An Automatic Response System based on Multi-layer Perceptual Neural Network and Web Crawler

This study proposed a deep-learning based Chinese natural language processing to effectively combine traditional hard-coded judgment and crawler to predict user intention. This study uses Jieba to do word segmentation and TF-IDF for keyword statistics and feature extraction. A multi-layer perceptual neural network is then used to classify user intention. In order to improve accuracy, fixed judgments and crawlers are added to obtain the latest service news on the official website. Experiments were conducted on the training data set (questions) collected by Facebook Chat, compared with the Chinese classification models of the popular methods. 100 questions and answers were tested, the accuracy reached 80% aboveį This shows that our method is feasible for various applications.