Based on the Gemini Large-scale Model, Enhanced Accuracy in Semantic Similarity Detection With the Ernie Model

Abstract

The rapid development of deep learning models has been a hallmark of recent years. As the world moves towards greater intelligence, there is an urgent need for advancements in technologies like question-answering systems, chatbots, and search and recommendation engines, especially within the service and e-commerce sectors. At the heart of these advancements lies the task of detecting semantic similarity in natural language processing. Against this backdrop, this paper examines the accuracy of semantic similarity detection across different deep learning models, focusing specifically on the LSTM, Transformer, and ERNIE models under identical hyperparameters and configuration settings. The study reveals a common challenge among these models in achieving effective data generalization on the LCQMC dataset. To address this, the paper introduces an innovative approach by combining the highest-performing ERNIE model with the Gemini large-scale model and employing data augmentation techniques to enhance accuracy. This strategy increased accuracy from 82% with the ERNIE model alone to 85%.