Adaptive Feature-Weighted Co-Clustering With Local Coordinate Coding

Co-clustering enables the simultaneous clustering of features and samples by exploiting their associations. Based on the commonly used matrix tri-factorization objective function in co-clustering, we propose an adaptive feature-weighted co clustering with local coordinate coding (AFC-LCC) model in this paper, by making two improvements to enhance the data clustering performance. On one hand, a local coordinate constraint is introduced to enforce the smoothness of the sample cluster indicators along the scaled feature cluster spaces; on the other hand, a quantitative measurement is incorporated to adaptively learn the feature contributions in co-clustering for model discriminative ability enhancement. By co-optimizing the involved variables in the AFC-LCC model objective function, the experimental results not only show competitive clustering performance in comparison with related clustering models, but also depict the rationality and effectiveness of the local coordinate constraint and the feature importance descriptor.