MoveMentor: Dataset of Chess Positions paired with Textual Explanations

MoveMentor: Dataset of Chess Positions paired with Textual Explanations Authors: Brandon Panos, Michael Graber, Can-Elian Barth, Melanie Völlmin FHNW University of Applied Sciences and Arts, Institute of Data Science, Windisch Project: Hasler Stiftung small project grant, Project ID: 2025-05-05-522 Description: Data for training a LLM with Leela Chess Zero embeddings to explain Chesspositions. Collected and built by FHNW Institute for Data Science Windisch. The chess comments of the dataset "chess_commentary_data.jsonl" are from Wikibooks (https://en.wikibooks.org/wiki/Chess_Opening_Theory) under the Creative Commons Attribution-ShareAlike License (https://creativecommons.org/licenses/by-sa/4.0/). The dataset "chess_question_answer_data.jsonl" is built by FHNW Institute for Data Science Windisch (using Stockfish 11 Eval Function Features and Leela Chess Zero predictions) and is under Creative Commons Attribution 4.0 International License. Together these datasets contain around 259'000 Chess postions with comments, questions and answers about the positions. File structure: chess_commentary_data.jsonl: san_seq: Complete SAN move sequence from the starting chess position to the position described in the comment. comment: Comment which describes the position chess_question_answer_data.jsonl: san_seq: Complete SAN move sequence from the starting chess position to the position addressed by the question. question: Question about the position answer: Answer to the question about the position You can find the code for training LLMs with this data and generation of more chess datasets here: https://github.com/can020202/movementor