Analyzing and Predicting user Navigation Pattern from Weblogs

Abstract

v newlineABSTRACT newlineUser behavior analysis has been emerging research in recent years, newlineas it helps many end services to market their products. The service providers newlinealways prefer to study the intentions of the user and try to increase their newlinesales. For this purpose, the user behavior analysis models are developed, newlinewhich extract the behavior of the user by analyzing the weblog files. In this newlineresearch, user behavior analysis and web page recommendation models are newlinedeveloped by analyzing the weblog history of the user. Here, the user newlinebehavior analysis is done by developing a prediction model. The proposed newlinealgorithms predict the user navigation pattern, by converting the user log files newlineinto sequential patterns and predict the performance from the same. As the newlinesecond contribution, (Weighted Support and Bhattacharya distance) WS-BDbased newlineTwo-level match is proposed for predicting the webpage based on the newlineuser behavior. Through the weighted support approach, the interesting newlinesequence pattern is extracted from the sequence obtained based on a prefix newlinespan algorithm. The web sequence patterns are generated using a newlinetransaction database. The interested web sequence patterns that are applied newlineto Bayesian Fuzzy Clustering (BFC) uses the maximum value of the newlinesimilarities existing between the individual sequence patterns and the newlinecentroids for clustering the patterns. Then, the two-level match predicted the newlinenext query web page from the formed clusters, and Bhattacharya distance is newlineused for calculating the two-level match. Thus, the best web page is chosen newlinefrom the sequence pattern with improved Bhattacharya distance. newlineHere, Dice Similarity-based Bayesian Fuzzy Clustering is proposed for newlineclustering the log database, and the weighted support is created for newlinepredicting webpage depending on the user behavior. The results are newlineevaluated based on precision, recall, and f-measure, and the data log for the newlinesimulation is considered from five different databases. At last, the supremacy newlineof the presented approach is proved over other existing works concerning the newlineprecision, recall, and f-measure. The modified span algorithm has achieved newlinethe values as 0.7642, 0.7565, and0.8645, as f-measure, recall, and precision. newlinevi newlineSimilarly, the incspan algorithm achieved the values of 0.8685, 0.7665, and newline0.7765 for precision, recall, and f-measure. The proposed WS-BD algorithm newlinehas the values as 0.7865, 0.8546, and 0.8784, and for f-measure, recall and newlineprecision respectively. The overall best performance is attained by the newlineproposed Dice Similarity-based Bayesian Fuzzy Clustering with the values of newline0.8415, 0.9, and 0.8945 as f-measure, recall, and precision. newline