Kristina T outanova Joint work with entities relations city_of - - PowerPoint PPT Presentation

β–Ά
kristina t outanova
SMART_READER_LITE
LIVE PREVIEW

Kristina T outanova Joint work with entities relations city_of - - PowerPoint PPT Presentation

Aug 15, 2023 399 likes β€’675 views

Kristina T outanova Joint work with entities relations city_of Honolulu born_in United States Barack Obama spouse Michelle Obama city_of Honolulu born_in lived_in United States Barack Obama nationality spouse Michelle Obama

slide-1
SLIDE 1

Kristina T

  • utanova

Joint work with

slide-2
SLIDE 2

entities relations

Barack Obama Honolulu United States Michelle Obama

spouse city_of born_in

slide-3
SLIDE 3

Barack Obama Honolulu United States Michelle Obama

spouse

city_of born_in nationality lived_in

slide-4
SLIDE 4

Barack Obama Honolulu United States Michelle Obama

spouse

city_of born_in nationality nationality

slide-5
SLIDE 5

Chicago

lived_in

Barack Obama Honolulu United States Michelle Obama

spouse

city_of born_in

Barack Obama worked in Chicago. A photo of Barack Obama’s Chicago house. Goal: use both sources of information. [Lao et al. 2012], [Gardner et al.

2013,2014] [Riedel et al. 2013] [Neelakantan et al 2015]

slide-6
SLIDE 6

Embedding-based models for KB completion

  • Using text: Universal Schema
slide-7
SLIDE 7

lived_in

f(Michelle Obama, lived_in, Chicago)

RESCAL [Nickel et al 2011], TransE [Bordes et al 2011, 2013], Bilinear-diagonal [Yang et al. 2015] Encoding relevant properties of the entities, predictive of their relationships. Encoding relevant properties of the relations that help define the set of entity pairs for which the relation holds.

  • 0.1

2.3

  • 1.4
  • 0.7
  • 3.4

1.6

Michelle Obama

lived_in Michelle Chicago

slide-8
SLIDE 8

KB relations Textual relations

founder

Chicago

lived_in

Barack Obama Honolulu United States Michelle Obama

spouse

city_of born_in

slide-9
SLIDE 9

Embedding-based models for KB completion

  • Using text: Universal Schema

Sparsity of textual relations

  • Compositional representations of text
slide-10
SLIDE 10
slide-11
SLIDE 11

KB relations Textual relations KB relations Textual relations

Basic Conv

slide-12
SLIDE 12

[Toutanova, Chen, Pantel, Poon, Choudhury, Gamon, EMNLP 2015]

slide-13
SLIDE 13

37.3 37.7 40.1

35 36 37 38 39 40 41 Mean Reciprocal Rank (MRR) KB Inference KB + text-basic KB+text compositional Evaluation on held out queries : Where did Michelle Obama live? Mean reciprocal rank of first correct answer (times 100)

FB15K-237 new dataset from Freebase and ClueWeb. http://research.microsoft.com/en-us/downloads/3a9bf02d-b791-4e95-b88d-389feef3e421/

slide-14
SLIDE 14

Embedding-based models for KB completion

  • Using text: Universal Schema

Are node/relation embeddings sufficient

  • Using observed features
slide-15
SLIDE 15

Barack Obama Honolulu Michelle Obama

spouse

born_in

Michelle Obama

spouse

? Honolulu

PEOPLE_BORN_IN

?

slide-16
SLIDE 16

Barack Obama Honolulu Michelle Obama

spouse

born_in

f(x=Michelle Obama, spouse, y=Barack Obama) = π‘₯1𝜚1 = 99.8

y

spouse

x

slide-17
SLIDE 17

56.2 82.2 20 40 60 80 100 MRR E+Bilinear-diag Simple Observed

slide-18
SLIDE 18

Embedding-based models for KB completion

  • Using text: Universal Schema

Inference from multi-step relation paths from KB and text

  • Efficient compositional representation and learning
slide-19
SLIDE 19
slide-20
SLIDE 20
slide-21
SLIDE 21
slide-22
SLIDE 22

IL2 MAPK1 [Toutanova, Lin, Yih, Poon, Quirk, ACL 2016]

slide-23
SLIDE 23

IL2 MAPK1

slide-24
SLIDE 24

πΊπ‘š 𝑑, 𝑒 = ෍

𝜌 =π‘š

𝑄 𝑒 𝑑, 𝜌 Ξ¦ 𝜌 ) πΊπ‘š+1 𝑑, 𝑒 = ෍

𝑙

𝐺

1 𝑑, 𝑙 πΊπ‘š(𝑙, 𝑒)

slide-25
SLIDE 25

39.92 48.27 48.6 52.53

36 38 40 42 44 46 48 50 52 54

Hits@10

Hits@10 on Gene Regulation Bilinear-diag NBestPaths All Paths All Paths+Nodes

slide-26
SLIDE 26