Research On Specified Player Category Extraction In MMORPGs

Due to the similarity to real world,Massive Multiplayer Online Role Playing games,termed MMORPGs,are growing as the research focuses in data mining.As these games grow with scales,greater demands are placed on applying anti-cheating,personalized recommendation and user research.Then extracting specified player category emerges as a reliable and effective approach to mining the target community from an enormous set of users.Conventional community finding algorithms under supervised or semi-supervised information depends on global prior knowledges.In this paper,the whole category sharing the same gaming style can be extracted by part of it,instead of designating all the potential categories and their labeled members.In this way we avoid the influence of imperfect labels and unbalanced-class problem against learning.Based on the known members of target community,we propose an extraction model towards specified MMORPGs player category,including the following two sections:1.Representation of MMORPGs structured features.It promises quantification of events and vectorization while keeping the time series information.Introducing column clustering into feature selection,their behaviors can be represented in lower-dimensional space.2.Specified player category extraction algorithm accepts semi-supervised information during bisected k-means,reducing the searching space and the clustering process converges towards the specified category.Extraction results are cohesive under unsupervised objective,while the convergence is adjusted by supervised information at the same time.Finally we combine the state-of-the-art methods of MMORPGs structured behavior features and semi-supervised community mining algorithms.With plenty of experimental demonstration,we can conclude that our framework can outperform others on different categories.