Logo image
Multi-Task Semi-Supervised Clustering via Feature Level Tags Embedding
Thesis

Multi-Task Semi-Supervised Clustering via Feature Level Tags Embedding

Yeh, Chia Hsin
Masters, 國立清華大學, 資訊工程學系
2016

Abstract

分群 多重任務 使用者標籤 clustering multi-task user tag
The modern solution to find personalized clustering is Semi-Supervised Clustering. However, Semi-Supervised Clustering usually has to face the problem of sampling bias. Our work aims to solve this problem by: 1. Use the user-given tags data on the data points as the side-information of semi-supervised clustering 2. Design a single model to embed the data points from feature space to tag space and cluster the data points in one phase 3. Use a high-mode tensor to transfer knowledge from user to user, and enhance the individual clustering performance Our experiments showed the superior performance and ability to transfer knowledge of our algorithm. Other contributions of this paper include: 1. Define a sparse regularization term on tensor 2. Design a CP-free update rule of our model

Metrics

1 Record Views

Details

Logo image