Context Prediction Based on Pattern Matching with Association Rules and Variable Weights

Abstract

In order to provide personalized services that use context with various properties and types in a mobile environment, a context prediction method(COPPAM) that can increase the prediction accuracy is presented in this paper. In COPPAM, each context property is relatively normalized to produce a pattern so that various types and properties can be dealt with. Also, association rules between the context properties are used to reflect the relevance between the context property to be predicted and other properties. Then, the weight of the prediction result(success/failure) is updated in real time, resulting in an improvement in prediction accuracy. When performing pattern matching for context prediction, the confidence and weight is used to calculate the prediction factor where the pattern with the largest prediction factor value is selected as the prediction pattern among the prediction candidate patterns with equal similarity. The evaluation results of COPPAM revealed that in the user location prediction using AILTB, the prediction accuracy average was 83.68%, which was an improvement over other methods by an average of 3.45%.