Table 11 Best method for each dataset - 3-class experiments
From: SentiBench - a benchmark comparison of state-of-the-practice sentiment analysis methods
Dataset | Method | F 1-Pos | F 1-Neg | F 1-Neu | Macro- F 1 |
|---|---|---|---|---|---|
Comments_BBC | Semantria | 36.76 | 67.94 | 43.49 | 49.40 |
Comments_Digg | Umigon | 49.62 | 62.04 | 44.27 | 51.98 |
Comments_NYT | SO-CAL | 56.99 | 60.08 | 15.34 | 44.14 |
Comments_TED | Opinion Lexicon | 64.95 | 56.59 | 30.77 | 50.77 |
Comments_YTB | LIWC15 | 73.68 | 49.72 | 48.79 | 57.4 |
Myspace | LIWC15 | 78.83 | 41.74 | 43.76 | 54.78 |
Tweets_DBT | Opinion Lexicon | 43.44 | 47.71 | 48.84 | 46.66 |
Tweets_RND_I | Umigon | 60.53 | 51.39 | 65.22 | 59.05 |
Tweets_RND_III | Umigon | 63.33 | 57.00 | 82.10 | 67.47 |
Tweets_RND_IV | Umigon | 75.86 | 76.33 | 71.54 | 74.58 |
Tweets_SAN | Umigon | 44.16 | 45.95 | 70.45 | 53.52 |
Tweets_Semeval | Umigon | 64.28 | 46.41 | 73.13 | 61.27 |
RW | Sentiment140 | 62.24 | 51.17 | 42.66 | 52.02 |