Design of a Financial Automatic Question Answering Engine Based on Combination of BERT and Seq2Seq

By comparing and analyzing the application effects of different automatic question answering models in the financial field, this study explores effective ways to improve the performance of financial automatic question answering systems. This paper uses bidirectional encoder representation from transformers (BERT) to deeply encode user questions and candidate answers, extract semantic features, and then construct a Transformer-based sequence-to-sequence (Seq2Seq) generation model. Taking the BERT encoding results as input, it generates natural language answers closely related to the questions. By calculating the performance of each modelon key indicators, the study found that the BERTa nd Seq2Seq fusion model (BSFM) demonstrated excellent performance on multiple datasets,especially in generating coherent and realistic answers in the financial field. This discovery not only validates the effectiveness of the BSFM model in financial automatic question answering tasks but also provides important reference for subsequent model optimization and fusion. Meanwhile, the study also found that BSFM exhibits good adaptability in certain specific scenarios. This comparative result reveals the advantages and limitations of different models in handling financial question answering tasks, providing useful guidance for future model selection and customized development.