Collaborative Manufacturing Enterprise Chart Question-answering Method Based on Multimodal Cross-fusion
WANG Xinxin
CHEN Liang
DONG Qianqian
Abstract：To aid collaborative manufacturing enterprises in extracting information from unstructured chart data,a chart ques-tion-answering method based on multimodal cross-fusion is proposed.Initially,text detection and recognition technologies are em-ployed to extract textual and numerical information from the charts,achieving the conversion of chart data to tabular data.Subse-quently,the queried chart,user question,and generated table are input into a multimodal cross-fusion model.This model com-bines image features with table-text features,providing a more comprehensive chart information for accurately answering user ques-tions.The method addresses challenges of traditional text recognition,improving accuracy to 91.58%.In the context of a real enter-prise chart dataset,the accuracy of chart question-answering reaches 82.24%.
Keywords：chart question-answeringmultimodal fusiondeep learningtext recognition
Publication Date：2025-09-20
Online Publishing Date：2025-12-23(First online date of this platform, not the publication date of the document)
Pages：6( 51-56 )