| There are three paradigms in scientific data research: experimental science,theoretical science,and computational science.Nowadays,a fourth paradigm that combines big data and artificial intelligence has emerged.Emerging technologies in computer science are widely used for data mining and analysis in some scientific fields.Scientific research data in the field of laser typically include information such as experimental materials,methods,laser parameters and experimental results.However,these data are usually found in scientific papers,manuals,files and laboratory records,and are difficult to obtain efficiently,which poses challenges to conducting scientific research and hinders its study rapidly.Efficient management and utilization of such scientific research data is the key to improving the experimental efficiency and accelerating laser research and development.In the management and use of the laser scientific research data,there are some following deficiencies: Firstly,the process of extracting important data from the laser literature by having experts read through it is timeconsuming and labor-intensive.Secondly,during the process of data transfer,the data quality may be compromised due to missing data,human errors,system errors,and other factors.Thirdly,scientific data on laser and optical materials experiments are missing in the publicly available laser database.Therefore,the main research content of this paper is as follows:(1)Laser Literature Knowledge Entities and Its Types Recognition.We use an unsupervised method to identify knowledge entities in laser literature.This method uses unlabeled laser text to pre-train the BERT-WWM model,and then combines Contrastive Learning and Clustering methods to construct a collection including representative words for Knowledge Entities.Then use the model to predict the nouns in the text,and finally calculate the Semantic Similarity between the predicted word and the representative words to judge whether the text noun is a knowledge entity.Our experiments show that the overall accuracy of 5 types of the laser Knowledge Entities recognition is 52.44%,and the optimal accuracy for a specific type of recognition is 67.12%.(2)Research on Data Quality Evaluation Method.Based on the six items specified in the Chinese national standards,we designed 12 secondary indicators according to the characteristics of the laser research data,and calculated the secondary indicators by developing the basic statistical indicators.Then,the weights for each indicator are allocated using the AHP,and form the data quality evaluation theory finally.We have designed the Quality Evaluation Module in the system to ensure the quality of the data in the system.(3)Laser Science Experiment Database Platform.We design three roles of the system administrator,the data administrator and the data user in the platform,and built modules such as the System Management,the Data Quality Management,the Laser Science Data Management and the Visualization.The platform also provides services such as storage,management,retrieval and download of different types of laser scientific data. |
