Research And Implementation Of Data Labeling And Retrieval Platform For Scientific And Technological Consultation

As a emerging and high value-added industry,modern technology consulting relies on independent intellectual groups consisting of experts with their scientific and technological knowledge.Modern technology consulting provides intellectual services for organizations,government and enterprises to make their own operation decisions,becoming one of the fastest-growing industry nowadays.Currently,technology consulting services can be divided into two different types:providing directly consulting services by professional experts or providing database and retrieval platforms.Retrieval platforms mainly provide retrieval services for technology data(such as papers,patents,expert and enterprise information,etc.)and to help users make their own decisions.While directly consulting service will organize expert groups who have prior domain experience to provide services and finally provide their effective investigation report.Both types of technology consulting services are still difficult to meet the actual needs of technology consulting,facing the following challenges:1)Massive scientific and technological data is heterogeneous,makes it difficult to pre-label the data uniformly and effectively.Furthermore,performance of existing data labeling methods are too poor to meet the needs of labeling rapidly increasing scientific data.3)It’s difficult to automate the labeling process,and apply to a novel domain quickly.4)The retrieval method is inflexible and hard to use.According to the above problems and challenges,this paper aims to design and develop a novel scientific data pre-labeling and retrieval platform for technology consulting services.It provides highly automated and efficient labeling program and end-to-end retrieval services for massive technology consulting data belonging to different domains.This platform allows single customers or technology consulter to organize the data they need and finish their works quickly and efficiently.The whole paper includes the following three aspects:1)Propose and implement a technology data labeling model based on graph neural networks(GNN)to take the rich relationship information between technology resources into consideration,to improve the performance of data labeling model and support relational and joint queries for different scientific data.2)Propose and implement an automatic process for data labeling and retrieval,which can apply to various technology domains quickly.By deploying search engines and graph databases,our platform can easily construct the training dataset and train the labeling model.After that,trained GNN model is able to tag the raw data under a specific label set,and write back to the search engine.At the same time,besides retrieving data through specific labels,the raw data are also modeled at the semantic level and embedded into vectors,which can be used by deep learning recommendation models to construct end-to-end and flexible technology resource retrieval services.3)Design and implement a technology resource labeling and retrieval platform for technology consulting services,which includes useful functions such as label set management,dataset construction,model training,data labeling,data retrieval and so on.This platform provides convenient services in the form of API interfaces and website to meet the flexible need of technology consulting.Finally,the platform is applied to the national key research and development project "Research and Resource Construction of Scientific and Technological Consulting Data Resource System",to provide the project with the ability to mass-produce models,which verifies the functional integrity of the platform,and verifies the effectiveness and practicality of the platform and methods.