A Course Recommender System Of MOOC Based On Collaborative Filtering Algorithm With Improved Pearson Correlation Coefficient

Nowadays,more and more users choose to take an online course on massive open online courses(MOOC)platform.Compared with traditional courses,MOOC can help students ignore the time limit and the location limit to taking courses and it is convenient for users to take courses in a variety of fields.A course recommender system is an important part of MOOC,which recommends appropriate courses to users.The recommender system uses historical records to analyze a user's historical actions,mining user's potential preference and constructing the interest model for each user and recommend course resources for each user.Users' rating records can directly reflect users' preference.When the user-rating dataset is used for the course recommendation,the recommender system should deal with the following questions:1.In the MOOC system,there are a large number of courses in many different fields.It is difficult to recommend appropriate courses for users.The user rating dataset reveals users' potential preferences.It is significant to construct a course recommender system based on the user rating dataset.2.Data sparsity will cause a loss of precision.Users of MOOC usually take a small number of courses and only a few users tend to rate a course after finishing it.It will be difficult to find similar users for the target user based on the sparse rating records.3.After predicting items' rating based on similar users' history,some courses with high quality may not be located in the recommendation list.When a course with high quality has been given low rating during predicting,it is necessary to propose a method to optimize the rating for this course.The collaborative filtering algorithm is a widely-used recommendation algorithm,which constructs a user interest model based on the target user's historical records and the similar users'historical records to predict a user's preference.The Pearson correlation coefficient is a frequently-used method to calculate similarity.But it is a universal method that is not optimized for the MOOC recommender system.In order to meet the requirement of MOOC course recommendation and construct a course recommender system based on sparse rating records from MOOC users,this thesis constructs an improved Pearson correlation coefficient oriented to the user rating dataset of MOOC and construct a course recommender system based on collaborative filtering algorithm with improved Pearson correlation coefficient.The main works of this thesis are concluded as follows;1.In the user rating dataset of MOOC,there are some key users who have taken and rated a large number of courses.These users',enrollment records will cover the records of most of the common users.It will cause an error in the recommendation for common users when the system uses the historical records of key users to calculate the user similarity.Besides,some users rating records may fluctuate in a large range that will cause an error as well.In order to address the problems above,a method using extra multipliers to improve the Pearson correlation coefficient will be proposed to construct a collaborative filtering algorithm.Compared with the existing collaborative filtering method,this method improves the algorithm procedure of calculating the similarity of users.Meanwhile,aiming at improving the performance of recommendation,many experiments based on a realistic dataset are taken to determine the performance of different combinations of additive multipliers during the calculation of user similar-ity.The experimental results show that the improved Pearson correlation coefficient performs better in course recommendation.2.The improved Pearson correlation coefficient optimized the calculation of user simi-larity that improves the recommendation of courses.However,when the user rating dataset is excessively sparse,the recommendation based on user similarity is not precise enough.In order to address this problem,the thesis proposes a course recommender system based on mixed similarity and the multipliers of improved Pearson correlation coefficient.The mixed similarity is calculated to reveal the user's preference for item properties.Besides,in order to optimize the recommendation of high-quality courses,a rating correction module is used to adjust the predictive ratings of recommended courses.The experiments based on a realistic dataset compare the performance of this recommender system with the recommender system based the traditional Pearson cor-relation coefficient and the improved Pearson correlation coefficient,the results show that the recommender system performs better after combining mixed similarity,the multipliers of improved Pearson correlation coefficient,and rating correction module.