Cross-lingual Domain Knowledge Transfer And Classification Method Based On Deep Semantics

Emergency management of social crises in cyberspace is an important part of maintaining national security,social stability and long-term peace and security,while the large-scale,complex and multilingual nature of content in cyberspace poses a number of challenges for online content regulation,online opinion management and combating cyber security incidents.The transfer learning and classification of crosslingual domain knowledge through AI-related technologies can effectively transform and expand knowledge in small language domains,which helps to improve the risk identification capability of online content and plays a crucial role in preventing and resolving major risks.But the following problems still exist for transfer and classification of cross-lingual domain knowledge:(1)Emergency management domain knowledge has the characteristics of low resources and low density,leading to barriers to in-depth research on domain knowledge;(2)Multilingual knowledge has different grammatical structures,and the high cost of artificial learning and experts leads to processing of unfamiliar languages analysis is very difficult;(3)Most existing transfer methods for cross-lingual knowledge make use of information other than the relational structure in the knowledge graph,but in practical problems,other information is more difficult to obtain,which directly affects the accuracy of crosslingual knowledge transfer;(4)Knowledge itself is not restricted by language,and multilingualism can provide more supplementary information for domain knowledge,but this part of knowledge in the classification task.To address the above problems,this paper proposes an improved approach to cross-linguistic domain knowledge transfer and classification.(1)To address the problems of low resources and high costs of cross-linguistic domain knowledge,SHACUT,a cross-lingual knowledge unit transfer method based on semantic hierarchy modelling,is proposed.The model is divided into two parts:the knowledge model embeds monolingual knowledge maps into their respective polar coordinate systems,explicitly reflects the semantic hierarchy of different knowledge,and completes the vector representation of multilingual knowledge units.The transfer model aligns and migrates the knowledge units in the map by linearly transforming the different language vector spaces and calculating the confidence level with the help of an alignment seed library.Experiments of cross-lingual knowledge unit transfer are carried out on the open common language dataset and the domain small language dataset respectively.The experiments show that SHACUT can effectively migrate cross-lingual knowledge units not only on the open common language dataset,but also on the domain small language dataset.unit transfer.(2)In response to the wide variation in grammatical structure and high relevance of multilingual domain knowledge,GAIA,A graph structure network aggregated feature representation method that fuses initial features and attention mechanism is proposed.The model additionally fuses the initial features of nodes on top of GCN,while replaces the original symmetric normalized neighbourhood matrix with the neighbourhood attention matrix obtained from the self-attention mechanism,and aggregates the multi-layer feature representation to obtain the final node feature representation.By conducting experiments on several mainstream datasets for semisupervised node classification tasks,the experimental results of the GAIA model perform the art-of-state.It can be demonstrated that the GAIA model can effectively alleviate the over-smoothing problem while effectively fusing local and global neighbourhood features,extending the GCN to a deeper level,and the experimental results outperform the state-of-the-art methods,which can provide more accurate features for downstream tasks in graph structured data applications representation.(3)In response to the high difficulty and high labour cost of natural language processing in small languages,A multilingual domain knowledge graph classification method based on cross-lingual knowledge transfer is proposed,which constructs a multilingual knowledge graph with labels by using the SHACUT model to build a deep fusion with the densely annotated and information-rich domain knowledge graph and the sparsely annotated small language domain knowledge graph,complementing node features,inter-node relationships and node category labels at the micro level The new multilingual knowledge graph is then aggregated with neighbourhood features using the GAIA model to mine the neighbourhood multilingual node features and obtain a distributed representation of the multilingual knowledge graph,based on which the classification of the small language domain knowledge graph is achieved.Through experimental comparison,the effectiveness and advantages of the knowledge graph classification method based on cross-language knowledge transfer for small language domain proposed in this chapter are verified,which provides a solution to the problem of processing small language domain knowledge graphs with sparse or no annotation,and can help decision makers make better and faster determination of small language domain knowledge categories.Finally,based on the above method,a cross-language domain knowledge transfer and classification system is implemented,which analyses and processes the datasets uploaded by users,completes the knowledge transfer from the source language to the target language,realises the node classification of the target language knowledge graph,and finally presents it to users in a visual way.The system has been tested to meet users’ needs for cross-language domain knowledge transfer and classification tasks.