Hierarchical Semi-Supervised Classification with Incomplete Class Hierarchies

This research explores the challenges and solutions in semi-supervised entity classification within incomplete class hierarchies. It addresses issues related to food, animals, vegetables, mammals, reptiles, and fruits, presenting an optimized divide-and-conquer strategy. The goal is to achieve semi-supervised classification and ontology extension in a unified framework, overcoming constraints within the classification model. Various optimization problems and methodologies are discussed, aiming to maximize data likelihood while tackling missing classes. This work builds upon prior studies in coupled semi-supervised learning and gloss finding for knowledge bases, offering insights into efficient hierarchical classification in complex datasets.

• Hierarchical Classification
• Semi-Supervised Learning
• Entity Classification
• Incomplete Class Hierarchies

westlake_c Follow

Uploaded on Sep 26, 2024 | 0 Views

Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

Presentation Transcript

1. Hierarchical Semi-supervised Classification with Incomplete Class Hierarchies Bhavana Dalvi, Aditya Mishra, William W. Cohen

2. Semi-supervised Entity Classification Everything Food Animals Vegetables Mammals Reptiles Fruits 2

3. Semi-supervised Entity Classification Everything Food Animals Subset Vegetables Mammals Reptiles Fruits Disjoint 3

4. Semi-supervised Entity Classification Everything Food Animals Vegetables Mammals Reptiles Fruits 4

5. Prior work Everything Coupled Semi-Supervised Learning for Information Extraction, Carlson et al. WSDM 2010 Automatic Gloss Finding for a Knowledge Base using Ontological Constraints, Dalvi et al. WSDM 2015 Food Animals Vegetables Mammals Reptiles Fruits 5

6. Challenge: Challenge: Incomplete Class Hierarchies Everything Food Animals Vegetables Mammals Reptiles Fruits 6

7. Challenge: Challenge: Incomplete Class Hierarchies Everything C9 Food Animals Location C8 Vegetables Mammals Reptiles Fruits Beverages 7

8. Challenge: Challenge: Incomplete Class Hierarchies Everything C9 Animals GOAL: Do semi-supervised classification and ontology extension in a single unified framework. Food Location C8 Vegetables Mammals Reptiles Fruits Beverages 8

9. Optimization Problem Maximize { Log Data Likelihood Model Penalty } m: #clusters, Params{C1 Cm} subject to, Class constraints: Zm Expectation Maximization 9

10. Optimized Divide-And-Conquer Strategy Class constraints: Mixed Integer Linear program Missing classes: Soft Divide-And-Conquer method 10

11. Class constraints: Class constraints: Mixed Integer Linear program Max {likelihood of assignment constraint violation penalty} 11

12. Class constraints: Class constraints: Mixed Integer Linear program Max {likelihood of assignment constraint violation penalty} Subset constraint Penalty Disjoint constraint Penalty Score of label assignment Subset constraint Disjoint Constraint 12

13. Missing classes: Missing classes: Soft Divide-And-Conquer 1 3 4 7 9 8 10 11 Near uniform? 13

14. Missing classes: Missing classes: Soft Divide-And-Conquer 1 3 4 7 9 8 10 11 Near uniform? ???? 14

15. Results: Results: 10% improvement F1 scores Flat Explore EM 75 OptDAC ExploreEM Macro avg. seed class F1 65 55 45 35 25 Level = 2 3 4 15

16. Results: Results: Ontology Extension 16

17. Datasets are made public Four hierarchical entity classification datasets are made publicly available at http://rtw.ml.cmu.edu/wk/WebSets/hierarchical_ ExploratoryLearning_WSDM2016/index.html 17

18. Thank You bhavanad@allenai.org 18