The Next Generation’s Personal File System Management

The current file systems are hierarchical, which can cause duplicate storage and cannot represent human’s mind map. In this paper, we explore the possibility of a heuristic, relational personal file system. Regarding each file as a node in the graph, we implement K-means, EM, LDA and Tree Bagging algorithms respectively to group the related files. In this way, we convert the current hierarchical file system to relational file system. We compare the results of these algorithms, the error of K-means, EM, LDA and random forest are 62.33%, 61.4%, 42.33% and 16% respectively. Among all the unsupervised learning, LDA a popular and generative model in topic modeling gives the best accuracy, but still does not surpass supervised learning. Therefore we propose to combine LDA and Tree Bagging algorithm, using the semi-supervised learning to classify the files in the future. In the end, we also discussed the potential of combining latent fator model with above methods to classify a large scale of file sets, thus extending our method from personal file system to corporate file system.