Skip to content

Latest commit

 

History

History
52 lines (30 loc) · 1.41 KB

emotion_classification.md

File metadata and controls

52 lines (30 loc) · 1.41 KB
Raw

Chinese Emotion Classification

Background

Emotion classification identifies the emotional state of a writer or speaker. This is distinct from sentiment classification, which describes the viewpoint of writers toward their subjects.

Example input/output

Input:

讨厌！你骗我！

Output:

Angry

Standard metrics

  • Accuracy of classification.
  • F1 score.

Cheng emotion corpus.

Cheng et al 2017 introduce an emotion corpus for Chinese Microblogs. It consists of short posts marked with the following distribution of emotion tags: Joy (11.3%), Angry (3.5%), Sad (2.6%), Fearful (0.6%), Positive (8.2%), Neutral (4.4%), Negative (9.9%), Non-emotion (59.5%). Furthermore, the corpus identifies the sub-span of the post that is the cause of the emotion.

Chen et al 2018 reports that the corpus includes “~3,000 subtweets, ~11,000 instances for EClass, and ~13,000 instances for ECause.”

Test set Genre
Cheng emotion corpus Microblog

Metrics

  • F1 score.

Results

System F1
[Chen et al 2018] 62.4
[Cheng et al 2017] 58.2

Suggestions? Changes? Please send email to [email protected]