A dataset of post-comment on Pantip, a popular Thai web board.

main.py hosts all web crawler code for both posts and comments. example/ folder hosts an example on how to use dataset with machine learning techniques with result. Inside the folder also contains the result of the program. post-processing/ folder hosts post-processing script and result. small-dataset-generator/ folder contains utility script for generating a small version of dataset. labeled/ folder contains labeled data. baseline-model/ contains code for three basedline model (LSTM, CNN, BERT) trained on the labeled dataset.

The final version of the datasets is located at post-processing/posts.csv and post-processing/comments.csv. You can also grab the unprocessed version of the dataset at posts.csv and comments.csv.

There are labeled version of small dataset located at labeled/comments-small-labeled.csv. The small version of comments selected the first ten comments of each posts. It has a classification label for sentiment of the text by considering the post body context. The class labels are NEU for neutral, NEG for negative, and POS for positive comments.

There are three possible labels for setiment labeling. POS for positive comments, NEU for neutral comments, and NEG for negative comments.

• Good vibe
• Expressing gratitude
• Encouraging people
• Optimistic comment

• Containing bad words
• Expressing anger, sadness
• Discouraging people
• Pessimistic comment

• Discussion/Explanation without emotion
• Tell a story
• Mixture of positive and negative stories
• Does not belong to either POS or NEG

You can help contribute to the annotated corpus by either

1. Annotate other parts of the dataset
2. Re-annotate the annotated dataset for cross checking

Please also to the Label explanation section above for more

Please feel free to open any pull requests for contributing to the repository either fixing bugs, add more model, or add more data.