Research On Concept Serialization Recommendation Algorithm In MOOC

In MOOCs,commonly speaking,curriculum designing,course selection and knowledge concept recommendation are the three major steps that systematically instruct users to learn.This paper focuses on the knowledge concept recommendation in MOOCs,which strives to recommend related topics to users to facilitate their online study.A course or video may contain one or more conceptions.Compared with course or video recommendation,fine-grained concept recommendation for users is more in line with user learning habits.Concept recommendation research has extensive research value and application prospects in the field of recommendation systems.Despite the existing efforts in some related areas such as course recommendation,there are still two main issues hindering them from being extent to accurately recommend concept to users that with limited information.First,the existing approaches only consider the historical behavior of users,but ignoring various kinds of auxiliary data(a.k.a.,side information in heterogeneous information network),which is also critical for better user embedding.Second,traditional recommendation models only consider the immediate user response on the recommended item,not explicitly taking into account the long-term interest.To tackle the two aforementioned issues,this paper proposes HAN-RL(Heterogeneous graph Attention Network – Reinforce Learning),a novel reinforced concept recommendation model in MOOCs with heterogeneous information networks learning.First,we formulate the concept recommendation in MOOCs as a reinforcement learning problem to offer personalized and dynamic knowledge concept label list to users/students.To consider more auxiliary information of users,we construct a heterogeneous information network among user,course,concept,and then exploit metapath and double-layer attention mechanism to aggregate information from neighbors for better representation.The addition of network structural features can more abundantly and deeply express the user’s learning pattern and long-term interest.Second,for the sequential recommendation,the user embedding is entered as status into reinforcement learning,and then a serialized recommendation result for the user is generated.Model is complete and self-consistent.Since the serialized recommendation of reinforcement learning,the users’ interaction with recommendation results contribute to the establishment and adjustment of the learning pattern and long-term interest.Finally,comprehensive experiments and analysis on a large-scale real-world dataset collected from Tsinghua University Xuetang X,one of the largest MOOC in China,verifies the feasibility and effectiveness of algorithm models,task processes and training strategies.Meanwhile,by comparing the results of the baseline model on various evaluation indicators and in-depth analysis of the recommended examples,the performance of the method proposed in this article in terms of efficiency,accuracy and personalization is explained.