Research On Personalized Learning Resource Recommendation Algorithm Based On Deep Learning

With the development of Internet technology,the field of online education has been rapidly developed.It has become more convenient for people to acquire the way of knowledge in learning.Online users can flexibly obtain learning resources on the online learning platform and study online courses.In the period of big data,with the number of online learning users the continuously increase.When students face the massive amount of learning resource data,they need to spend more time and energy to select the related course content.Moreover,the recommendation system can be one of the useful methods to deal with these "information overload" problem.It can provide student with personalized online learning resource course recommendation services.Although the recommendation algorithm has achieved good results in other fields,the application of online course resource recommendation system is relatively few.On the one hand,when using traditional recommendation algorithms to solve course recommendation,problems such as data sparseness and cold start still exist.On the other hand,due to lack of digging deep into learner and course characteristics,leading to the recommendation results need to be further improved.In response to these problems,this paper proposes an online learning resource course recommendation algorithm which based on the autoencoder neural network combination the self-attention mechanism encoder with the course relevance decoder.The specific research content are as follows:(1)Aiming at the sparseness of online education data in online course recommendation,it is not possible to better distinguish the importance of different courses on student preferences.And the problem of the in-depth features of students and courses cannot be effectively learned in the training process,which can result in unsatisfactory recommendation outcome.We propose the self-attention encoder method that utilizes the complex non-linear interaction data between the student and course.And then adds the self-attention coefficient to the course feature for mining the importance level of different historical courses on the student,further obtains the deep feature representation of the interactive relationship between the student and the course.It make the recommendation outcomes more accurate.(2)Aiming at the problems of cold start in online course recommendation,lack of interpretability of recommendation results,and ignoring the implicit features of courses for students to better accept recommendation results.We propose the course relevance decoder method that make the course description text information is converted into the word vector representation,and input it into the Siamese Long Short Term Memory neural network.Meanwhile it embeds the attention mechanism to capture the hierarchical relationship between courses.And uses the Manhattan distance function to calculate the course relevance.In this paper,we apply the proposed based on self-attention encoder and course relevance decoder on autoencoder algorithm to the online course recommendation problem.It not only model the preferences of student,but also combines the attributes feature of courses to obtain the list of course recommendation scores ranking,and provides with the personalized recommendation service.The experimental results show that the recommendation algorithm has the good recommendation performance.