A Cross-Modal Sequence-to-Sequence Generation Method Based on Topic Awareness

Abstract

The invention discloses a cross-modal sequence-to-sequence generation method based on topic perception, including: 1. Using a bidirectional long-short-term memory network to learn the contextual semantic representation of each record in a data table, and obtain the hidden vector representation sequence of the data record 2. According to the text corresponding to the title of the data table and the data table, learn the word distribution corresponding to each topic and obtain the vector representation of the topic by weighting and summing the vector representations of the words; 3. Based on the data records obtained by the coding layer in step 1 The hidden vector represents the sequence and the topic representation obtained in step 2, and uses the LSTM structure based on the attention mechanism as a decoder to generate analytical text; 4. Build a loss function to optimize the model parameters in step 1-3; 5. Inference In the process, for a given data table, beam search is used to approximate the best text generation results. This method can enhance the thematic consistency between the data table and the generated text, and improve the quality of the generated text.

Description

technical field [0001] The invention relates to the field of data processing, in particular to a cross-modal sequence-to-sequence generation method based on topic perception. Background technique [0002] With the advent of the era of big data, various industries have gradually accumulated massive industry data. These data are closely related to the production management of human society and are the main objects of analysis and research in various fields. Among these industry data, structured data has become the most common form of data due to its simple format and easy recording and storage, such as company financial statements, equipment sensor records, etc. However, structured data is usually very domain-specific, and it is difficult for people who lack industry knowledge to understand the meaning behind its values ​​and indicators. Therefore, how to accurately and efficiently convey the semantic information contained in structured data is an important cross-modal gener...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem of topic consistency and numerical encoding existing in existing data-to-text generation tasks, the present invention provides a cross-modal sequence-to-sequence generation method based on topic awareness

Images