date: 2014/05/27 author: xiangnan kong , bokai cao , philip s. yu source: kdd’13
DESCRIPTION
Multi-Label Classification by Mining Label and Instance Correlations from Heterogeneous Information Networks. Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13 Advisor : Jia -ling Koh Speaker: Sheng- Chih Chu. Outline. Introduction - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/1.jpg)
Date: 2014/05/27Author: Xiangnan Kong, Bokai Cao, Philip S. YuSource: KDD’13Advisor: Jia-ling KohSpeaker: Sheng-Chih Chu
Multi-Label Classification by Mining Label and Instance Correlations from Heterogeneous Information Networks
![Page 2: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/2.jpg)
2
Outline• Introduction•Meta-path-base Correlation•PIPL Algorithm•Experiment•Conclusion
![Page 3: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/3.jpg)
3
Introduction•The label correlations are not given and can be to learn from moderate-sized data.•Use heterogeneous information networks to facilitate the multi-label classication process.
![Page 4: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/4.jpg)
4
Single-label Classification• Ex:Single-label Classification
d1 d2 d3Economy 1 0 0Art 0 1 0Polity 0 0 1
• Ex: Muti-Label Classification d1 d2 d3
Economy 1 1 0Art 0 1 1Polity 1 0 1
![Page 5: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/5.jpg)
5
•EX: Drug-Target Binding Prediction
Multi-label Classificantion
Instance
label
![Page 6: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/6.jpg)
6
•EX:
Heterogeneous Information Networks
![Page 7: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/7.jpg)
7
FrameworkMeta-path
ConstructureMeta-path- based Label and Instance Correlation
Training Initialization
Bootstrap
Model
Iterative Inference
Output
![Page 8: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/8.jpg)
8
Outline• Introduction•Meta-path-base Correlation•PIPL Algorithm•Experiment•Conclusion
![Page 9: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/9.jpg)
9
Label and Instance correlationLabel :• The same gene correlation • Share similar pathway• Inter-connected through PPI link
Instance:• Similar side effects• Chemical ontologies• Similar substructures (feature)
![Page 10: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/10.jpg)
10
Meta-path-base Correlation• Meta-path-base Label Correlation
• Meta-path-base Instance Correlation
![Page 11: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/11.jpg)
11
Outline• Introduction•Meta-path-base Correlation•PIPL Algorithm•Experiment•Conclusion
![Page 12: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/12.jpg)
12
PIPL Algorithm•Meta-path Constructure
![Page 13: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/13.jpg)
13
•Training Initialization
• Yi: each Instance has a label set.• Pj(i):link i-th label through
meta-Path jArray(2-dimention)
考慮本身之外 xi,跟 xi有關係之 label,跟xi有關係之 Instabces
![Page 14: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/14.jpg)
14
•Bootstrap & Iterative Inference
•μ: unlabeled instances
![Page 15: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/15.jpg)
15
Outline• Introduction•Meta-path-base Correlation•PIPL Algorithm•Experiment•Conclusion
![Page 16: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/16.jpg)
16
Experiment• Heterogeneous Information networks: 290K nodes , 720K edge(SLAP)• Gene-Disease Association Prediction: 1943 instances , 300 feature , 50 labels• Drug-Target Binding Prediction: 5651 instances,1500 feature, 50 labels• 5-fold cross validation
![Page 17: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/17.jpg)
17
Evaluation Metrics
• Micro-F1 ↑,Better• HammingLoss ↓,Better• SubsetLoss↓,Better
![Page 18: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/18.jpg)
18
![Page 19: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/19.jpg)
19
![Page 20: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/20.jpg)
20
Outline• Introduction•Meta-path-base Correlation•PIPL Algorithm•Experiment•Conclusion
![Page 21: Date: 2014/05/27 Author: Xiangnan Kong , Bokai Cao , Philip S. Yu Source: KDD’13](https://reader035.vdocuments.net/reader035/viewer/2022062323/56816252550346895dd29ba8/html5/thumbnails/21.jpg)
21
Conclusion• The Paper proposed to use heterogeneous information networks to facilitate the learning process of multi-label classication by mining label correlations and instance correlations from the network.• And propose a novel solution to multi-label classication, called PIPL by exploiting complex linkage information in heterogeneous information networks.