Social robot detection method and apparatus

Abstract

The application provides a social robot detection method and device, which comprises the following steps: inputting the acquired multi-source data of a user to be detected into a first detection model, so that the model jointly models the data and outputs a preliminary detection result; based on a two-stage confidence evaluation mechanism, evaluating the confidence of the preliminary detection result, and determining a user whose predicted uncertainty meets a first preset condition and whose modal evidence strength does not meet a second preset condition as a low-confidence sample; if the user to be detected is a low-confidence sample, inputting the multi-source data into a multi-modal large language model, so that the model reviews to obtain a review detection result, and when it is determined that the result meets a credibility condition, outputting the review detection result as a target detection result; if the user to be detected is a non-low-confidence sample, outputting the preliminary detection result as a target detection result. The application can realize hierarchical processing of different difficulty samples and improve detection performance and computing efficiency.

Images