An empirical study on using hidden markov model for search interface segmentation

This paper describes a hidden Markov model (HMM) based approach to perform search interface segmentation. Automatic processing of an interface is a must to access the invisible contents of deep Web. This entails automatic segmentation, i.e., the task of grouping related components of an interface together. While it is easy for a human to discern the logical relationships among interface components, machine processing of an interface is difficult. In this paper, we propose an approach to segmentation that leverages the probabilistic nature of the interface design process. The design process involves choosing components based on the underlying database query requirements, and organizing them into suitable patterns. We simulate this process by creating an "artificial designer" in the form of a 2-layered HMM. The learned HMM acquires the implicit design knowledge required for segmentation. We empirically study the effectiveness of the approach across several representative domains of deep Web. In terms of segmentation accuracy, the HMM-based approach outperforms an existing state-of-the-art approach by at least 10% in most cases. Furthermore, our cross-domain investigation shows that a single HMM trained on data having varied and frequent design patterns can accurately segment interfaces from multiple domains.