In the Persian encyclopedia, one of the necessary capabilities is to recognize contradictory sentences in each text in order to provide appropriate services to users. So far, few research works have been done inside the country to detect contradictions in Persian texts, which do not have enough accuracy. This research investigates and evaluates a new method for detecting contradictory sentences in a Persian text using the RWKV neural network. The main goal of this research is to detect contradictory sentences in a Persian text using data from FarsTail dataset with high accuracy. Based on the research method, the appropriate neural network RWKV was selected as a deep learning model. Next, fine-tuning the mentioned model for the task of detecting contradictions in the text, loading the model and data, pre-processing the data, fine-tuning the model with the pre-processed data and then evaluating the model was done. It should be mentioned that the data was divided into three categories: inferential, contradictory and neutral, and after pre-processing, the model was trained and used to recognize contradictory sentences in a Persian text. The evaluation of the model was done by using the confusion matrix and calculating various criteria including accuracy, positive prediction accuracy, readability and F1 score. The evaluation results showed that the RWKV model with an accuracy of 95.06% and an F1 score equal to 92.03% has a high ability to identify contradictory sentences in a text, compared to previous similar works. Comparing the performance of the RWKV model with other models such as mBERT, ESIM, HBMP and DecompAtt indicates the superiority
)