Hierarchical mixtures of experts and the EM algorithm
Neural Computation
Adaptive mixtures of local experts
Neural Computation
Local linear perceptrons for classification
IEEE Transactions on Neural Networks
| Hi-index | 12.05 |
The Social Web holds significant value for consumer intelligence. Today, tremendous valuable information behind the social Web has not yet been fully utilized by firms to achieve competitive advantages. In this paper, we propose a machine learning approach to model latent, heterogeneous consumer tastes from social Web reviews. Our approach employs the Mixture of Experts (ME) method to derive a set of distinct consumer clusters, each of which has a cluster-conditional taste template regression model. Furthermore, our approach is able to predict the cluster membership as well as the overall response solely from a consumer's review ratings. We provide system architecture, model specification, and result analysis. We also compare the ME model with the Latent Regression Model. In practice, our ME learning approach will enable businesses to identify distinct consumer clusters directly from the social Web in order to satisfy consumers through customization and differentiation based on cluster-specific taste templates and cluster membership.