Abstract: Incremental streaming data clustering is an important research consideration in social media applications and performance of clustering on the streaming data is largely depends on the data representation quality with respect to clustering effectiveness and data efficiency. Machine learning technique is a employed as traditional approach to streaming data clustering but faces huge complications due to increasing evolution of data contents. In order to tackle those issues, a multi-view convolution neural network has been presented in this paper. The Proposed model uses convolution layer for feature reduction and extraction. Pooling layer for feature selection which select the feature with maximum weight. Selected feature is projected to fully connected layer. Fully connected layer maps the feature into generated clusters on basis of objective function with maximum margin cluster. Those cluster further fine tuned to modify the hyper parameters of various layers in the convolution neural network to maintain the classification error on managing the ReLu activation layer in specified limit. Softmax layer minimizes the feature variance and cluster feature seperability in the feature space. Hyper parametric tuning is carried out in the output layer to make the data instance in the cluster to be close to each other by determining the similarity of the data instances on cluster representation. It results significantly enhancement in the clustering performance using the discriminative information’s. Detailed experiments of the current model have been evaluated against state of art techniques using facebook datasets. The implementation outcome of the multiview convolution neural network for data learning architecture represent the high accuracy and efficiency on clustering the streaming data.

Deep Learning, Streaming data, Deep Clustering, Convolution Neural Network Unstructured Data