Research On Product Marketing Copywriting Generation Based On Transformer Improvement Model

With the rapid development of e-commerce and the improvement of people’s living standard,more and more people choose online shopping.On e-commerce platforms,consumers have no access to products.High-quality product marketing copy can not only show the culture and products of merchants,but also better attract consumers to buy.For e-commerce platforms such as JD.com and Taobao,which have a large amount of commodity data,creating commodity marketing copy by hand is a time-consuming and inefficient job.Therefore,it is necessary to study the automatic generation of commodity marketing copy.In the area of text generation,Transformer is better able to understand and process contextual information with high accuracy and generalization capabilities.Therefore,this paper takes Transformer model as the basic model to study the generation of commodity marketing copy.Secondly,this paper builds TP model by adding pointer network to Transformer model,so that the model can help accurately copy information while retaining the ability to generate new words through generator,so as to solve the problem of unknown words.Then,the self-attention layer of TP model and the feedforward fully connected layer are merged into a unified full attention layer,thus simplifying the model structure and building the AATP model of automatically generating commodity marketing copy.Finally,in this paper,the commodity title,attribute,picture text and reference commodity marketing copy data of Jingdong e-commerce platform "Discover good goods" column are used for experiments,and the ROUGE criterion is used as the main evaluation index of the model to generate commodity marketing copy.Then compared with Seq2Seq,pointer generation network,Transformer,TP and other models with attention mechanism added,the results show that the AATP model proposed in this paper is superior to other models.Transformer model,which integrates pointer network and full attention mechanism,not only improves the generation efficiency of commodity marketing copy,but also saves labor cost and meets the needs of rapid iteration.It also lays a theoretical foundation and reference significance for text generation in other tasks.