NTU NLP 2017 - Term Project 1

Hotel Review Opinion Mining

Task

From a hotel review as the following,

總體感覺充其量四星吧，根本達不到五星。尤其是二樓的西餐廳的飯菜實在做的不敢恭維，根本不像西餐，口味也有待提高，品種也少。環境有點亂哄哄，唯一的優勢可能是便宜。

extract aspect-specific opinions like:

• 環境 negative
• 價格 positive
• 餐廳 negative

Data

• aspect_review.txt - small review dataset with aspect-polarity labels
• polarity_review.txt - large review dataset with only polarity labels
• NTUSD_pos.txt / NTUSD_neg.txt - sentiment dictionary
• aspect_term.txt - example terms that refer to certain aspect
• test_review.txt - test reviews to extract aspect-specific opinions
• test.csv - test questions
• sample_submission.csv
• Preprocessed data (Segmentation)
  • test_seg.csv - test.csv combine with test_review.txt
  • aspect_review_seg.csv
  • polarity_review_seg.csv
• Preprocessed data (Segmentation & Pos tagged)
  • test_posseg.csv - test.csv combine with test_review.txt
  • aspect_review_posseg.csv
  • polarity_review_posseg.csv

Public Data Label Distribution

Label	Freq
-1	0.23963
0	0.57719
1	0.18318