Download as pdf or txt
Download as pdf or txt
You are on page 1of 78

A FIRST APPROACH TO BELIEF

DYNAMICS IN COMPLEX
SOCIAL NETWORKS
Fabio R. Gallo, Gerardo I. Simari, María V. Martínez,
Natalia Abad Santos, and Marcelo A. Falappa

Institute for Computer Science and Engineering (UNS-CONICET)


Department of Computer Science and Engineering
Department of Mathematics
Universidad Nacional del Sur
Bahía Blanca, Argentina

1
Motivation

22
Motivation

32
Motivation

42
Motivation

52
Motivation

62
Motivation

72
Preliminaries

83
Social Network as Complex Network

: language, 𝑃𝑟𝑒𝑑: propositional symbols, only connective is ¬,


𝑉𝑃, 𝐸𝑃: vertex/1 and edge/2 predicate symbols.

A Social Network is a 4-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 ) where:


1. 𝑉 is a finite set whose elements are called vertices.
2. 𝐸 ⊆ 𝑉 × 𝑉 is a finite set whose elements are called edges.
3. 𝑙𝑣𝑒𝑟𝑡 : 𝑉 × 2 is a function called a vertex labelling function.
4. 𝑙𝑒𝑑𝑔𝑒 : 𝐸 → 2 is a function called an edge labelling function,
where = 𝑏, 𝑤 𝑏∈ , 𝑤 ∈ [0,1]}.

94
Social Network as Complex Network

: language, 𝑃𝑟𝑒𝑑: propositional symbols, only connective is ¬,


𝑉𝑃, 𝐸𝑃: vertex/1 and edge/2 predicate symbols.

A Social Network is a 4-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 ) where:


1. 𝑉 is a finite set whose elements are called vertices.
2. 𝐸 ⊆ 𝑉 × 𝑉 is a finite set whose elements are called edges.
3. 𝑙𝑣𝑒𝑟𝑡 : 𝑉 × 2 is a function called a vertex labelling function.
4. 𝑙𝑒𝑑𝑔𝑒 : 𝐸 → 2 is a function called an edge labelling function,
where = 𝑏, 𝑤 𝑏∈ , 𝑤 ∈ [0,1]}.

104
Social Network as Complex Network

: language, 𝑃𝑟𝑒𝑑: propositional symbols, only connective is ¬,


𝑉𝑃, 𝐸𝑃: vertex/1 and edge/2 predicate symbols.

A Social Network is a 4-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 ) where:


1. 𝑉 is a finite set whose elements are called vertices.
2. 𝐸 ⊆ 𝑉 × 𝑉 is a finite set whose elements are called edges.
3. 𝑙𝑣𝑒𝑟𝑡 : 𝑉 × 2 is a function called a vertex labelling function.
4. 𝑙𝑒𝑑𝑔𝑒 : 𝐸 → 2 is a function called an edge labelling function,
where = 𝑏, 𝑤 𝑏∈ , 𝑤 ∈ [0,1]}.

114
Social Network as Complex Network

: language, 𝑃𝑟𝑒𝑑: propositional symbols, only connective is ¬,


𝑉𝑃, 𝐸𝑃: vertex/1 and edge/2 predicate symbols.

A Social Network is a 4-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 ) where:


1. 𝑉 is a finite set whose elements are called vertices.
2. 𝐸 ⊆ 𝑉 × 𝑉 is a finite set whose elements are called edges.
3. 𝑙𝑣𝑒𝑟𝑡 : 𝑉 × 2 is a function called a vertex labelling function.
4. 𝑙𝑒𝑑𝑔𝑒 : 𝐸 → 2 is a function called an edge labelling function,
where = 𝑏, 𝑤 𝑏∈ , 𝑤 ∈ [0,1]}.

124
Social Network as Complex Network

: language, 𝑃𝑟𝑒𝑑: propositional symbols, only connective is ¬,


𝑉𝑃, 𝐸𝑃: vertex/1 and edge/2 predicate symbols.

A Social Network is a 4-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 ) where:


1. 𝑉 is a finite set whose elements are called vertices.
2. 𝐸 ⊆ 𝑉 × 𝑉 is a finite set whose elements are called edges.
3. 𝑙𝑣𝑒𝑟𝑡 : 𝑉 × 2 is a function called a vertex labelling function.
4. 𝑙𝑒𝑑𝑔𝑒 : 𝐸 → 2 is a function called an edge labelling function,
where = 𝑏, 𝑤 𝑏∈ , 𝑤 ∈ [0,1]}.

134
Social Network as Complex Network

: language, 𝑃𝑟𝑒𝑑: propositional symbols, only connective is ¬,


𝑉𝑃, 𝐸𝑃: vertex/1 and edge/2 predicate symbols.

A Social Network is a 4-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 ) where:


1. 𝑉 is a finite set whose elements are called vertices.
2. 𝐸 ⊆ 𝑉 × 𝑉 is a finite set whose elements are called edges.
3. 𝑙𝑣𝑒𝑟𝑡 : 𝑉 × 2 is a function called a vertex labelling function.
4. 𝑙𝑒𝑑𝑔𝑒 : 𝐸 → 2 is a function called an edge labelling function,
where = 𝑏, 𝑤 𝑏∈ , 𝑤 ∈ [0,1]}.

144
Social Network as Complex Network
Example:

155
Social Network as Complex Network
Example:

165
Social Network as Complex Network
Example:

175
Network Knowledge Base

A Network KB (NKB for short) is a 5-tuple (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 , 𝐾):

• The first four elements comprise a social network.

• 𝐾: 𝑉 → 2 is a mapping assigning a knowledge base to each


vertex.

For every vertex 𝑣𝑖 , 𝐾 𝑣𝑖 : knowledge base associated with 𝑣𝑖 .

186
Constraints

A constraint C over an 𝑁𝐾𝐵 (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 , 𝐾) is a pair


(𝑆, 𝐵) where, given 𝑣1 , … , 𝑣𝑛 ∈ 𝑉, and 𝑒1 , … , 𝑒𝑚 ∈ 𝐸 ∩
𝑣1 , … , 𝑣𝑛 × {𝑣1 , … , 𝑣𝑛 },
1. 𝑆, called the structural part, contains a conjuction of
conditions that can be of either of the following forms:
• 𝑙𝑣𝑒𝑟𝑡 𝑣 = 𝑎, 𝑎 ∈ 𝑉𝑃, 𝑣 ∈ {𝑣1 , … , 𝑣𝑛 };
• 𝑏, 𝑤 ∈ 𝑙𝑒𝑑𝑔𝑒 𝑒 , 𝑏, 𝑤 ∈ 𝑇, 𝑤 ∈ 𝛼, 𝛽 , for some 0 ≤
𝛼 ≤ 𝛽 ≤ 1.
2. 𝐵 is called the belief part and contains a conjuction of
conditions involving elements in 𝐾 𝑣1 , … , 𝐾(𝑣𝑛 ).

197
Constraints

A constraint C over an 𝑁𝐾𝐵 (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 , 𝐾) is a pair


(𝑆, 𝐵) where, given 𝑣1 , … , 𝑣𝑛 ∈ 𝑉, and 𝑒1 , … , 𝑒𝑚 ∈ 𝐸 ∩
𝑣1 , … , 𝑣𝑛 × {𝑣1 , … , 𝑣𝑛 },
1. 𝑆, called the structural part, contains a conjuction of
conditions that can be of either of the following forms:
• 𝑙𝑣𝑒𝑟𝑡 𝑣 = 𝑎, 𝑎 ∈ 𝑉𝑃, 𝑣 ∈ {𝑣1 , … , 𝑣𝑛 };
• 𝑏, 𝑤 ∈ 𝑙𝑒𝑑𝑔𝑒 𝑒 , 𝑏, 𝑤 ∈ 𝑇, 𝑤 ∈ 𝛼, 𝛽 , for some 0 ≤
𝛼 ≤ 𝛽 ≤ 1.
2. 𝐵 is called the belief part and contains a conjuction of
conditions involving elements in 𝐾 𝑣1 , … , 𝐾(𝑣𝑛 ).

207
Constraints

A constraint C over an 𝑁𝐾𝐵 (𝑉, 𝐸, 𝑙𝑣𝑒𝑟𝑡 , 𝑙𝑒𝑑𝑔𝑒 , 𝐾) is a pair


(𝑆, 𝐵) where, given 𝑣1 , … , 𝑣𝑛 ∈ 𝑉, and 𝑒1 , … , 𝑒𝑚 ∈ 𝐸 ∩
𝑣1 , … , 𝑣𝑛 × {𝑣1 , … , 𝑣𝑛 },
1. 𝑆, called the structural part, contains a conjuction of
conditions that can be of either of the following forms:
• 𝑙𝑣𝑒𝑟𝑡 𝑣 = 𝑎, 𝑎 ∈ 𝑉𝑃, 𝑣 ∈ {𝑣1 , … , 𝑣𝑛 };
• 𝑏, 𝑤 ∈ 𝑙𝑒𝑑𝑔𝑒 𝑒 , 𝑏, 𝑤 ∈ 𝑇, 𝑤 ∈ 𝛼, 𝛽 , for some 0 ≤
𝛼 ≤ 𝛽 ≤ 1.
2. 𝐵 is called the belief part and contains a conjuction of
conditions involving elements in 𝐾 𝑣1 , … , 𝐾(𝑣𝑛 ).

217
NKB example

228
NKB example

238
NKB example

248
NKB example

258
NKB example

268
Overview of local and
global revision

279
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

10
28
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
29
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
30
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
31
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
32
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
33
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

𝜷 represents a
support for iPhone.

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
34
News items
A news item consists of a triple 〈𝒐, 𝒍, 𝒅〉, where:
𝒐: 𝑜𝑟𝑖𝑔𝑖𝑛
𝒍: 𝑙𝑖𝑡𝑒𝑟𝑎𝑙
𝒅: 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ( add / remove / flipped)

Example:

𝜷 represents a
support for iPhone.

〈𝑓𝑜𝑙𝑙-T, 0.8〉

10
35
11
36
Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠

11
37
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠

11
38
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

11
39
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

𝑮𝒍𝒐𝒃𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏

11
40
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

𝑁𝐾𝐵′

𝐼𝑛𝑡𝑒𝑔𝑟𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

𝑮𝒍𝒐𝒃𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏

11
41
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

𝑁𝐾𝐵′

𝐼𝑛𝑡𝑒𝑔𝑟𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

𝑮𝒍𝒐𝒃𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏

11
42
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

𝑁𝐾𝐵′

𝐼𝑛𝑡𝑒𝑔𝑟𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

𝑮𝒍𝒐𝒃𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏

11
43
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

𝐴𝑙𝑙 𝐼𝐶𝑠 𝑠ℎ𝑜𝑢𝑙𝑑


𝑏𝑒 𝑠𝑎𝑡𝑖𝑠𝑓𝑖𝑒𝑑

𝑁𝐾𝐵
𝑁𝐾𝐵′

𝐼𝑛𝑡𝑒𝑔𝑟𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

𝑮𝒍𝒐𝒃𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏

11
44
𝑳𝒐𝒄𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏𝒔

Updates
𝑎𝑛𝑑 𝑛𝑒𝑤𝑠 𝑖𝑡𝑒𝑚𝑠
Some ICs could
be violated after
the local changes

𝐴𝑙𝑙 𝐼𝐶𝑠 𝑠ℎ𝑜𝑢𝑙𝑑


𝑏𝑒 𝑠𝑎𝑡𝑖𝑠𝑓𝑖𝑒𝑑

𝑁𝐾𝐵
𝑁𝐾𝐵′

𝐼𝑛𝑡𝑒𝑔𝑟𝑖𝑡𝑦 𝐶𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠

𝑮𝒍𝒐𝒃𝒂𝒍 𝒓𝒆𝒗𝒊𝒔𝒊𝒐𝒏

11
45
Local
Belief Revision

12
46
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
1. Inclusion: 𝐾 ′ 𝑣 ⊆ 𝐾 𝑣 ∪ 𝑙𝑖𝑡(𝑃𝑣 ).
No unwarranted information should be added as part of a revision.

2. Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣).


If this property holds, the input is guaranteed to be accepted.

3. Weak Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣) when 𝑙𝑖𝑡 𝑃𝑣 is consistent.


Same as Success, except that it only applies when the new information is
consistent with the existing one.
13
47
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
1. Inclusion: 𝐾 ′ 𝑣 ⊆ 𝐾 𝑣 ∪ 𝑙𝑖𝑡(𝑃𝑣 ).
No unwarranted information should be added as part of a revision.

2. Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣).


If this property holds, the input is guaranteed to be accepted.

3. Weak Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣) when 𝑙𝑖𝑡 𝑃𝑣 is consistent.


Same as Success, except that it only applies when the new information is
consistent with the existing one.
13
48
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
1. Inclusion: 𝐾 ′ 𝑣 ⊆ 𝐾 𝑣 ∪ 𝑙𝑖𝑡(𝑃𝑣 ).
No unwarranted information should be added as part of a revision.

2. Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣).


If this property holds, the input is guaranteed to be accepted.

3. Weak Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣) when 𝑙𝑖𝑡 𝑃𝑣 is consistent.


Same as Success, except that it only applies when the new information is
consistent with the existing one.
13
49
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
1. Inclusion: 𝐾 ′ 𝑣 ⊆ 𝐾 𝑣 ∪ 𝑙𝑖𝑡(𝑃𝑣 ).
No unwarranted information should be added as part of a revision.

2. Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣).


If this property holds, the input is guaranteed to be accepted.

3. Weak Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣) when 𝑙𝑖𝑡 𝑃𝑣 is consistent.


Same as Success, except that it only applies when the new information is
consistent with the existing one.
13
50
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
1. Inclusion: 𝐾 ′ 𝑣 ⊆ 𝐾 𝑣 ∪ 𝑙𝑖𝑡(𝑃𝑣 ).
No unwarranted information should be added as part of a revision.

2. Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣).


If this property holds, the input is guaranteed to be accepted.

3. Weak Success: 𝑙𝑖𝑡 𝑃𝑣 ⊆ 𝐾′(𝑣) when 𝑙𝑖𝑡 𝑃𝑣 is consistent.


Same as Success, except that it only applies when the new information is
consistent with the existing one.
13
51
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
4. Consistency: 𝐾 ′ 𝑣 ⊢⊥.
The result of the operation must be consistent.

5. Vacuity 1: If 𝑙 ∉ 𝐾(𝑣), 𝑙𝑖𝑡 𝑝 = 𝑙 implies 𝑑𝑒𝑐 𝑝 = 𝑟, for all


𝑝 ∈ 𝑃𝑣 , then 𝑙 ∉ 𝐾 ′ (𝑣).
If the current knowledge base contains an element e, and all the news items
in the input that refer to e have removed it, then e is not part of the output.

14
52
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
4. Consistency: 𝐾 ′ 𝑣 ⊢⊥.
The result of the operation must be consistent.

5. Vacuity 1: If 𝑙 ∉ 𝐾(𝑣), 𝑙𝑖𝑡 𝑝 = 𝑙 implies 𝑑𝑒𝑐 𝑝 = 𝑟, for all


𝑝 ∈ 𝑃𝑣 , then 𝑙 ∉ 𝐾 ′ (𝑣).
If the current knowledge base contains an element e, and all the news items
in the input that refer to e have removed it, then e is not part of the output.

14
53
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
6. Vacuity 2: If 𝑙 ∈ 𝐾(𝑣), 𝑙𝑖𝑡 𝑝 = ¬𝑙 implies 𝑑𝑒𝑐 𝑝 = 𝑟, for all
𝑝 ∈ 𝑃𝑣 , then 𝑙 ∈ 𝐾 ′ (𝑣).
As a kind of dual to Vacuity 1, this property states that if the current knowledge
base contains an element 𝑒, and all the news items in the input that refer to 𝑒
have removed its negation, then the output should still contain 𝑒.

7. Weak Vacuity 1: If 𝑙 ∉ 𝐾(𝑣) and 𝑙𝑖𝑡 𝑝 ≠ 𝑙 for all 𝑝 ∈ 𝑃𝑣 , then


𝑙 ∉ 𝐾 ′ (𝑣).
If the current knowledge base does not contain an element 𝑒, and none of the
news items in the input refer to 𝑒, then 𝑒 is not part of the output.
15
54
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
6. Vacuity 2: If 𝑙 ∈ 𝐾(𝑣), 𝑙𝑖𝑡 𝑝 = ¬𝑙 implies 𝑑𝑒𝑐 𝑝 = 𝑟, for all
𝑝 ∈ 𝑃𝑣 , then 𝑙 ∈ 𝐾 ′ (𝑣).
As a kind of dual to Vacuity 1, this property states that if the current knowledge
base contains an element 𝑒, and all the news items in the input that refer to 𝑒
have removed its negation, then the output should still contain 𝑒.

7. Weak Vacuity 1: If 𝑙 ∉ 𝐾(𝑣) and 𝑙𝑖𝑡 𝑝 ≠ 𝑙 for all 𝑝 ∈ 𝑃𝑣 , then


𝑙 ∉ 𝐾 ′ (𝑣).
If the current knowledge base does not contain an element 𝑒, and none of the
news items in the input refer to 𝑒, then 𝑒 is not part of the output.
15
55
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
8. Weak Vacuity 2: If 𝑙 ∈ 𝐾(𝑣) and 𝑙𝑖𝑡 𝑝 ≠ ¬𝑙 for all 𝑝 ∈ 𝑃𝑣 ,
then 𝑙 ∈ 𝐾 ′ (𝑣).
As a kind of dual to Weak Vacuity 1, this property states that if the current
knowledge base contains an element 𝑒, and none of the news items in the
input refer to its negation then the output should still contain 𝑒.

9. Strong Congruence: Let 𝑃𝑣∗ a set of news ítems and NKB∗ = ⊛


(𝑁𝐾𝐵, 𝑣, 𝑃𝑣∗ ) , if 𝜒 𝑃𝑣 = 𝜒(𝑃𝑣∗ ) then 𝐾 ′ 𝑣 = 𝐾 ∗ (𝑣).
If two sets of news item sets are equivalent, the respective outputs should be
identical; i.e., the result does not depend on the syntax used to express the input.
16
56
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
8. Weak Vacuity 2: If 𝑙 ∈ 𝐾(𝑣) and 𝑙𝑖𝑡 𝑝 ≠ ¬𝑙 for all 𝑝 ∈ 𝑃𝑣 ,
then 𝑙 ∈ 𝐾 ′ (𝑣).
As a kind of dual to Weak Vacuity 1, this property states that if the current
knowledge base contains an element 𝑒, and none of the news items in the
input refer to its negation then the output should still contain 𝑒.

9. Strong Congruence: Let 𝑃𝑣∗ a set of news ítems and NKB∗ = ⊛


(𝑁𝐾𝐵, 𝑣, 𝑃𝑣∗ ) , if 𝜒 𝑃𝑣 = 𝜒(𝑃𝑣∗ ) then 𝐾 ′ 𝑣 = 𝐾 ∗ (𝑣).
If two sets of news item sets are equivalent, the respective outputs should be
identical; i.e., the result does not depend on the syntax used to express the input.
16
57
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
10. Weak Congruence: Let 𝑃+ = 𝑝 ∈ 𝜒 𝑃 𝑑𝑒𝑐 𝑝 = 𝑎} be a
set of all positive/added news items of 𝜒 𝑃 , 𝑃− =
𝑝 ∈ 𝑃 𝑑𝑒𝑐 𝑝 = 𝑟} be a set of all negative/removed news
ítems of 𝑃 and let 𝑃𝑣∗ be a set of news items and 𝑁𝐾𝐵∗ = ⊛
(NKB, 𝑣, 𝑃𝑣∗ ) = (𝑉 ∗ , 𝐸 ∗ , 𝑙𝑣𝑒𝑟𝑡
∗ ∗
, 𝑙𝑒𝑑𝑔𝑒 , 𝐾 ∗ ). If 𝑃𝑣+ = 𝑃𝑣∗+ , and 𝑃𝑣− ⊆ 𝑃𝑣∗−
then 𝐾 ∗ 𝑣 ⊆ 𝐾′(𝑣).
Weaker version of the previous one: if both inputs are identical with respect to
the content they add, but one of the inputs removes a superset of what the other
removes, then the output for the former is a subset of the output for the latter.

17
58
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:

11. Majority: Let 𝑃𝑟𝑜 = 𝑝 ∈ 𝑃 𝑙𝑖𝑡 𝑝 = 𝑙, 𝑑𝑒𝑐 𝑝 ≠ 𝑟} ⊆ 𝑃𝑣 and


𝐶𝑜𝑛 = 𝑝 ∈ 𝑃𝑣 𝑙𝑖𝑡 𝑝 = ¬𝑙, 𝑑𝑒𝑐 𝑝 ≠ 𝑟} ⊆ 𝑃𝑣 . Given 𝑙 ∈
s.t. 𝑙 ∉ 𝐾(𝑣) and ¬𝑙 ∉ 𝐾(𝑣), if 𝑃𝑟𝑜 > 𝐶𝑜𝑛 then ¬𝑙 ∉
𝐾 ′ (𝑣), and if 𝑃𝑟𝑜 < 𝐶𝑜𝑛 then 𝑙 ∉ 𝐾′(𝑣).

18
59
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:
12. Weighted Majority: For NKBs that have one label per edge, given
𝑙 ∉ 𝐾(𝑣) and ¬𝑙 ∉ 𝐾(𝑣), let 𝑠𝑢𝑚𝑊𝑃𝑜𝑠𝑡 and 𝑠𝑢𝑚𝑊𝑁𝑒𝑔 be
two values calculated as follows:
𝑠𝑢𝑚𝑊𝑃𝑜𝑠 = Σ𝑒∈𝐼 𝑤𝑒𝑖𝑔ℎ𝑡 𝑒 ; where I = 𝑒 ∈ 𝐸 𝑒 =
𝑣, 𝑠𝑟𝑐 𝑝 , 𝑙𝑖𝑡 𝑝 = 𝑙, 𝑝 ∈ }.
𝑠𝑢𝑚𝑊𝑁𝑒𝑔 = Σ𝑒∈𝐽 𝑤𝑒𝑖𝑔ℎ𝑡 𝑒 ; where J = 𝑒 ∈ 𝐸 𝑒 =
𝑣, 𝑠𝑟𝑐 𝑝 , 𝑙𝑖𝑡 𝑝 = ¬𝑙, 𝑝 ∈ }.
If 𝑠𝑢𝑚𝑊𝑃𝑜𝑠 > 𝑠𝑢𝑚𝑊𝑁𝑒𝑔 then ¬𝑙 ∉ 𝐾′(𝑣);
if 𝑠𝑢𝑚𝑊𝑃𝑜𝑠 < 𝑠𝑢𝑚𝑊𝑁𝑒𝑔 then 𝑙 ∉ 𝐾′(𝑣)
19
60
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:

13. Local Effect: ∀𝑤 ∈ 𝑉 s.t. 𝑤 ≠ 𝑣, 𝐾 ′ 𝑤 = 𝐾(𝑤).


Applying an operator must not have any effect on other nodes’ KBs.

14. Structural Preservation: 𝑉 = 𝑉 ′ , 𝐸 = 𝐸 ′ , 𝑙𝑣𝑒𝑟𝑡 = 𝑙′𝑣𝑒𝑟𝑡 and


𝑙𝑒𝑑𝑔𝑒 = 𝑙′𝑒𝑑𝑔𝑒 .

No vertex, edge, or label is modified by a local revision operation.

20
61
Postulates
Local BR operators are functions ⊛: ×𝑽×𝟐 → .
Let 𝑵𝑲𝑩′ = ⊛ (NKB, 𝒗, 𝑷) = (𝑽′ , 𝑬′ , 𝒍′𝒗𝒆𝒓𝒕 , 𝒍′𝒆𝒅𝒈𝒆 , 𝑲′).
We propose a set of postulates as reasonable properties for
local NKB revision:

13. Local Effect: ∀𝑤 ∈ 𝑉 s.t. 𝑤 ≠ 𝑣, 𝐾 ′ 𝑤 = 𝐾(𝑤).


Applying an operator must not have any effect on other nodes’ KBs.

14. Structural Preservation: 𝑉 = 𝑉 ′ , 𝐸 = 𝐸 ′ , 𝑙𝑣𝑒𝑟𝑡 = 𝑙′𝑣𝑒𝑟𝑡 and


𝑙𝑒𝑑𝑔𝑒 = 𝑙′𝑒𝑑𝑔𝑒 .

No vertex, edge, or label is modified by a local revision operation.

21
62
NKB Operator
A local NKB revision operator is basic if it satisfies Structural
Preservation, Local Effect, Consistency, Uniformity, and
Inclusion.
Let ⊛ be a basic local NKB revision operator:
⊛ is restrained if it satisfied Strong Congruence, Vacuity 1,
and Vacuity 2.
⊛ is weakly restrained if it satisfies W. Congruence, and W.
Vacuity 1 and 2.
⊛ is social if it satisfies Weak Success, and either Majority or
Weighted Majority.

22
63
Towards Constructions based on User Types

Credulous Incredulous Herd behavior

Blind follower Cautious Self-confident

23
64
Towards Constructions based on User Types

Credulous Incredulous Herd behavior

Blind follower Cautious Self-confident

23
65
Experimental
Evaluation

24
66
Experimental Evaluation
Twitter dataset from several election cycles in India (between
Jul-2013 and May-2015)

 184,654 users
 66,827,454 follow relationships
 18,292,721 tweets
• 5,107,986 tweets with hashtags
• 136,809 distinct hashtags
Sentiment Analysis
 We obtain the sentiment associated with the use of each
hashtag (PHPInsight).
25
67
Experimental Evaluation
User Avg. Tweet User Avg. Tweet User Avg. Tweet
Reaction Reaction Reaction
3% (61% / 36%) 0% (0% / 100%) 67% (0% / 33)
𝒂𝟏 𝒂𝟓 𝒂𝟗
2% (73% / 25%) 0% (0% / 100%) 0% (92% / 8%)
38% (8% / 54%) 0% (0% / 100%) 0% (62% / 38%)
0% (39% / 61%) 9% (2% / 89%) 1% (51% / 48%)
𝒂𝟐 𝒂𝟔 𝒂𝟏𝟎
0% (53% / 47%) 0% (31% / 69%) 1% (64% / 35%)
20% (0% / 80%) 1% (9% / 90%) 32% (2% / 66%)
2% (54% / 44%) 24% (2% / 74%) 11% (0% / 89%)
𝒂𝟑 𝒂𝟕 𝒂𝟏𝟏
0% (67% / 33%) 1% (55% / 44%) 0% (33% / 67%)
34% (4% / 62%) 1% (23% / 76%) 0% (10% / 90)
23% (13% / 64%) 1% (43% / 56%) 9% (2% / 89%)
𝒂𝟒 𝒂𝟖 𝒂𝟏𝟐
1% (60% / 39%) 22% (12% / 66%) 0% (31% / 69%)
7% (25% / 68%) 2% (23% / 75%) 1% (9% / 90%)

26
68
Experimental Evaluation
User Avg. Tweet User Avg. Tweet User Avg. Tweet
Reaction Reaction Reaction
3% (61% / 36%) 0% (0% / 100%) 67% (0% / 33)
𝒂𝟏 𝒂𝟓 𝒂𝟗
2% (73% / 25%) 0% (0% / 100%) 0% (92% / 8%)
38% (8% / 54%) 0% (0% / 100%) 0% (62% / 38%)
0% (39% / 61%) 9% (2% / 89%) 1% (51% / 48%)
𝒂𝟐 𝒂𝟔 𝒂𝟏𝟎
0% (53% / 47%) 0% (31% / 69%) 1% (64% / 35%)
20% (0% / 80%) 1% (9% / 90%) 32% (2% / 66%)
2% (54% / 44%) 24% (2% / 74%) 11% (0% / 89%)
𝒂𝟑 𝒂𝟕 𝒂𝟏𝟏
0% (67% / 33%) 1% (55% / 44%) 0% (33% / 67%)
34% (4% / 62%) 1% (23% / 76%) 0% (10% / 90)
23% (13% / 64%) 1% (43% / 56%) 9% (2% / 89%)
𝒂𝟒 𝒂𝟖 𝒂𝟏𝟐
1% (60% / 39%) 22% (12% / 66%) 0% (31% / 69%)
7% (25% / 68%) 2% (23% / 75%) 1% (9% / 90%)

26
69
Experimental Evaluation
User Avg. Tweet User Avg. Tweet User Avg. Tweet
Reaction Reaction Reaction
3% (61% / 36%) 0% (0% / 100%) 67% (0% / 33)
𝒂𝟏 𝒂𝟓 𝒂𝟗
2% (73% / 25%) 0% (0% / 100%) 0% (92% / 8%)
38% (8% / 54%) 0% (0% / 100%) 0% (62% / 38%)
0% (39% / 61%) 9% (2% / 89%) 1% (51% / 48%)
𝒂𝟐 𝒂𝟔 𝒂𝟏𝟎
0% (53% / 47%) 0% (31% / 69%) 1% (64% / 35%)
20% (0% / 80%) 1% (9% / 90%) 32% (2% / 66%)
2% (54% / 44%) 24% (2% / 74%) 11% (0% / 89%)
𝒂𝟑 𝒂𝟕 𝒂𝟏𝟏
0% (67% / 33%) 1% (55% / 44%) 0% (33% / 67%)
34% (4% / 62%) 1% (23% / 76%) 0% (10% / 90)
23% (13% / 64%) 1% (43% / 56%) 9% (2% / 89%)
𝒂𝟒 𝒂𝟖 𝒂𝟏𝟐
1% (60% / 39%) 22% (12% / 66%) 0% (31% / 69%)
7% (25% / 68%) 2% (23% / 75%) 1% (9% / 90%)

26
70
Experimental Evaluation
User Avg. Tweet User Avg. Tweet User Avg. Tweet
Reaction Reaction Reaction
3% (61% / 36%) 0% (0% / 100%) 67% (0% / 33)
𝒂𝟏 𝒂𝟓 𝒂𝟗
2% (73% / 25%) 0% (0% / 100%) 0% (92% / 8%)
38% (8% / 54%) 0% (0% / 100%) 0% (62% / 38%)
0% (39% / 61%) 9% (2% / 89%) 1% (51% / 48%)
𝒂𝟐 𝒂𝟔 𝒂𝟏𝟎
0% (53% / 47%) 0% (31% / 69%) 1% (64% / 35%)
20% (0% / 80%) 1% (9% / 90%) 32% (2% / 66%)
2% (54% / 44%) 24% (2% / 74%) 11% (0% / 89%)
𝒂𝟑 𝒂𝟕 𝒂𝟏𝟏
0% (67% / 33%) 1% (55% / 44%) 0% (33% / 67%)
34% (4% / 62%) 1% (23% / 76%) 0% (10% / 90)
23% (13% / 64%) 1% (43% / 56%) 9% (2% / 89%)
𝒂𝟒 𝒂𝟖 𝒂𝟏𝟐
1% (60% / 39%) 22% (12% / 66%) 0% (31% / 69%)
7% (25% / 68%) 2% (23% / 75%) 1% (9% / 90%)

26
71
Experimental Evaluation
User Avg. Tweet User Avg. Tweet User Avg. Tweet
Reaction Reaction Reaction
3% (61% / 36%) 0% (0% / 100%) 67% (0% / 33)
𝒂𝟏 𝒂𝟓 𝒂𝟗
2% (73% / 25%) 0% (0% / 100%) 0% (92% / 8%)
38% (8% / 54%) 0% (0% / 100%) 0% (62% / 38%)
0% (39% / 61%) 9% (2% / 89%) 1% (51% / 48%)
𝒂𝟐 𝒂𝟔 𝒂𝟏𝟎
0% (53% / 47%) 0% (31% / 69%) 1% (64% / 35%)
20% (0% / 80%) 1% (9% / 90%) 32% (2% / 66%)
2% (54% / 44%) 24% (2% / 74%) 11% (0% / 89%)
𝒂𝟑 𝒂𝟕 𝒂𝟏𝟏
0% (67% / 33%) 1% (55% / 44%) 0% (33% / 67%)
34% (4% / 62%) 1% (23% / 76%) 0% (10% / 90)
23% (13% / 64%) 1% (43% / 56%) 9% (2% / 89%)
𝒂𝟒 𝒂𝟖 𝒂𝟏𝟐
1% (60% / 39%) 22% (12% / 66%) 0% (31% / 69%)
7% (25% / 68%) 2% (23% / 75%) 1% (9% / 90%)

26
72
Experimental Evaluation
User Avg. Tweet User Avg. Tweet User Avg. Tweet
Reaction Reaction Reaction
3% (61% / 36%) 0% (0% / 100%) 67% (0% / 33)
𝒂𝟏 𝒂𝟓 𝒂𝟗
2% (73% / 25%) 0% (0% / 100%) 0% (92% / 8%)
38% (8% / 54%) 0% (0% / 100%) 0% (62% / 38%)
0% (39% / 61%) 9% (2% / 89%) 1% (51% / 48%)
𝒂𝟐 𝒂𝟔 𝒂𝟏𝟎
0% (53% / 47%) 0% (31% / 69%) 1% (64% / 35%)
20% (0% / 80%) 1% (9% / 90%) 32% (2% / 66%)
2% (54% / 44%) 24% (2% / 74%) 11% (0% / 89%)
𝒂𝟑 𝒂𝟕 𝒂𝟏𝟏
0% (67% / 33%) 1% (55% / 44%) 0% (33% / 67%)
34% (4% / 62%) 1% (23% / 76%) 0% (10% / 90)
23% (13% / 64%) 1% (43% / 56%) 9% (2% / 89%)
𝒂𝟒 𝒂𝟖 𝒂𝟏𝟐
1% (60% / 39%) 22% (12% / 66%) 0% (31% / 69%)
7% (25% / 68%) 2% (23% / 75%) 1% (9% / 90%)
𝑢1 , 𝑢2 , 𝑢3 , 𝑎𝑛𝑑 𝑢10 can be seen as self-confident users.
26
73
Conclusions
and
Future Work

27
74
Conclusions and Future Work

In this work we:

 characterized NKBs.

 proposed suitable postulates for local revision in NKBs.

 discussed possible ideas for constructions.

 presented the results of a preliminary experiment.

In future work we will further formalize and evaluate local


operators, and also work on global revision.

28
75
29
76
30
77
References
1. Allcott, H., Gentzkow, M.: Social media and fake news in the 2016 election.
Tech. rep., National Bureau of Economic Research (2017)
2. Gallo, F.R., Abad Santos, N., Simari, G.I., Falappa, M.A.: A desiderata for
modeling and reasoning with social knowledge. In: Proc. of CACIC 2015 (2015)
3. Gallo, F.R., Abad Santos, N., Simari, G.I., Martinez, M.V., Falappa, M.A.: Belief
dynamics in complex social networks. In: Proc. of ASAI 2016–JAIIO 45 (2016)
4. Kempe, D., Kleinberg, J., Tardos, E.: Maximizing the spread of influence through
a social network. In: Proc. of KDD ’03. pp. 137–146. ACM (2003)
5. Shakarian, P., Broecheler, M., Subrahmanian, V.S., Molinaro, C.: Using
generalized annotated programs to solve social network diffusion optimization
problems. ACM Trans. Comput. Logic 14(2), 10:1–10:40 (2013)
6. Shakarian, P., Simari, G.I., Callahan, D.: Reasoning about complex networks: A
logic programming approach. TPLP 13(4-5-Online-Supplement) (2013)
7. Shao, C., Ciampaglia, G.L., Flammini, A., Menczer, F.: Hoaxy: A platform for
tracking online misinformation. In: Proc. WWW ’16 Comp. pp. 745–750 (2016)

31
78

You might also like