Research On Educational Resource Recommendation Method Based On Knowledge Graph

The development of educational informatization has transformed the learners'learning style from traditional classroom learning to online learning.With the development of big data in education,educational resources are characterized by huge amount,information overload and uneven quality.Under this network environment,learners are faced with the problems of "information overload" and "knowledge trek".Personalized recommendation system is an effective way to solve the problem of "information overload".The recommendation system excavates the potential interest preferences of learners according to the learners' historical behavior,calculates the similarity between learners and resources,and recommends the educational resources that learners may be interested in to learners.In the current research on Personalized Educational Resource Recommendation,researchers focus on learner feature modeling and resource attribute modeling,and recommend by calculating the similarity between learners and resources.Such recommendation algorithms tend to recommend resources similar to learners' preferences,without considering the logical relation between educational resources.The learning process of learners is gradual and learning knowledge is from shallow to deep.There are rich logical relations among knowledge points contained in educational resources.Therefore,it is necessary to consider the relationship between knowledge points contained in resources for personalized resource recommendation.Based on the above considerations,this paper constructs a small-scale knowledge graph in the filed of machine learning,and proposes a educational resource recommendation algorithm based on knowledge graph,which integrates knowledge connect degree and learner interest.The main work of this paper is as follows:(1)Using python web crawler to crawl all the book information labeled "machine learning" on Douban book and the related interactive behavior data as the dataset of this paper.(2)Constructing a small-scale knowledge graph in machine learning domain using the Chinese catalog text data in the dataset.In the process of building knowledge graph,we firstly use the method of word frequency analysis and the neural network model(BILSTM+CRF)combined with manual screening to complete the entity recognition of knowledge-side entity and knowledge-point entity.Secondly,according to the hierarchical relationship of catalog text,we use template-based method to extract the relationship between entities.Finally,we use neo4i graph database to store the acquired entities and relation information and complete the construction of knowledge graph.(3)Based on the constructed knowledge graph,we propose a collaborative filtering recommendation algorithm that integrates learners,interest and knowledge connect degree.In the model design part knowledge connect degree and learner interest are defined in detail,and a detailed recommendation process is given.Aiming at "cold start" problem of users in the proposed recommendation algorithm,we propose a recommendation strategy for learning beginners(new users).(4)In order to verify the effectiveness of the proposed recommendation algorithm,we use offline experiments to evaluate the recommendation algorithm.The experimental results show that,compared with the traditional userCF and itemCF,the proposed algorithm has improved the accuracy,recall,Fl-score and other indicators.The innovations of this paper are:(1)A small-scale knowledge graph in the field of machine learning is constructed.Knowledge graph can be used to show learners the relationship between knowledge-points and related learning paths,to alleviate the problem of "knowledge trek";(2)A collaborative filtering recommendation algorithm integrating learners' interest and knowledge connect degree is proposed.This algorithm not only considers the interest of learners,but also the cognitive level of learners and the logical relationship between knowledge points.The results of recommendation are more in line with learners' learning process needs;(3)The proposed recommendation method is not limited to the interaction data of new resources,which can alleviate the problem of ncold start" of resources in recommendation system to a certain extent.