@TechReport{Domokos2005, author = {Domokos, Csaba and Széll, Erika and Karpati, Peter and Böszörmenyi, Laszlo}, institution = {Institute of Information Technology ({ITEC}), Klagenfurt University}, title = {Mixed and Weighted Measures for Client Behavior Prediction in a Proactive Video Server}, year = {2005}, address = {Klagenfurt, Austria}, month = {jan}, number = {TR/ITEC/05/2.09}, abstract = {The precision of the predictors used in the ADMS[1] can be determined by similarity. There are already such measures[2] given, but we do not know exactly what efficiency they have and how well they show the difference between two lists. Kendall’s tau • Spearman’s footrule • Ulam’s distance We examined the characteristics of these similarity measures and developed some more measures that fit better our needs. One of the main goals is to consider the similarity more important at the begin of list, than at the end of list. Because the clients at the begin of the list probably will request more videos. During our work we defined 20 special ordered lists with 17 elements each. We tested the different measures on these lists. We also tested the Kemeny distance, which was defined in paper[3]. We modified the Spearman’s footrule and the Ulam’s distance according to the goal defined above (the top of the list considerate with higher weight (Weighted Spearman’s footrule, Weighted Ulam’s distance). Using the already known measures we developed a more complex, mixed measure, which uses more factors when defining the similarity. Finally we compared the 7 different measures using the artificially defined lists. With using the similarity measures we can tell how good the predictors[2] work in ADMS project. We could order the predictors by goodness, testing them on a real database (the World Cup ’98 Website’s access log).}, language = {EN}, pages = {40} }