A Simulative Question Recommendation System For Driving Schools Based On Learner Clustering

With the publication of The Guidance of the State Council on Actively Promoting the "Internet +" Action,China’s internet education industry has entered a fast track of development,and the new mode of education has become a powerful supplement to the traditional mode of education.In recent years,with the accumulation of network user information and educational resource information,according to the characteristics of different learners,how to effectively improve learners’ mastery of knowledge,recommend questions to learners personalized,and improve the quality of teaching is a worthy topic of "internet + education".At present,the simulation question system of domestic driving school for subject I provides learners with simulation test methods such as sequential practice and random practice.The tactics of question sea can achieve certain training effect,but the target is not strong,the learner cannot carry on the target test exercise according to their own knowledge weak spot.To solve the above problems,this paper does the following work:(1)Obtain test data of car subject I from the website of domestic driving school.After data preprocessing,the characteristics of "test score" and "test time" of learners are retained.According to these two characteristics,different clustering algorithms are used to cluster learners,and the clustering effect of each clustering algorithm on samples is compared,from which the clustering algorithm with excellent clustering indexes is selected as the early-stage data processing algorithm of the late-stage recommendation strategy.Experimental data show that K-means and FCM algorithms have good clustering effect on samples.(2)Mark the difficulty degree,trap degree,test point and other information of each question,calculate the weight of test points and the learner’s error rate,and finally get the learner’s weakness.According to the weakness of learner test points,the clustering algorithm is combined with recommendation strategies(content-based recommendation and collaborative filtering recommendation)to form a hybrid recommendation algorithm to recommend simulated test questions to learners.The experiment shows that compared with the single recommendation strategy,the mixed recommendation strategy is superior to the single recommendation strategy in terms of accuracy,recall rate and coverage rate for sample sets after clustering.(3)According to the learner clustering based mixed recommendation algorithm for questions,a driving school simulation question recommendation system is designed to realize the rational application of mixed recommendation algorithm in personalized question recommendation.