Diagnosis Of Learning Adaptability Based On Multi-relational Classification And Clustering

The adpative learning system supplies students with the suitable teaching contents according to the students's foundation and ability,it has become one of the most challenging and cross-subject research topic. Among these adaptability patterns,the diagnosis of learning adaptability is one of the most important steps. Therefore,we need to discuss the new pattern of the diagnosis,and estimate the students' ability and their comprehension of the knowledge, in this way,we can know the learning needs of different students and provide the students with the right teaching contents. This has a important significance to the e-learning system.Based on the research background and researches in domestic and on abroad, this dissertation applies multi-classification and multi-clusting technologies to mine the students's learing trace. Afterwards,we build the model of the diagnosis of learning adaptability. The main works of the dissertation are listed as follows:1. We make diagnosis based on two aspects. One is the students's learning ability,the other is the students's cognitive ability. We have build the diagnostic model and divide the diagnosing process. The model use logistic method to make diagnosis before learing,then use multi-clustering to set up the criterion of learning resources,thus we finish the diagnosis in the course of learning. Finally, we make use of multi-classification to update the diagnosis.2. Learning data is the imbalanced data set,We have studied the multi-relational classifier which is applied to this kind of data. We build sub-classifiers in combination with One-vs-All method and CrossMine algorithm,and all the sub-classifiers are validated by their AUC values; We compare the ECOC of the target class with the Hamming distance of the linked word produced by the sub-classifiers,and choose the class which has the shortest Hamming distance for the final result.3. There are a lot of information of properties and links in the database,we study the clusting method based on the information of properties and links.This method makes use of the user's guide to search the multi-relational properties which is tightly related to the dusting task. We generate a vector of multi-relational properties for each target tuple.Then we calculate the similarity from the property vetor and the link. It increases the accuracy of the clustering method and sets up the criterion of learning resources,meanwhile ,it provides the references to the diagnosis while learning.4. At last,we have designed the prototype system and the achieve the diagnostic model, and by analyzing and evaluating the effectiveness and performance of the system we have proved its legitimacy and usability.