Research On Relation Graph Clustering Algorithm Based On Matrix Volume

Big data springs up and people make the corresponding countermeasures and achievesome progress when facing the opportunities and challenges of big data. We entered abig-data era. Big data, whether in the internet or in the internet for things, is a profound andcardinal diathesis. Then, how to find and mine the implicit and valuable informationeffectively, using the techniques to solve the practical problems, is an urgent agenda.Clustering analysis for multi-relational data is one of the important techniques ofmulti-relational data mining. It undertakes the heavy responsibilities in mining topologystructure, explicating function, identifying the class cluster model and predicting behavior etc.It has widely applied in some research areas such as the world-wide web, counter-terrorism,viral marketing, social networks and computational biology and so on. It can address someissues like identifying a terrorist organization, network topology, image segmentation,anticipating connotative function, community mining, gene prognosing etc.Firstly, I study some documents about clustering analysis, which enlighten me withmulti-relational data, so I fully understand that cluster analysis techniques elaborate amomentous part in our daily lives and know the related categories and features of clusteringanalysis for multi-relational data algorithms. I also learn the methods and techniques used byresearchers when they deal with issues of multi-relational data, and learn the new progressof the research at home and abroad. Meanwhile, I also further study on data mining, graphtheory, statistics and probability theory etc, and use the theory and applications intothe study in this article.The article pays attention to the high dimensional and related closely objects.The main innovative research work can be summarized as follows:Firstly, the relations include the links of the same type of objects and the links of thedifferent types of objects.Secondly, considering the confines of the methods whose yardstick is distance, theconcept of matrix volume is introduced and treated as a distance metric for the highdimensional and related closely objects. Then, the article describes three matrix volumeapproaches such as determinant method, parallelotope volume method and matrix volumemethod, which are deemed three yardsticks. And the paper makes a comprehensivecomparison though the advantages and disadvantages, the applications space and thecategory of problems etc. Thirdly, on the basis of the above, the article integrates matrix volume with theclustering algorithm based on relation graph.Finally, the article simulates the algorithm proposed and original algorithm and doessome experiments in the artificial dataset and real-world dataset. The results make clear thatthe algorithm this paper elucidated, just in the clustering results, is better than the algorithmbased on distances like Euclidean distance, Manhattan distance and so on.It exploits the idea of matrix volume to dispose the issue of multi-relational dataclustering in the high dimensional space. In the further, we can do some attempts bycombining the idea of matrix volume with Bayesian, Markov and some popular intelligentoptimization algorithms like PSO, Bees etc. Believe that matrix volume has a broaderapplication space.