A learning-to-rank method for information updating task

摘要

Our paper addresses the information updating task which is to determine the most appropriate location in an existing document to place a new piece of related information. We propose a new learning-to-rank method for the information updating task. The updating task is formalized as a learning-to-rank problem, and in training, a heuristic method of automatically assigning labels for training examples is proposed to exploit structural information of documents. With the proposed formulation, state-of-the-art learning-to-rank algorithms can be applied to the task. We deal with the problem of the lack of semantic information by incorporating semantic features derived from word clusters to further improve the performance of information updating. The proposed method is applied in updating Wikipedia biographical articles and Legal documents. Experimental results achieved on both Wikipedia biographical data set and Legal data set showed that our proposed learning-to-rank method with cluster-based features outperforms previously reported methods for information updating task.