Regularized Distance Metric Learning and its Application to Knowledge Discovery  

Oleh:

Mikawa, Kenta ; Ishida, Takashi ; Goto, Masayuki

Jenis: Article from Proceeding
Dalam koleksi: The 14th Asia Pacific Industrial Engineering and Management Systems Conference (APIEMS), 3-6 December 2013 Cebu, Philippines, page 1-9.
Topik: Metric Learning; Regularization; Document Classification ; Vector Space Model
Fulltext: 1240.pdf (553.24KB)

Isi artikel

Due to the development of Information Technology, there are huge amount of text data posted on the Internet. In this study, we focus on the Metric Learning which is one of the fields of machine learning. Metric learning is a method estimating the metric matrix of Mahalanobis distance from the training data under the appropriate constraint. Mochihashi et al. proposed a method which can derive the optimal metric matrix analytically. However, document data is normally very high dimensional and sparse. Therefore, when this method is applied to document data directly, it may occur over-fitting because the number of estimated parameters is in proportion to square of the input data’s dimension. To avoid the problem of over-fitting, a regularizer term is introduced in this study. Therefore the purpose of this study is to formulate the regularized estimation of the metric matrix analytically. On the other hand, the metric matrix which is derived as the result of optimization problem expresses the relationship between variables, it is possible to grasp the relationship and the importance of each feature from the estimated metric matrix. In this study, in addition to show the way of deriving the optimal metric matrix, we demonstrate how to make use of the estimated metric matrix by using the Japanese newspaper article.

Opini Anda

Klik untuk menuliskan opini Anda tentang koleksi ini!

Kembali