Knowledge Understanding and Citation Ability Improvement Strategy Based on End-to-end GenIR Model

In order to improve the machine’s knowledge understanding and citation capabilities in complex and heterogeneous information environments and ensure accuracy and efficiency, this paper constructs and optimizes an end-to-end GenIR model. The model combines the advantages of generative AI technology and information retrieval technology, aiming to automatically extract, understand and cite relevant knowledge from large amounts of text data. The paper first collects a large amount of text data in related fields, then preprocesses the data, and then builds an end-to-end GenIR model. The model is then trained and the model parameters are optimized using the back propagation algorithm. During the training process, the attention mechanism is used to enhance the model’s ability to capture key information, and adversarial training is used to improve the robustness of the model. After training, the model is fine-tuned. Experimental results show that the GenIR model reaches a maximum MAP value of 99.9% and also has significant advantages in nDCG. The average inference time is about 29% faster than the BM25 model, demonstrating dual optimization in accuracy and efficiency. This study successfully improves the machine’s knowledge understanding and citation capabilities in complex and heterogeneous information environments by building and optimizing an end-to-end GenIR model. It not only achieves optimization in accuracy and efficiency, but also provides a new solution for the field of information retrieval.