Research On Data Mining Methods For Understanding Students Behavior

In recent years,with the rapid rise of educational data mining,combinng data min-ing methods to the student behavior data analysis has become a popular trend.It focuses on the discovery of future behaviors and interests,the prediction of students' learning performance,and the extraction of students' individual or group characteristics.Re-search on student behavior patterns has received widespread attention.Studies in many fields such as pedagogy and sociology have found that students' behavioral patterns have an objective response to their learning performance,emotional state,and mental health.How to model student behavior and how to describe the behavioral characteris-tics accurately and comprehensively are all important issues that need to be considered.With the improvement of the card system and the upgrade of cloud storage technology,a large number of behavior records have been collected,which provides powerful data support for us to portray students' individual behavior patterns.This paper mainly ex-plores the data mining methods of student behavior patterns,and applies the understand-ing of behavior patterns to the relevant prediction problems of student performance.The specific research content is summarized below.(1)Student behavior identification and behavior modeling.We standardize a series of processes for identifying student behavior sequences from structured storage records and extracting user behavior patterns The Markov model is currently one of the most widely used models for behavioral modeling.This paper supposes that students' daily behavior is consistent with the Markov character,and constructs HMM to capture the regularity of student behavior.The construction of student behavior patterns is the basis of the description of behavior characteristics It is also an important prerequisite for the establishment of student portrait systems.(2)Learning performance prediction based on campus behavior patterns.Learning performance is the most critical indicator for measuring a school's teaching level,and it is also one of the most concerned research points in the field of educational data mining.To some extent,behavior patterns reflect the students,psychological state,health sta-tus,and efforts to a certain extent.We propose a behavior-based learning performance prediction framework for students,and describe behaviors more comprehensively from multiple perspectives,including statistical features and associated features.And we use a multi-task model to make a fine-grained prediction of student performance in the course.Experiments have found that the proposed framework has a high recall,and it also shows some practicality for early warning.Further,we discussed the correlation between student behavior patterns and learning performance.(3)The prediction of mastery based on online learning behavior.The online learn-ing method frees the user's learning from the limitations of time and space,but also eas-ily makes users ambiguously determine their own mastery,blindly focus on what they want to learn,and result in a decrease in learning effectiveness.We propose a mastery prediction framework based on online learning behaviors,adding additional contextual information to the collaborative filtering algorithm including student-knowledge point mastery and lesson-knowledge information,and predicting students' mastery based on the learned learning paths.Taking the time-varying degree of mastery into account,we use Ebbinghaus Forgetting curve to approximate the student's mastery of knowl-edge points.Experiments have found that the addition of additional information has improved the forecasting effect,especially the running time.At the same time,the pro-poserd framework can not only dynamically assess students' knowledge of the knowl-edge,but also predict the results of the system is also easy to review feedback or adjust the learning sequence,to provide personalized learning services.