Research On The Generation Method From Structured Data To Interpreted Text

In the era of big data,information continues to explode in forms such as table data,image data and text data,among which text data based on natural language is intuitive and easy to read,and is an effective source for human to obtain information.In recent years,with the increasing dimension and quantity of data,the realistic demand of automatically generating text summaries from the structured table data which is difficult to read directly to improve the efficiency of information acquisition becomes more and more prominent.In order to meet the demand of improving the efficiency of information acquisition,researchers in related fields have gradually designed various methods to convert tabular data into text data.Converting structured data into readable text requires three steps: content selection,content planning,and text generation.At present,researchers tend to use data-driven end-to-end deep learning methods,but content planning is not emphasized in the existing research.However,in scenes that need to focus on the order of physical appearance,such as game commentary,content planning is particularly important.On the other hand,it is necessary to construct a huge model to comprehensively consider the three links with the help of neural network,which will lead to a longer training time for the model.This paper aims to explore the current research status of structured data generation and interpretation text,and proposes an improved Transformer based architecture,with the addition of a content planning link using sorting learning.This paper introduces the process of the model in detail,including three stages:data preprocessing,model training and model application.In the model training stage,this paper models content planning through sorting learning,and uses data set ROTOWIRE to test the model effect.In addition,a data enhancement method is proposed to further improve the model performance.Finally,this paper observed the influence of content selection and content planning tasks on the model effect by modifying their weights.Multi-angle experiments show that the structured data generation and interpretation text scheme proposed in this paper is superior to the comparison method.