Research On Creative Portrait Construction Technology Based On Document Cognitive Understanding

As document information management and document knowledge analysis become increasingly important,some fields have urgent requirements for project-level semantic understanding.It's required to be able to accurately understand the current status of the research content and technical indicators in new application projects during the evaluation of research projects and accurately identify the capabilities of each research direction in the process of generating scientific information.This article focuses on the creative identification and weighting requirements of project concept documents,and conducts research on creative portrait construction technology based on document cognitive understanding.Based on the project documents,it studies the precise construction ability of creative portraits.Using natural language processing and document-level semantic understanding improve the ability to understand research content,technical indicator capabilities,and other content.This article adopts the idea of combining text structure traction and natural language processing.First,a document structure model is constructed for the project document.In document structure model,the key structural elements are identified around the project concept document assessment requirements,including research objectives,research content,and technology indicators,etc.,to transform unstructured project document data into a model.On this basis,around the creative portrait,firstly achieve the position of creative knowledge elements through subtitles and semantic paragraphs,and then candidate knowledge structure element sets are identified through paragraph-level and sentence-level textual structures on locked paragraphs.Paragraph-level natural language processing and feature recognition are used to verify candidate element sets to improve the accuracy of knowledge extraction.Combined with the paper's researches,a creative portrait service software was realized.The software was compared with the classic paragraph-level knowledge extraction algorithms——TextRank,LexRank and TF-IDF.The experimental results show that the average accuracy of the method is 6.39%higher than TextRank,5.59%higher than LexRank,and 17.08%higher than TF-IDF.