Research And Implementation Of Knowledge Graph-based Machine Learning Method Recommendation Model

With the rapid evolution of the machine learning field,the number of machine learning methods is continuously growing at an increasingly rapid rate.Although the massive amount of machine learning methods provides rich resources,it also brings a serious information overload problem,which puts a heavy retrieval pressure on the learning and usage of researchers related to the machine learning field.Therefore,how to efficiently recommend suitable machine learning methods according to the practical task scenarios of researchers has become an urgent problem to be solved at present.Research has shown that the rich entities and their relationships in knowledge graphs can be utilized as auxiliary information to improve the accuracy of recommender systems.However,introducing knowledge graphs into the machine learning method recommendation scenario still has the following problems.(1)There is a lack of knowledge graphs in the field of machine learning that contain the main entities such as datasets and methods and the relationships among entities in the field.(2)Traditional knowledge graph-based recommendation models tend to focus on structural information while ignoring important descriptive information.(3)Existing knowledge graphs fail to distinguish descriptive attributes and structural connections in relationships,nor do they exploit the supervisory role between multiple information.To address the above challenges,this paper conducts the following in-depth research and implementation for the knowledge graph-based machine learning method recommendation.1.A machine learning domain knowledge graph MLKG is constructed,which defines the main entities in the machine learning field,contains the relevant attributes of the entities and the rich connection relationships between them,and can provide auxiliary information and an interpretable basis for machine learning method recommendation.2.We propose a description-enhanced knowledge graph-based model DEKR for machine learning method recommendation,which considers the problems that machine learning datasets and methods often have condensed names,lack specific explanations as well as suboptimal recommendations occurring when entity links are sparse,so that the description information related to core entities is introduced into the machine learning knowledge graph,overcoming the limitation that traditional knowledge graph-based recommendation models mainly rely on structural information and ignore description information.3.We propose a machine learning method recommendation model KGCL with knowledge graph-based comparative learning,based on point 2 above,which addresses the limitation that the description information in DEKR is only textual information,expands the scope of description features in the knowledge graph by distinguishing the relationships in the knowledge graph into description attributes and structural connections,it also conducts contrastive learning between the representations obtained from the description view and the structural view of central entities,so as to obtain a more generalized representation.4.A prototype system of machine learning method recommendation platform is implemented,which can recommend suitable machine learning methods based on the machine learning datasets selected by users or input by themselves,and display the relevant attribute information of the methods.Through the above research,this thesis have fully explored the entities and relationships among entities in the field of machine learning,and utilized the constructed machine learning knowledge graphs as auxiliary information to generate accurate and interpretable recommendation results,which has effectively solved the information overload problem in the field of machine learning and significantly improved the retrieval efficiency.