Sparse fuzzy classification for profiling online users and relevant user-generated content

Extracting information and knowledge from users’ online activity is of great significance for a variety of practical purposes. Yet, existing research suffers from limitations including requiring prior knowledge and poor interpretability. In this study, we develop a novel classification algorithm based on the dual concepts of fuzzy set and sparsity regularization. Specifically, the proposed algorithm introduces two types of fuzzy sets designed to fuzzify samples, then the membership criteria from resultant fuzzy sets is further cast as the soft feature for training a sparse classifier. To demonstrate the practical benefits of this process and the performance of the proposed classification algorithm, we carefully examine its application to several benchmarking datasets, in addition to a unique real-world data resource containing 49,252 users worldwide and their 55,539 online historical records collected over a ten-month period. Experimental results demonstrate that the proposed algorithm outperforms other state-of-the-art methods by generating a more accurate classification. In turn, our findings can be utilized to shape industry efforts to identify users’ preferences and to personalize recommendation services.