Research On Intelligent Generation Of Page And Application On WeChat Applet

With the development of front-end technology and the emphasis on product experience,front-end development work has become more and more important.Most of the front-end development work is to build and modify the page layout.The time cost and labor cost will be a lot of waste.Therefore,the technology of quickly building pages and generating code will greatly speed up this process and save a lot of costs.After analyzing the shortcomings in the front-end development process,based on the existing development of front-end technology and recent deep learning research results,this thesis analyzes the difficulties of intelligent generation of page code,that is,the processing of page screenshots and code,and proposes a encoder-decoder model to solve the problems of quickly generating complex layout pages.It uses a multi-scale feature extraction method based on attention mechanisms to generate vector representations of page screenshots.The corresponding description texts are encoded using word embeddings and LSTM networks,and then uses the decoder to process their connection vectors and predict the next code words.In addition to the research and optimization of algorithms,based on the development experience,this thesis also analyzes the existing page layout,and proposes a new dataset production method.By using this method,20,000 page screenshots composed of 27 layout and label words and description file were generated.This dataset is named Page2code.The page layout of this dataset is more complicated,which reflects the real structure of most web pages in the Internet to a certain extent,and can better verify the effectiveness of the algorithm.After sufficient experimental comparison on the public data set and the homemade data set,the evaluation results of the model proposed in this paper are 0.33 and 0.41 higher than the existing pix2code research respectively.On some samples different from the training set,the generation results are also more accurate than the pix2code model.These results verify the feasibility of this model on the problem of intelligent page generation,and show that it can effectively generate page code with complex layouts.Based on this model,this thesis also designs and implements a page intelligent generation system.Through the multi-module communication among front-end,back-end,model and database,the WeChat applet page code is generated,which verifies that it has high practical value.