| At present,with the continuous expansion of Internet application fields,a large amount of knowledge appears and spreads on the network.Due to the relative freedom of network speech,it is easy to cause the problem that the trustworthiness of knowledge is difficult to distinguish.In addition,the complex relationship between knowledge brings challenges to traditional knowledge management methods.Knowledge graph has emerged rapidly with its intuitive network structure.However,the traditional construction method is difficult to meet the requirements,and the automated construction mechanism inevitably causes entity or relationship errors,which has a negative impact on knowledge reasoning,knowledge application and other aspects.Therefore,in the face of the rapid growth of knowledge,effective evaluation methods of knowledge trustworthiness degree are urgently needed to help filter out correct and reliable knowledge.This thesis focuses on the network knowledge and knowledge graph,studies the knowledge trustworthiness evaluation method based on representation learning,and proposes the knowledge trustworthiness evaluation model based on representation learning.It deeply analyzes the factors that affect the knowledge trustworthiness and the characteristics of knowledge trustworthiness,and constructs the evaluation score system and calculation method.Finally,a knowledge trustworthiness evaluation system is designed and developed based on the evaluation model.The main research work of this thesis is as follows:(1)Representation learning is an important method for knowledge trustworthiness evaluation,but the existing knowledge representation learning methods have the shortcomings of single loss function and poor knowledge representation effect in the face of complex relations.To solve this problem,this thesis proposes a knowledge representation learning method based on background information and adaptive weight measure.This method can better represent the entity and relationship vectors,and introduce knowledge background information to further improve the effect of the model and achieve higher calculation accuracy.Finally,the experimental results in link prediction and complex relationship prediction tasks show that the knowledge representation learning proposed in this thesis has better performance,which provides support for the establishment of knowledge trustworthiness evaluation model based on representation learning.(2)A knowledge trustworthiness evaluation model based on representation learning method is constructed.At present,knowledge trustworthiness evaluation is mostly limited to the evaluation of knowledge in a certain domain,which has some limitations.This thesis analyzes the network knowledge from the source,establishes the corresponding evaluation index,on the basis of improving the evaluation ability of knowledge correctness,fuses external knowledge sources and other factors,constructs a comprehensive evaluation index of knowledge credibility including external and internal factors,designs an evaluation strategy to calculate the final score of the data,and divides the trustworthiness level according to the results.A more comprehensive,reasonable and effective knowledge trustworthiness evaluation model is formed.(3)A knowledge trustworthiness evaluation system based on representation learning is designed and developed.This thesis analyzes the functional and non-functional requirements of the trustworthiness evaluation system,designs the system process,structure and functional framework,designs the database model and table structure according to the functional requirements,selects the appropriate technical framework,builds a stable and safe development environment,and realizes the development of knowledge trustworthiness evaluation system based on representation learning.Finally,the system performance and each functional page were tested to ensure the normal operation of the system. |
