Comparison of Proposed PPMM with Other PPM Methods for Link Completion Problem

Abstract

Prediction by Partial Matching (PPM) is a good statistical data compression technique based on conditional probability idea. Various researches have proposed methods for estimating the probability of novel entities, called zero frequency problem. However, the performance of each proposed PPM applied to the problem of link completion is not substantially compared. The problem of link completion is more complex than the problem of data compression. A symbol can be missed at any position in a given string. A new probability estimation called PPMM is introduced in this paper and the accuracy comparison of all previously proposed methods and our introduced probability estimation on the link completion problem is reported. In addition, the static as well as dynamic schemes with exclusion and inclusion approaches are involved in this comparison. The experiments are performed by using the data of co-authorship obtained from scientific publication DBLP. In comparison to other PPM methods, our proposed method is the best for link completion with the accuracy of more than 83%.