A Big Data Conversational Recommender Framework Based on LLMs and HRAG

This research quantitatively evaluates the quality of answers provided by the framework and the severity of hallucination problems through two indicators: accuracy and hallucination rate, and evaluates the different performance of HRAG and the rewritten prompt words for different tasks. The dataset is from production data of big data domain knowledge, system knowledge, and application metadata. And this experiment uses the Qwen-72B-Chat model as all LLMs in the conversational recommender process, and the Text2Vec model as the embedding model.