Thai Economic Sentiment Dataset (2014–2023): Labeled Complaints and Full-Corpus Predictions from Social Media

This repository contains two datasets supporting the study "Public Perception Meets Economic Data: An Analysis of Online Sentiment and Economic Indicators in Thailand" (currently under peer review). Both datasets consist of Thai-language social media posts related to the economy, collected using the keyword เศรษฐกิจ ("economy") over the period January 2014 – December 2023. Files labeled_dataset.csv (772 rows) — the gold-standard, human-annotated dataset used to fine-tune the classification model. date: post date rawContent: post text (URLs redacted and replaced with [URL]) sentiment: gold label, one of complaint / not-complaint, assigned by three independent human annotators following a fixed labeling rubric (Fleiss' κ = 0.55) full_corpus_predictions.csv (24,670 rows) — the complete corpus of collected posts, with labels predicted by the fine-tuned WangchanBERTa classifier (test-set F1 = 0.87). date: post date rawContent: post text (URLs redacted and replaced with [URL]) predicted_label: model-assigned label, one of complaint / not-complaint Collection and annotation. Posts were collected from a single, widely used Thai social media platform. Annotators were calibrated on a shared batch of 50 posts against a written rubric before independently labeling the remainder of the gold dataset. Full methodological detail, including annotation guidelines, class-imbalance handling, and model training procedure, is described in the accompanying manuscript. Privacy. All posts were publicly accessible at the time of collection. No usernames, handles, or other direct identifiers are included. URLs embedded in post text have been redacted and replaced with the placeholder [URL]. Citation. If you use this dataset, please cite the associated manuscript (citation to be added upon publication; in the meantime, please cite this Zenodo record).