High dimensional data is a function of feature space inputs and noise. This makes it difficult to analyse and visualize the data, specially, in case of social network analysis where the aim is to analyse user’s behaviour. Heterogeneous contents (text, audio, video, images etc.) make it more difficult to model. However, it is known that the actual feature vectors lie on much lower dimensions which can be obtained through non-linear manifold learning techniques. In this paper, we propose iterative local tangent space alignment with adaptive neighbourhood to extract the true low dimensional representation of data by sequentially aligning the tangent spaces and thereby reducing the overall reconstruction error. As the sub-manifold regions become linear, its neighbourhood size increases which leads to more information fusion. Extensive experiments on both synthetic and real world dataset proves that proposed method outperforms existing non-linear dimensionality reduction technique in both low dimensional representation and classification.

Social Network Mining, Dimensionality reduction, Manifold learning, Local tangent space alignment, Adaptive neighbourhood

[1] S. Moro, P. Rita, and B. Vala, “Predicting social media performance metrics and evaluation of the impact on brand building: A data mining approach,” Journal of Business Research, vol. 69, no. 9, pp. 3341–3351, 2016.
[2] J. Surma and A. Furmanek, “Data mining in on-line social network for marketing response analysis,” in 2011 IEEE Third International Conference on Privacy, Security, Risk and Trust and 2011 IEEE Third International Conference on Social Computing, Oct 2011, pp. 537–540.
[3] L. del Carmen Contreras Chinchilla and K. A. R. Ferreira, “Analysis of the behavior of customers in the social networks using data mining techniques,” in 2016 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM), Aug 2016, pp. 623–625.
[4] Hussain, Sadiq, et al. "Big data and learning analytics model." International Journal of Computer Sciences and Engineering 6.7 (2018): 654-663.
[5] T. Jia, M. Jian, L. Wu, and Y. He, “Modular manifold ranking for image recommendation,” in 2018 IEEE Fourth International Conference on Multimedia Big Data (BigMM), Sep. 2018, pp. 1–5.
[6] W. Zhao, S. Tan, Z. Guan, B. Zhang, M. Gong, Z. Cao, and Q. Wang, “Learning to map social network users by unified manifold alignment on hypergraph,” IEEE Transactions on Neural Networks and Learning Systems, vol. 29, no. 12, pp. 5834–5846, Dec 2018.
[7] J. B. Tenenbaum, V. De Silva, and J. C. Langford, “A global geometric framework for nonlinear dimensionality reduction,” science, vol. 290, no. 5500, pp. 2319–2323, 2000
[8] L. K. Saul and S. T. Roweis, “An introduction to locally linear embedding.”
[9] M. Belkin and P. Niyogi, “Laplacian eigenmaps for dimensionality reduction and data representation,” Neural computation, vol. 15, no. 6, pp. 1373–1396, 2003.
[10] Z. Zhang and H. Zha, “Principal manifolds and nonlinear dimensionality reduction via tangent space alignment,” SIAM JOURNAL ON SCIENTIFIC COMPUTING, pp. 313–338, 2004.
[11] Vidya, G., and S. Omprakash. "Survey on Recent Researches on High Level Image Retrieval." International Journal of Computer Sciences and Engineering (2016): 72.
[12] C. Cortes and V. Vapnik, “Support-vector networks,” Machine learning, vol. 20, no. 3, pp. 273–297, 1995.
[13] P. Roy, S. Ghosh, S. Bhattacharya, and U. Pal, “Effects of degradations on deep neural network architectures,” arXiv preprint arXiv:1807.10108, 2018.
[14] H. Xiao, K. Rasul, and R. Vollgraf. (2017) Fashion-mnist: a novel image dataset for benchmarking machine learning algorithm.