https://github.com/jacksonchen1998/chinese-dimensional-sentiment-analysis

2024 NYCU Natural Language Processing
https://github.com/jacksonchen1998/chinese-dimensional-sentiment-analysis

Last synced: 2 months ago
JSON representation

2024 NYCU Natural Language Processing

• Host: GitHub
• URL: https://github.com/jacksonchen1998/chinese-dimensional-sentiment-analysis
• Owner: jacksonchen1998
• Created: 2024-04-15T06:08:41.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2024-05-14T06:04:27.000Z (about 2 years ago)
• Last Synced: 2025-12-09T13:03:56.780Z (6 months ago)
• Language: Jupyter Notebook
• Size: 559 KB
• Stars: 1
• Watchers: 1
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Chinese-dimensional-sentiment-analysis

2024 NYCU Natural Language Processing

## Method

This method aims to leverage a set of six independently trained BERT-based models, each specializing in a different category. 

The final prediction is derived from the average of all model outputs, providing a robust and reliable estimate.

## Evaluation

Mean Absolute Error (MAE):

$$

  MAE = \frac{1}{n} \sum_{i=1}^n |a_i - p_i|

$$

Person Correlation Coefficient ($r$):

$$

  r = \frac{1}{n-1} \sum_{i}^n (\frac{a_i - \mu_A}{\sigma_A})(\frac{p_i - \mu_P}{\sigma_A})

$$

- $a_i \in A$: $i$-th actual value

- $p_i \in P$: $i$-th predicted value

- $\mu$: mean value

- $\sigma$: standard deviation

- $n$: the number of test sample

A lower MAE and a higher r indicate more accurate prediction performance.

## Experiment

### Dataset

- Training Dataset: Chinese EmoBank (CVAT)

- Testing Dataset: 1000+ Mental Health Texts

### On Chinese EmoBank (CVAT) with Bert

  

    

      

        Method

      

      

        MAE

      

      

        r

      

    

    

      

        Valence

      

      

        0.3342

      

      

        0.95883

      

    

    

      

        Arousal

      

      

        0.36525

      

      

        0.88947

      

    

  

> Valence

![bert_v_mae](./image/bert_v_mae.png)

![bert_v_r](./image/bert_v_r.png)

> Arousal

![bert_a_mae](./image/bert_a_mae.png)

![bert_a_r](./image/bert_a_r.png)

### On private dataset with Bert

  

    

      

        Method

      

      

        MAE

      

      

        r

      

    

    

      

        Valence

      

      

        0.75

      

      

        0.68

      

    

    

      

        Arousal

      

      

        1.1

      

      

        0.47

      

    

  

### On Chinese EmoBank (CVAT) with MoE Bert

### On private dataset with MoE Bert

## Reference

- [bert-base-chinese](https://huggingface.co/google-bert/bert-base-chinese)

- [Chinese EmoBank: Building Valence-Arousal Resources for Dimensional Sentiment Analysis](https://dl.acm.org/doi/full/10.1145/3489141)