Intelligent Question Answering Design of Marketing Expert System Combining BERT and TF-IDF Algorithm

In order to improve the response precision of intelligent question answering function of marketing expert system, an intelligent question answering model combining BERT and TF-IDF algorithm is proposed. On the one hand, the model uses TF-IDF weighted statistical method for text retrieval, and selects the paragraphs and questions whose similarity is ranked first to fifth in the corpus to be spliced together as the output of the algorithm. On the other hand, BERT model is used to predict the probability of each word in the output paragraph, determine the beginning and end position of the answer, and then extract the answer to reply to the user's question. The results show that compared with model without TF-IDF algorithm, model without BERT, and popular neural network models such as CNN, BiLSTM, TextCNN and Transformer, the intelligent question answering matching results obtained by the proposed method have higher F1 value, EM score and average accuracy (AVERAGE), and they reach 98.77%, 94.92% and 96.74% respectively, which proves the effectiveness of the proposed method. In addition, the accuracy of the response results of the marketing expert system constructed by combining BERT and TF-IDF algorithm intelligent question answering model reaches 94% in the test of random 50 questions, and the average time taken to respond to each question is 0.52s, which meets the accuracy and immediacy requirements of the marketing expert system, has certain practical value, and is worthy of further research and promotion.