A new weighted ensemble model-based method for text implication recognition

In the context of text entailment recognition, there exists a risk of getting stuck in local optima when using a single pretrained model. To address this issue and improve the accuracy of model recognition, this paper builds a text entailment recognition model based on the ensemble learning approach and proposes a weighted ensemble model-based method for text entailment recognition. For the task of text entailment recognition, multiple base pretrained models are trained, and a weighted ensemble model for text entailment recognition is constructed by applying the sparrow search algorithm to assign weights to these base models. This paper compares the performance of the base models, ensemble models, and weighted ensemble models in terms of model recognition accuracy, demonstrating the strong performance of the weighted ensemble model. Furthermore, the weighted ensemble model combines the advantages of pretrained models and ensemble learning, making the final model more capable of generalization. Compared with the basic model, the model recognition accuracy proposed in this paper is improved by 2-3%, and the model accuracy, recall rate and F1 value are also improved by about 3%.