Student Forum

Metadata is crucial for the administration and use of data. In this work, the use of recurrent neural network in deep convolutional neural network of generative adversarial networks is optimized to improve metadata quality, particularly for video anomaly detection. Conventional models frequently fall short in producing high-caliber synthetic metadata. Our recurrent neural network with deep convolutional neural network of generative adversarial networks method addresses class imbalance, increases the accuracy of anomaly identification, and enhances the production of synthetic metadata. In comparison to previous models, our model demonstrated higher AUC values and better syntactic image metadata quality generated using CUHK Avenue dataset but this is research work is in progress.

Learning from safety accidents and sharing safety knowledge has become an important part of accident prevention and improving construction safety management. It is difficult to reuse unstructured data in the construction industry, and there is an abundance of multi-modal resources represented by images or text, the knowledge of which is difficult to be used directly for safety analysis. In this paper, we propose a metadata-based security knowledge organization model from an intelligent construction perspective. Firstly, according to the safety management needs and construction information multi-source heterogeneous characteristics, the construction safety event knowledge connotation and characteristics are analyzed. Secondly, a metadata model for intelligent construction safety information is designed. The third, the construction image feature recognition method by fusing metadata is proposed. Finally, a multi-modal knowledge service path is explored, which provides methods for data sharing and fusion in intelligent construction management.

Author name ambiguity significantly complicates the retrieval and attribution of journal articles. To address this issue, this research proposes a synergistic approach that integrates a data-driven strategy with curated external knowledge bases. This approach aims to effectively disambiguate author names, thereby improving the reliability of metadata by ensuring that publications are correctly attributed to their true authors. Experimental results demonstrate the superiority of the proposed method over existing approaches in similar contexts, showcasing its potential to enhance the accuracy and efficiency of scholarly communication.

The distinction between the bibliographic records representation of musical works in MARC record-based Web Public Access Catalogs and data-centric Linked Data catalogue is examined in this preliminary study. How MARC and LD resource descriptions are used to support the PAC’s representation for musical works is also examined.

Chinese paper-cutting as a valued intangible cultural heritage (ICH) lacks comprehensive metadata descriptions. This study proposes a multi-dimensional metadata framework with six dimensions: basic attributes, connotative features, creators, ICH projects, inheritors, and derivative innovations. It incorporates elements from Dublin Core, CDWA Lite, and VRA Core for basic attributes, connotative features, and creators, and designs attributes for ICH projects and inheritors based on Chinese ICH project declarations. The derivative innovation dimension covers derived works. Usability of the proposed framework was verified using RDF metadata for "Flower Cutting Maiden". The study notes limitations in validation data and connotative feature depth, and puts forward semantic and linked data direction for the further future research.