en
0.25
0.5
0.75
1.25
1.5
1.75
2
Knowledge Vault: A Web-Scale Approach to Probabilistic Knowledge Fusion
Published on Oct 07, 20145967 Views
Recent years have witnessed a proliferation of large-scale knowledge bases, including Wikipedia, Freebase, YAGO, Microsoft's Satori, and Google's Knowledge Graph. To increase the scale even further, w
Related categories
Chapter list
Knowledge Vault: a web-scale approach to probabilistic knowledge fusion00:00
Outline of the talk - 100:37
A Knowledge Graph is a multi-graph where nodes = entities, edges = relations 00:50
Example Knowledge Graphs 01:13
Freebase is created by fusing structured data sources and human contributions01:43
The long tail of knowledge02:19
Outline of the talk - 203:07
From Knowledge Graph to Knowledge Vault 03:10
Previous projects on automatically building KBs (eg NELL, YAGO) predict facts based on text03:59
KV: Predict new facts based on text AND existing edges in FB 04:48
High Level Graph - 105:17
KV is 50x bigger than comparable KBs05:40
Uses for KV's uncertain triples 06:18
Outline of the talk - 307:14
Fact extraction from the web07:23
Fact extraction from text (TXT) 07:46
Fact extraction from DOM trees*08:49
Fact extraction from tables (TBL)*09:10
Fact extraction from schema.org annotation (ANO) 09:27
Combine outputs from all extractors10:19
ROC for each extraction system10:39
Confidence of true facts rises given more evidence 11:29
Outline of the talk - 412:17
Mining facts from graphs 12:30
Link prediction using tensor factorization12:38
(Deep) neural network for link prediction13:37
Path Ranking Algorithm [Lao et al., EMNLP11]14:37
Example of paths / rules learned by PRA15:38
PRA similar in performance to neural network15:55
Outline of the talk - 516:12
High Level Graph - 216:18
Fusing web extractions with graph priors 16:29
Example: (Barry Richter, studiedAt, UW-Madison) 16:42
Summary and future work17:38