Data entropy: Artificial intelligence credible assessment

As large language models are increasingly applied in different fields of artificial intelligence, phenomena such as ''hallucinations’’ frequently occur in reasoning and problem-solving, especially influenced by dataset quality and implicit biases. To solve the credible problem of artificial intelligence,This paper proposes a method for identifying and quantifying emotional bias in text datasets through emotional bias analysis. By analyzing the uncertainty in the distribution of emotional categories in texts, the method can effectively reveal biases in datasets, providing insights for optimizing large model training. The research in this paper presents a novel approach for credible assessment in artificial intelligence, using an information-theoretic framework through data entropy calculation.