基于图卷积的科技成果分类分级无监督哈希检索研究

打开文本图片集

中图分类号：TP309 文献标识码：A 文章编号：1672-3791（2026）02-0231-05

Research on Unsupervised Hash Retrieval for Classification and Grading of Scientific and Technological Achievements Based on Graph Convolution

MA Fengxia

TianshuiProductivity PromotionCenter，Tianshui，Gansu Province，741OOo China

Abstract：Intheprocessofscientificand technologicalachievementretrieval，text-formscientificandtechnological achievement datais limited bytheabilityof natural language understanding，which makes the features extracted fromthe data not accurate enough，andthus leads to the decline of the accuracy of the retrieval results.In order to address this isse anunsupervised hash retrieval method for theclasificationand grading ofscientificand technologicalachievements basedongraph convolution is proposed.Byconverting thescientificand technological text data into vector expresson form，a soft clustering asignment matrix is applied to define the node membership of text vectors，soas toestablish thegraph structuredataof scientificandtechnologicalachievements.Basedonthis data form，a graphconvolutional neural network is introduced to extract data features under Fourier inversion.On this basis，through unsupervised learning，thecoresponding hash code is established for thedata features，andthen the hash code distance between the hash code of the target search term and the hash code of the scientific and technologicalachievementdata iscalculated togeneratethescientificandtechnologicalachievementranking searchresults ofclasificationand clasification.Theexperimental results show thatthe retrieval results of this method have high accuracy，and the Precision Recall（PR） curve is relatively good， with good practical application prospects.

Keywords： Information retrieval; Scientific and technological achievements； Classification search；Grading seGraph convolution; Unsupervised hashing

为了帮助科研人员快速定位并获取相关领域的核心成果，对科研成果信息进行精准检索十分关键。（剩余4555字）