Research On The Improvement Of Students' Data Structure Performance Prediction By Baidu Library

With the development of information technology, education has undergone tremendous changes, and great amounts of education-related data has been generated. At the same time, the era of “big data” does come gradually with the prosperity of data mining techniques and information processing techniques. The application of data mining technology in the field of education is called educational data mining, which is called EDM. EDM is a cross subject, involving many fields including such as computer technology, education, statistics and so on. Achievement prediction is one of the classic application scenarios in EDM research. At present, the results of EDM mainly regarding the digital teaching software and intelligent tutoring system, which have strong pertinence and weak commonality, only can be used to analyze the specific system.This article employed data mining techniques to predict whether a student can pass Data Structure based on their access records on Baidu Wenku. Previous study had shown that it is possible to predict whether a student can pass Data Structure based on their access frequencies on general websites. This study first estimated the access time as well as access frequencies on Baidu Wenku. Then these features were added to the original data set to improve the prediction performance.By observing web logs, a state transition model is constructed to simulate Baidu Wenku accessing sequence. The model is then used to estimate the access time. Besides, the category of each document is obtained and used to for performance prediction. And the seven most relevant categories to student performance were determined.In order to increase prediction performance, these new features were gradually added to the model in certain orders. Combination of these features involved using new features or not using them, using coarse access time or accurate access time, and using key categories or non-key categories. It is found that the accurate access time is more effective than the coarse time, and the key categories is more effective than the non-key categories. The adoption of optional feature set achieves a specificity 70.59%, while the sensitivity is still greater than 80%.Finally, the present work is concluded and future work is pointed out. The results of this study shows that the way using network access logs and previous test scores to predict performance is feasible. The data set used in this study does not depend on some specific teaching system, therefore, it has a strong commonality, which overcomes the shortcomings of the current research on most performance prediction.