EmoTweetJawa: A Dataset of Sentiment-Annotated Low-Resource Javanese Tweets

Overview EmoTweetJawa is an open dataset of Javanese-language tweets collected from Twitter using Javanese-identity hashtags (#jawapride, #bosojowo, #quotesjowo, #jowobijak, etc.), human-annotated into three sentiment classes: Positif (Positive), Negatif (Negative), and Netral (Neutral). The dataset is designed to support research on sentiment analysis and natural language processing for Javanese, a low-resource regional language of Indonesia spoken natively by an estimated 68 million people. Each of the 1,428 tweets was independently labelled by three human annotators, reaching a Fleiss' Kappa of 0.59 (moderate-to-substantial agreement); disagreements were resolved by majority vote and manual reconciliation to produce a reliable ground-truth label for every tweet. Dataset Structure & Files The dataset includes the following files: EmoTweetJawa-Corpus.csv: raw tweet text and public engagement metadata (date, retweet/reply/favorite/quote counts) for each of the 1,428 annotated tweets. User mentions in the full_text column are replaced with the placeholder "@user" and t.co shortened links are removed, to avoid redistributing personally identifiable information. EmoTweetJawa-Corpus_EN.csv: same as above, with an added english_translated column (machine-translated); the same anonymization (user mentions replaced with "@user", t.co links removed) is applied to both the full_text and english_translated columns. EmoTweetJawa-Annotations.csv: the individual sentiment label assigned by each of three independent human annotators for every cleaned tweet. EmoTweetJawa-Annotations_EN.csv: same as above, with an added english_translated column. EmoTweetJawa-GroundTruth.csv: the final, reconciled sentiment label for each tweet, resolved by majority vote with manual reconciliation of remaining ties. EmoTweetJawa-GroundTruth_EN.csv: same as above, with an added english_translated column. Key Features & Fields Temporal Scope: tweets dated 2011-2025 (collected in 2024-2025). Language: Javanese (jv) and translated English (en). Annotations: 3-class sentiment labels (Positive/Negative/Neutral) from 3 independent human annotators plus a reconciled ground truth. Usage & Ethical Considerations This dataset is provided strictly for academic, educational, and research purposes. Only tweet text and public engagement metadata are released; no usernames, user IDs, or other personally identifiable information are included. Re-users must comply with Twitter's Developer Agreement and Policy and applicable data protection/privacy regulations regarding social media data distribution.