A pattern-based voting approach for concept discovery on the web

Automatically discovering concepts is not only a fundamental task in knowledge capturing and ontology engineering processes, but also a key step of many applications in information retrieval. For such a task, pattern-based approaches and statistics-based approaches are widely used, between which the former ones eventually turned out to be more precise. However, the effective patterns in such approaches are usually denned manually. It involves much time and human labor, and considers only a limited set of effective patterns. In our research, we accomplish automatically obtaining patterns through frequent sequence mining. A voting approach is then presented that can determine whether a sentence contains a concept and accurately identify it. Our algorithm includes three steps: pattern mining, pattern refining and concept discovery. In our experimental study, we use several traditional measures, precision, recall and F1 value, to evaluate the performance of our approach. The experimental results not only verify the validity of the approach, but also illustrate the relationship between performance and the parameters of the algorithm.