User Behavior Prediction Research Based On Log Analysis In Educational Data

Recently,most websites predict user behavior based on user static attributes,dynamic behaviors and other data.However,the relations are rarely considered between user dynamic behaviors.The design of business process determines user behaviors in business information system.Therefore,this paper collects user behavior log data with the business information system in the education field,and then conducts three aspects:(1)mining user behavior relations from log data in business information system;(2)studying brain cortical learning model BRHTM that includes user behavior relationships;(3)using BRHTM to predict the behavior of users about visiting the regional education statistics platform.The specific research is as follows:Firstly,we use the method of process mining to build complete workflow network by applying log data collected from information system.Due to the two-loop problem is existed in the initial log order relationship and existing improvements,this paper puts forward a new log order relationship to solve the current problems so that forms a relatively complete workflow network.Then the experimental results are compared with five quality dimensions,such as fitness,simplicity,precision,generalization and behavioral appropriateness.The degree of simplicity,precision and generalization is boosted in experiment.The results indicate that the log order relationship defined in this paper is valid.Subsequently,the relations of causality and parallelism are obtained based on complete workflow network,and the correlation analysis is carried out with the number of user behaviors,the number of previous operation and subsequent operation.The results show that there is a strong correlation between five features.At the end of this section,the user behavior relations are mined with user behavior category labels by clustering algorithm.Secondly,this paper designs a cortical learning model that integrates user behavior relationship,namely BRHTM.In the first place,sparse distribution representation is discussed,and processing units are formed based on it,which is a kind of exclusive artificial model of cell.The training data set of model consists of processing units.Afterwards,with the active status and predictive state of cell are recognized and the cell synapse value is updated,the model of predicting cell is formed.Then the prediction results of the model BRHTM are inspected.The relations of parallelism,causality and clustering are used to modify the prediction results in turn.Finally,the user operations data in UCI data sets is inputed and the predictions of one-step and multi-step are carried out about the model of HTM and BRHTM,and decision tree algorithm(C4.5)respectively.The final experiment shows that the effect of BRHTM model in single-step prediction is not obvious,and it has good effect and improves the accuracy in multi-step prediction.Finally,we use the brain cortical learning model BRHTM which combines the user behavior relations,design the implementation scheme and software system about user behavior prediction of the regional education statistics basic database platform.Meanwhile,the user behavior prediction module is set in platform.The BRHTM model is used to predict user behavior and gives hints to users.We collected the log data of thirteen districts in a certain area,and the processing units are formed by sparse distribution representation,then the BRHTM model is used to predict user behavior and the prediction results is revised.The prediction results are showed to users in the system,and ultimately improve the user operation efficiency.At the same time,the experimental results also show that users in certain regions have relevant business experience,which also have a great significance for the evaluation of regional education statistics.The availability of model is verified and solutions are provided for similar prediction problems.The model process is relatively complete,so that the problem of user behavior prediction is solved.