligatory 5-Star," as primary drivers of this mismatch, with museums exhibiting the highest rates of incongruence. Factors like venue type, reviewer expertise, review length, and temporal elements were found to contribute to this divergence, highlighting that star ratings should not be automatically treated as reliable ground-truth labels in natural language processing tasks. AI
IMPACT Highlights limitations of using star ratings as ground truth for NLP tasks, potentially impacting sentiment analysis model development.
RANK_REASON Academic paper published on arXiv detailing a new analysis of rating-sentiment incongruence. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Conservative Rater
- Hugging Face
- natural language processing
- Obligatory 5-Star
- random forest
- Shap
- Sri Lanka
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
