TRMA_supporting_data_release

This dataset contains 2,000 de-identified synthetic conversational recommendation dialogues derived from the Amazon Reviews 2023 Movies & TV subset. Dialogue length varies from 6 to 10 turns. Initial user turns were generated with DeepSeek-chat from de-identified preference-title contexts; assistant and length-extension turns are deterministic title-only templates grounded in the released target item. The dataset includes fixed train, validation, and test splits (1,200/300/500) and excludes source user identifiers, raw review text, exact timestamps, source code, API credentials, and model logs.