Research On Three Dimensional Accurate Portrait Of Cross Media Science And Technology Resources Based On Deep Learning

In recent years,with the rapid development of science and technology,the number of scientific research achievements continues to rise in an explosive growth trend,and nearly 10000 new academic papers are published every day.With the rise of mobile Internet,universities and related data service companies have opened up a large number of academic data,and the related scientific and technological information also tends to burst out,increasing the exposure of scientific and technological content.The above resources generally contain multimodal data,such as text and image types.There are huge differences in the data structure of resource information between different modes,which are usually displayed in unstructured form.This will lead us to get a lot of invalid information even if we put in energy.How to mine and analyze the core effective information from these massive scientific and technological resources is of great significance.The main work of this thesis includes the following aspects(1)In order to realize the semantic feature extraction of cross media science and technology resource information,the data acquisition scheme of science and technology resource information and the cross media semantic feature extraction algorithm based on deep learning are proposed.The distributed crawler technology and mass information storage technology are used to collect the data of scientific and technological resources information,and the deep network model is used to extract the semantic feature vector of scientific and technological resources text and image.(2)This thesis proposes a mining and discovery method of entity information and entity association relationship of scientific and technological resources,and proposes a scientific and technological entity extraction algorithm(BBLAC)based on the attention mechanism of Bert fusion of local features,which realizes the filtering of invalid information in scientific and technological resources information and the extraction of core effective information.The experimental results show that the results of this algorithm are better than the comparison algorithm.This thesis proposes algorithm(MDESJ)based on multi-dimensional mechanism,which can expand the similarity relationship between scientific and technological entities and complete the three-dimensional portrait of scientific and technological resources.(3)This thesis proposes a method to analyze and extract the cross media semantic association relationship of scientific and technological resource entities,and proposes a cross media semantic association algorithm(SSGACA)based on generating confrontation network and sharing semantic structure,Cross media retrieval technology is used to realize text to image retrieval,to display multimodal information in a more intuitive and efficient image way,and to supplement the content of multimedia resources of the three-dimensional portrait of scientific and technological resources data.(4)A three-dimensional accurate portrait system of cross media science and technology resources based on deep learning is designed and implemented.The system mainly includes the following modules:deep learning based cross media semantic feature extraction and expression module of science and technology resource entity,mining and discovery module of science and technology resource entity information and entity Association,cross media semantic association analysis and extraction of science and technology resource entity.It mainly realizes the following functions:data collection,three-dimensional portrait of science and technology resources,cross media science and technology resources retrieval and display.The system is tested and verified.