ECHO-PM Dataset v1.0: GitHub Messages with Model-Based Sentiment/Emotion Labels, Empathetic Rewriting, and ROUGE/Perplexity Metrics

ECHO-PM Dataset (v1.0): GitHub Messages with Sentiment, Emotion, and Empathetic Rewriting Outputs This record provides a CSV dataset with 677 GitHub messages (issue bodies and comments) collected from public repositories and enriched with (i) text preprocessing artifacts, (ii) sentiment and emotion model outputs, (iii) heuristic rewrite strategies and reformulated messages, and (iv) automatic validation metrics (ROUGE-L and perplexity). Scope of this release (facts from the CSV):- Rows: 677 | Columns: 28- Repositories: pallets/flask (343), psf/requests (334)- Message types: comment (556), issue_body (121)- Time range (created_at): 2023-10-19 to 2025-11-27 Main contents:1) Metadata and raw text: repo, issue_number, kind, author, created_at, title, comment_id, message2) Preprocessing: clean_text, tokens, lemmas; plus heuristic flags (flag_long, flag_log, flag_code, flag_descartar)3) Sentiment (model outputs): sentiment_label (negative/neutral/positive), sentiment_score, sentiment_numeric (-1/0/1) Distribution: neutral (382), negative (236), positive (59)4) Emotion (model outputs): top_emotion, top_emotion_score, emotions_raw Top emotions (counts): neutral (419), gratitude (67), curiosity (55), confusion (43), approval (31), disapproval (22), ...5) Empathetic rewriting: rewrite_strategy, reformulated_text Strategy distribution: polish_neutral (455), enhance_gratitude (97), clarify_question (70), mitigate_negative (31), clarify_problem (24)6) Automatic evaluation: rougeL_f1, ppl_original, ppl_reformulated Valid rows: ROUGE-L F1 (n=671), PPL original (n=667), PPL reformulated (n=668) Ethics & privacy:All content originates from public GitHub sources. The author field contains public GitHub handles; for stricter privacy requirements, consider releasing an anonymized variant (hash/salt) in a future version. License:Creative Commons Attribution 4.0 International (CC BY 4.0). Documentation:This record includes README.md, CODEBOOK.md, DATASHEET.md, CITATION.cff, CHANGELOG.md, and checksums.txt (SHA256) for integrity.