Large language model illusion relieving method based on entity chain and prompt chain technology

Abstract

The invention provides a large language model illusion relieving method based on an entity chain and prompt chain technology. The method comprises the steps that 1, entities are extracted, and an entity set is constructed; step 2, entity linking and background knowledge acquisition; 3, key knowledge points are extracted; step 4, response generation; 5, performing evaluation and iterative prompt; and step 6, repeatedly executing the step 5 until a satisfactory result is achieved or an upper limit of repeated execution times and iteration times is reached. According to the method, a text prompting method from coarse granularity to fine granularity is provided, meanwhile, illusion caused by the fact that external knowledge cannot be obtained in the model is avoided, complex processes such as fine adjustment of a searcher are not needed, the method has good practicability and convenience, and illusion generation of the model can be effectively reduced.

Images