Method, device, device and storage medium for generating text based on tabular data

Abstract

The present invention provides a method, device, device and storage medium for generating text based on tabular data, including: constructing a data feature representation space, wherein the data feature representation space includes a recording coding layer, a content selection layer and a content planning layer, To achieve hierarchical feature representation; build a deep generative network by training the mapping relationship between feature representation space and output vector; build a generative model from tabular data to text; convert the input tabular data to be converted based on the generative model output as text. The invention integrates the data feature representation based on the hierarchical structure and the text generation technology of deep generation, can understand and reason about each sample data record in the entire table data, and can realize the conversion of structured data into global semantic coherent long Text, enriches the input data inference method and the diversity of output text expression, and effectively improves the performance evaluation indicators such as BLEU and PPL in the text generation model.

Description

technical field [0001] The present invention relates to the technical field of human-computer interaction, and in particular, to a method, apparatus, device and storage medium for generating text based on tabular data. Background technique [0002] Text generation technology is to use artificial intelligence algorithms to allow computers to complete writing tasks autonomously, without manual intervention during the writing process, and automatically write high-quality natural language texts. Text generation technology can not only be used for intelligent question answering and dialogue, machine translation and other systems, but also to achieve more intelligent and natural human-computer interaction; it can also be used to generate image descriptions in the field of visual images to speed up the combination of vision and natural language processing. [0003] Among them, data-to-text generation technology is the hotspot and difficult point of current research, and it shows im...

AI Technical Summary

Problems solved by technology

[0004] In the current research on deriving and generating descriptive text from tabular (structured) data (strings, numbers, etc.), in the reasoning of specific values, time and other information, there are still gaps between data on a time axis that cannot be effectively given. The defect of distribution law, and most of the descriptive text generation process separates semantics and syntax, making the text lack of logic and coherence, and still lacks diversity in text expression

Images