| Popis: |
In this paper, a graph-based feature selection method for learning to rank, called FS-SCPR, is proposed. FS-SCPR models feature relationships as a graph and selects a subset of features that have minimum redundancy with each other and have maximum relevance to the ranking problem. For minimizing redundancy, FS-SCPR abandons redundant features which are those being grouped into the same cluster. For maximizing relevance, FS-SCPR greedily collects from each cluster a representative feature which is with high relevance to the ranking problem. This paper utilizes FS-SCPR as a preprocessor for determining discriminative and useful features and employs Ranking SVM to derive a ranking model for document retrieval with the selected features. The proposed approach is evaluated using the LETOR datasets and found to perform competitively when being compared to another feature selection method, GAS-E. |
