SlideShare a Scribd company logo
On the Family of Concept Forming Operators in
Polyadic FCA
Dmitry I. Ignatov
National Research University Higher School of Economics, Moscow, Russia
Department of Data Analysis and Machine Intelligence, Faculty of Computer
Science &
Laboratory of Intelligent Systems and Structural Analysis
FCA4KD
Moscow
2017
1 / 24
Outline
1 Formal Concept Analysis (FCA)
2 Triadic FCA
3 Towards a closure operator for 3-adic case
2 / 24
Motivation
A large amount of structured and unstructured data generates
triadic relations.
E.g. folksonomy is a set of triples (user, object, tag)
Concrete examples:
Bibsonomy.org (user,
bookmark, tag)
Social networking sites
(user, group, interest)
Delicious (user, link,
tag)
Figure: Folksonomy as a graph.
3 / 24
Main research question
Are there concept-forming operators for n-ry case?
(answered more than two decades ago)
What about associated closure operators?
4 / 24
Formal Concept Analysis
‘‚F‡illeD IWVP“D ‘fFq—nter —nd ‚F‡illeD IWWW“
GD — set of o˜je™ts
MD — set of —ttri˜utes
rel—tion I ⊆ G × M su™h th—t (g, m) ∈ I if —nd only if the o˜je™t g h—s
the —ttri˜ute mF
K := (G, M, I) is — form—l ™ontextF
heriv—tion oper—torsX A ⊆ GD B ⊆ M
A
def
= {m ∈ M | gIm for all g ∈ A}, B
def
= {g ∈ G | gIm for all m ∈ B}
e form—l ™on™ept is — p—ir (A, B)X A ⊆ G, B ⊆ M, A = B, and B = A.
A is the extent —nd B is the intent of the ™on™ept (A, B)F
„he ™on™eptsD ordered ˜y (A1, B1) ≥ (A2, B2) ⇐⇒ A1 ⊇ A2
@B2 ⊇ B1A
form — ™omplete l—tti™eD ™—lled the ™on™ept l—tti™e B(G, M, I)F
(·) is — ™losure oper—tor @idempotentD extensiveD monotoneA
5 / 24
Road Map
1 Formal Concept Analysis (FCA)
2 Triadic FCA
3 Towards a closure operator for 3-adic case
6 / 24
Triadic Formal Concept Analysis
‘pFvehm—n 8 ‚F‡illeD IWWS“
Denition 1
Triadic context K = (G, M, B, I) consists of set G (objects), M
(attributes), B (conditions) and ternary relations
Y ⊆ G × M × B. Triple (g, m, b) ∈ Y means that the object g
has the attribute m under the condition b.
Denition 2
(Formal) triconcept of K is a triple (X, Y, Z) which is maximal
w.r.t. its components inclusion, i.e. X ⊆ G, Y ⊆ M, Z ⊆ B è
X × Y × Z ⊆ Y
7 / 24
Triadic Formal Concept Analysis
‘pFvehm—n 8 ‚F‡illeD IWWS“
For convenience, a triadic context is denoted by
K = (X1, X2, X3, Y ). A triadic context K = (X1, X2, X3, Y )
gives rise to the following dyadic contexts
K(1) = (X1, X2 × X3, Y (1)), K(2) = (X2, X1 × X3, Y (2)), and
K(3) = (X3, X1 × X2, Y (3)) where
gY (1)(m, b) :⇔ mY (2)(g, b) :⇔ bY (3)(g, m) :⇔ (g, m, b) ∈ Y .
The derivation operators (primes or concept-forming operators)
induced by K(i) are denoted by (.)(i). For each induced dyadic
context we have two kinds of such derivation operators.
8 / 24
Triadic Formal Concept Analysis
‘pFvehm—n 8 ‚F‡illeD IWWS“
That is, for {i, j, k} = {1, 2, 3} with j  k and for Z ⊆ Xi and
W ⊆ Xj × Xk, the (i)-derivation operators are dened by:
Z → Z(i)
= {(xj, xk) ∈ Xj × Xk|xi, xj, xk —re rel—ted ˜y ‰ for —ll xi ∈ Z},
W → W(i)
= {xi ∈ Xi|xi, xj, xk —re rel—ted ˜y ‰ for —ll (xj, xk) ∈ W}.
Formally, a triadic concept of a triadic context
K = (X1, X2, X3, Y ) is a triple (A1, A2, A3) of
A1 ⊆ X1, A2 ⊆ X2, A3 ⊆ X3 such that for every
{i, j, k} = {1, 2, 3} with j  k we have (Aj × Ak)(i) = Ai.
9 / 24
Triadic Formal Concept Analysis
‘pFvehm—n 8 ‚F‡illeD IWWS“
The set of all triadic concepts of K = (X1, X2, X3, Y ) is denoted
by T(K).
There is a quasiorder i for each i ∈ {1, 2, 3} and its
corresponding equivalence relation ∼i is dened by
(A1, A2, A3) i (B1, B2, B3) :⇐⇒ Ai ⊆ Bi and
(A1, A2, A3) ∼i (B1, B2, B3) :⇐⇒ Ai = Bi.
These quasiorders satisfy the antiordinal dependencies: For
{i, j, k} = {1, 2, 3} and all triconcepts (A1, A2, A3) and
(B1, B2, B3) from T(K) it holds that
(A1, A2, A3) i (B1, B2, B3) and (A1, A2, A3) j (B1, B2, B3)
imply (A1, A2, A3) k (B1, B2, B3).
10 / 24
Folksonomy
‘„F v—n der ‡—lD PHHR“
he(nition
e qu—druple (U, T, R, Y ) is ™—lled folksonomyD where U is — set of usersD T is
— set of t—gsD R is — set of resour™esD —nd Y ⊆ U × T × R.
e triple (u, t, r) ∈ Y denotes th—t the user u —ssigns the t—g t to the
resour™e rF
11 / 24
Folksonomy example
st is inspired ˜y fi˜sonomy @httpXGG˜i˜sonomyForgAF
12 / 24
TRIAS
‘t¤—s™hke et —lFD PHHT“
Trias is a method for nding triadic formal concepts, that are
closed 3-sets. Triadic formal concepts can be interpreted as
absolutely dense triclusters.
NextClosure algorithm enumerates all formal concepts of
the dyadic context in their lexicographical order
Trias is a NextClosure extension to the triadic case
Minimal support constraints are added (triclusters with too
small extent, intent or modus are skipped)
13 / 24
TRIAS
sgn—tov et —lFX „ri—di™ porm—l gon™ept en—lysis —nd tri™lusteringX se—r™hing for
optim—l p—tternsF w—™hine ve—rning @PHISA
righ el—psed time
„oo l—rge num˜er of sm—ll wellEinterpreted tri™lusters @tri™on™eptsA
ix—mples of the tri™on™epts for the swhf ™ontextX
1 {„he €rin™ess fride @IWVUAD €ir—tes of the g—ri˜˜e—nX „he gurse of
the fl—™k €e—rl @PHHQA}D {€ir—te}D {p—nt—syD edventure}
2 {€l—toon @IWVTAD vetters from swo tim— @PHHTA}D {f—ttle}D
{hr—m—D‡—r}
3 {† for †endett— @PHHSA}D {p—s™istD „erroristD qovernmentD ƒe™ret
€oli™e D
pight}D {e™tionD ƒ™iEpiD „hriller}
14 / 24
Trisets
Denition
(Trabelsi et al., 2012) Let K = (G, M, B, I) be a formal
tricontext. A triple (X, Y, Z) is called a triset of K i
X × Y × Z ⊆ I.
Note that Cerf et al.(2009) dene a triset of K dierently:
X × Y × Z ∈ 2G × 2M × 2B.
For trisets t1 = (A1, B1, C1) and t2 = (A2, B2, C2), t1 t2
means that A1 × B1 × C1 ⊆ A2 × B2 × C2, i.e. every triple
(a, b, c) ∈ (A1, B1, C1) is in (A2, B2, C2).
15 / 24
Triset-based concept forming operator
Denition
(Trabelsi et al., 2012) Let S = (X, Y, Z) be a tri-set of
K = (G, M, B, I ⊆ G × M × B). The mapping
h : 2G × 2M × 2B ∩ 2I → 2G × 2M × 2B is dened as follows:
h(S) = {(U, V, W) | U = {g ∈ G | ∀m ∈ Y, ∀b ∈ Z : (g, m, b) ∈
I}, V = {m ∈ M | ∀g ∈ U, ∀b ∈ Z : (g, m, b) ∈ Y },
W = {b ∈ B | ∀g ∈ U, ∀m ∈ V : (g, m, b) ∈ Y }
Note that every triconcept is a maximal or closed triset, i.e. a
triset that cannot be extended by triples from I being a triset.
16 / 24
Triset-based concept forming operator
Proposition
h(·) is extensive and idempotent by on
T = {t | t is — triset of K} = {(X, Y, Z) ∈ 2G
× 2M
× 2B
| (X, Y, Z) ⊆ I}
and every xpoint f of h (i.e. h(f) = f) is a triconcept of K.
Proof. One can nd the proof of extensivity and idempotency in
(Trabelsi et al, 2012). It is easy to see that every formal
triconcept is a xpoint of h(·) and every triset (X, Y, Z) is
transformed by h(·) to the triconcept
((Y ×Z)(1), ((Y ×Z)(1)×Z)(2), ((Y ×Z)(1)×((Y ×Z)(1)×Z)(2))(3).
Indeed, all formal triconcepts should be listed since a triset is
allowed to be a triple with at least one component being ∅.
17 / 24
Triset-based concept forming operator
Theorem
For a given tricontext K = (G, M, B, I ⊆ G × M × B) and its
associated triset system
T = {(X, Y, Z) ∈ 2G × 2M × 2B | (X, Y, Z) ⊆ I} operator h is
not monotone w.r.t. .
Proof. To construct a violating example, one needs two dierent
triconcepts with the same extent, c1 = (X, Y1, Z1) and
c2 = (X, Y2, Z2) of K such that Y1 ⊂ Y2 and Z1 ⊃ Z2.
Consider the tri-set s = (X, Y1, Z2):
s c1 ⇒ h(s) = c2 h(c1) = c1.
18 / 24
Triset-based concept forming operator
Example
For the tricontext in Figure 2, the violating example for
monotonicity of h(·) is as follows:
x = ({u1, u2}, {t1}, {r1}) y = ({u1, u2}, {t1}, {r1, r2}) ⇒
h(x)) = ({u1, u2}, {t1, t2}, {r1}) h(y) = ({u1, u2}, {t1}, {r1, r2}).
t1 t2 t3
u1 × ×
u2 × ×
u3
r1
t1 t2 t3
u1 ×
u2 ×
u3
r2
Figure: A small example with Bibsonomy data
19 / 24
Switching generators
Denition
Let K = (K1, K2, K3, I) be a triadic formal context. A triset S
is called a (maximal) switching generator of the context K i
this is a non-empty component-wise intersection of c1 and c2 ,
where c1 and c2 are concepts of K.
20 / 24
Triset-based concept forming operators
For tricontext K = (K1, K2, K3, I) we consider a family of
operators
{σijk|σijk : 2K1
× 2K2
× 2K3
→ 2K1
× 2K2
× 2K3
su™h th—t
σijk : (X1, X2, X3) → (Y1, Y2, Y3), where
Yi = (Xj×Xk)(i)
, Yj = (Yi×Xk)(j)
, Yk = (Yi×Yj)(k)
D —nd {i, j, k} = {1, 2, 3}}.
The cardinality of the family is 3! = 6 and n! for its n-ary case
generalisation.
21 / 24
Triset-based concept forming operators
Proposition
Operators σijk(·) are not commutative, i.e.
σijk(σlmn(·)) = σlmn(σijk(·)), where (i, j, k) = (l, m, n) and
{i, j, k} = {1, 2, 3}.
€roofF gonsider — tri™ontext given ˜elowF
t1 t2 t3 t4
u1 × × × ×
u2 × × ×
u3 × × ×
u4 ×
r1
t1 t2 t3 t4
u1 × × ×
u2 × × ×
u3 × ×
u4
r2
t1 t2 t3 t4
u1 ×
u2
u3
u4
r3
„he system of —ll swit™hing gener—tors S ™ont—ins s1 = {u1, t4, r1} —nd
s2 = {u1, u2, t3, t4, r1}F
s1 proves th—t σi__(·) = σj__(·) = σk__(·) —nd
s2 proves th—t σijk(·) = σikj(·) for {i, j, k} = {1, 2, 3}F
„he f—™t th—t σlmnσijk(·) = σijk(·) proves the propositionF
22 / 24
Triset-based concept forming operators
Theorem
For K = (K1, K2, K3, I) and the associated triset system T
there is no an associated closure operator in case there exist at
least two concepts c1 = (X1, Y1, Z1) and c2 = (X1, Y2, Z2) such
that they have the common non-empty maximal switching
generator s, i.e. the intersection of the unions of their
componets, respectively, is not an empty triset.
Proof. Let σ be a closure operator for K. Since s c1 and
s c2 then σ(s) should result in ci which is either c1 or c2 (or
one of other concepts ck with s ck if any exist). So let
σ(s) = ci and consider s cj; it implies that
σ(s) = ci σ(cj) = cj for i = j, and {i, j} = {1, 2}.
Contradiction.
23 / 24
For more details see https://arxiv.org/abs/1602.07267
Thank you!
24 / 24

More Related Content

PDF
8 fixed point theorem in complete fuzzy metric space 8 megha shrivastava
PPT
Integration
PDF
Ijmet 10 01_046
DOCX
Integration SPM
PDF
Chapter 9- Differentiation Add Maths Form 4 SPM
PPT
2 3 Bzca5e
PPT
Integration and its basic rules and function.
8 fixed point theorem in complete fuzzy metric space 8 megha shrivastava
Integration
Ijmet 10 01_046
Integration SPM
Chapter 9- Differentiation Add Maths Form 4 SPM
2 3 Bzca5e
Integration and its basic rules and function.

What's hot (17)

PDF
Finite elements : basis functions
PDF
ISI MSQE Entrance Question Paper (2009)
PPTX
Indefinite Integral
PDF
modul 2 add maths
PDF
Bc4103338340
PDF
Additional Mathematics form 4 (formula)
PPT
Application of derivatives 2 maxima and minima
PDF
Fixed point theorems for four mappings in fuzzy metric space using implicit r...
PDF
Extension of Some Common Fixed Point Theorems using Compatible Mappings in Fu...
PDF
Stochastic Schrödinger equations
PDF
ISI MSQE Entrance Question Paper (2011)
PPT
Solution 3
PDF
On the solvability of a system of forward-backward linear equations with unbo...
PDF
ISI MSQE Entrance Question Paper (2012)
PPTX
Lesson 10 techniques of integration
PDF
Ponencia en el XVIII CCM
PPT
Matdis 3.4
Finite elements : basis functions
ISI MSQE Entrance Question Paper (2009)
Indefinite Integral
modul 2 add maths
Bc4103338340
Additional Mathematics form 4 (formula)
Application of derivatives 2 maxima and minima
Fixed point theorems for four mappings in fuzzy metric space using implicit r...
Extension of Some Common Fixed Point Theorems using Compatible Mappings in Fu...
Stochastic Schrödinger equations
ISI MSQE Entrance Question Paper (2011)
Solution 3
On the solvability of a system of forward-backward linear equations with unbo...
ISI MSQE Entrance Question Paper (2012)
Lesson 10 techniques of integration
Ponencia en el XVIII CCM
Matdis 3.4
Ad

Viewers also liked (18)

PDF
Sequence mining
PDF
Поиск частых множеств признаков (товаров) и ассоциативные правила
PDF
Experimental Economics and Machine Learning workshop
PDF
A lattice-based consensus clustering
PPTX
NIPS 2016, Tensor-Learn@NIPS, and IEEE ICDM 2016
PDF
Pattern-based classification of demographic sequences
PDF
Context-Aware Recommender System Based on Boolean Matrix Factorisation
PDF
Введение в рекомендательные системы. 3 case-study без NetFlix.
PDF
Pattern Mining and Machine Learning for Demographic Sequences
PDF
Putting OAC-triclustering on MapReduce
PDF
A lattice-based consensus clustering
PPTX
AIST 2016 Opening Slides
PDF
Searching for optimal patterns in Boolean tensors
PDF
A One-Pass Triclustering Approach: Is There any Room for Big Data?
PPTX
RAPS: A Recommender Algorithm Based on Pattern Structures
PPTX
Boolean matrix factorisation for collaborative filtering
PDF
Intro to Data Mining and Machine Learning
PPTX
Online Recommender System for Radio Station Hosting: Experimental Results Rev...
Sequence mining
Поиск частых множеств признаков (товаров) и ассоциативные правила
Experimental Economics and Machine Learning workshop
A lattice-based consensus clustering
NIPS 2016, Tensor-Learn@NIPS, and IEEE ICDM 2016
Pattern-based classification of demographic sequences
Context-Aware Recommender System Based on Boolean Matrix Factorisation
Введение в рекомендательные системы. 3 case-study без NetFlix.
Pattern Mining and Machine Learning for Demographic Sequences
Putting OAC-triclustering on MapReduce
A lattice-based consensus clustering
AIST 2016 Opening Slides
Searching for optimal patterns in Boolean tensors
A One-Pass Triclustering Approach: Is There any Room for Big Data?
RAPS: A Recommender Algorithm Based on Pattern Structures
Boolean matrix factorisation for collaborative filtering
Intro to Data Mining and Machine Learning
Online Recommender System for Radio Station Hosting: Experimental Results Rev...
Ad

Similar to On the Family of Concept Forming Operators in Polyadic FCA (13)

PPTX
From Sensing to Decision
PDF
MINING TRIADIC ASSOCIATION RULES
PDF
MINING TRIADIC ASSOCIATION RULES
PDF
Existence and Uniqueness of Algebraic Closure
PDF
Assignment#16
PDF
Fuzzy formal concept analysis: Approaches, applications and issues
PDF
Formal Concept Analysis
PDF
Navigating and Exploring RDF Data using Formal Concept Analysis
PDF
Extracting biclusters of similar values with Triadic Concept Analysis
PDF
DBMS 11 | Design Theory [Normalization 1]
PDF
Interpretable Concept-Based Classification with Shapley Values
PPT
test
PPT
7. Relational Database Design in DBMS
From Sensing to Decision
MINING TRIADIC ASSOCIATION RULES
MINING TRIADIC ASSOCIATION RULES
Existence and Uniqueness of Algebraic Closure
Assignment#16
Fuzzy formal concept analysis: Approaches, applications and issues
Formal Concept Analysis
Navigating and Exploring RDF Data using Formal Concept Analysis
Extracting biclusters of similar values with Triadic Concept Analysis
DBMS 11 | Design Theory [Normalization 1]
Interpretable Concept-Based Classification with Shapley Values
test
7. Relational Database Design in DBMS

More from Dmitrii Ignatov (10)

PPTX
AIST2019 – opening slides
PDF
Turning Krimp into a Triclustering Technique on Sets of Attribute-Condition P...
PDF
Personal Experiences of Publishing with Springer from both Editor and Author ...
PPTX
Aist2014
PDF
Social Learning in Networks: Extraction Deterministic Rules
PPTX
Orpailleur -- triclustering talk
PDF
CoClus ICDM Workshop talk
PPTX
Pseudo-triclustering
PPTX
Radio recommender system for FMHost
PDF
CrowDM system
AIST2019 – opening slides
Turning Krimp into a Triclustering Technique on Sets of Attribute-Condition P...
Personal Experiences of Publishing with Springer from both Editor and Author ...
Aist2014
Social Learning in Networks: Extraction Deterministic Rules
Orpailleur -- triclustering talk
CoClus ICDM Workshop talk
Pseudo-triclustering
Radio recommender system for FMHost
CrowDM system

Recently uploaded (20)

PDF
Worlds Next Door: A Candidate Giant Planet Imaged in the Habitable Zone of ↵ ...
PPTX
perinatal infections 2-171220190027.pptx
PPT
1. INTRODUCTION TO EPIDEMIOLOGY.pptx for community medicine
PPTX
Seminar Hypertension and Kidney diseases.pptx
PPT
THE CELL THEORY AND ITS FUNDAMENTALS AND USE
PDF
Unit 5 Preparations, Reactions, Properties and Isomersim of Organic Compounds...
PPTX
Substance Disorders- part different drugs change body
PPTX
POULTRY PRODUCTION AND MANAGEMENTNNN.pptx
PDF
S2 SOIL BY TR. OKION.pdf based on the new lower secondary curriculum
PDF
CHAPTER 3 Cell Structures and Their Functions Lecture Outline.pdf
PPT
Mutation in dna of bacteria and repairss
PPT
veterinary parasitology ````````````.ppt
PPT
Presentation of a Romanian Institutee 2.
PDF
Communicating Health Policies to Diverse Populations (www.kiu.ac.ug)
PDF
GROUP 2 ORIGINAL PPT. pdf Hhfiwhwifhww0ojuwoadwsfjofjwsofjw
PDF
lecture 2026 of Sjogren's syndrome l .pdf
PDF
Looking into the jet cone of the neutrino-associated very high-energy blazar ...
PDF
Assessment of environmental effects of quarrying in Kitengela subcountyof Kaj...
PDF
CHAPTER 2 The Chemical Basis of Life Lecture Outline.pdf
PPTX
A powerpoint on colorectal cancer with brief background
Worlds Next Door: A Candidate Giant Planet Imaged in the Habitable Zone of ↵ ...
perinatal infections 2-171220190027.pptx
1. INTRODUCTION TO EPIDEMIOLOGY.pptx for community medicine
Seminar Hypertension and Kidney diseases.pptx
THE CELL THEORY AND ITS FUNDAMENTALS AND USE
Unit 5 Preparations, Reactions, Properties and Isomersim of Organic Compounds...
Substance Disorders- part different drugs change body
POULTRY PRODUCTION AND MANAGEMENTNNN.pptx
S2 SOIL BY TR. OKION.pdf based on the new lower secondary curriculum
CHAPTER 3 Cell Structures and Their Functions Lecture Outline.pdf
Mutation in dna of bacteria and repairss
veterinary parasitology ````````````.ppt
Presentation of a Romanian Institutee 2.
Communicating Health Policies to Diverse Populations (www.kiu.ac.ug)
GROUP 2 ORIGINAL PPT. pdf Hhfiwhwifhww0ojuwoadwsfjofjwsofjw
lecture 2026 of Sjogren's syndrome l .pdf
Looking into the jet cone of the neutrino-associated very high-energy blazar ...
Assessment of environmental effects of quarrying in Kitengela subcountyof Kaj...
CHAPTER 2 The Chemical Basis of Life Lecture Outline.pdf
A powerpoint on colorectal cancer with brief background

On the Family of Concept Forming Operators in Polyadic FCA

  • 1. On the Family of Concept Forming Operators in Polyadic FCA Dmitry I. Ignatov National Research University Higher School of Economics, Moscow, Russia Department of Data Analysis and Machine Intelligence, Faculty of Computer Science & Laboratory of Intelligent Systems and Structural Analysis FCA4KD Moscow 2017 1 / 24
  • 2. Outline 1 Formal Concept Analysis (FCA) 2 Triadic FCA 3 Towards a closure operator for 3-adic case 2 / 24
  • 3. Motivation A large amount of structured and unstructured data generates triadic relations. E.g. folksonomy is a set of triples (user, object, tag) Concrete examples: Bibsonomy.org (user, bookmark, tag) Social networking sites (user, group, interest) Delicious (user, link, tag) Figure: Folksonomy as a graph. 3 / 24
  • 4. Main research question Are there concept-forming operators for n-ry case? (answered more than two decades ago) What about associated closure operators? 4 / 24
  • 5. Formal Concept Analysis ‘‚F‡illeD IWVP“D ‘fFq—nter —nd ‚F‡illeD IWWW“ GD — set of o˜je™ts MD — set of —ttri˜utes rel—tion I ⊆ G × M su™h th—t (g, m) ∈ I if —nd only if the o˜je™t g h—s the —ttri˜ute mF K := (G, M, I) is — form—l ™ontextF heriv—tion oper—torsX A ⊆ GD B ⊆ M A def = {m ∈ M | gIm for all g ∈ A}, B def = {g ∈ G | gIm for all m ∈ B} e form—l ™on™ept is — p—ir (A, B)X A ⊆ G, B ⊆ M, A = B, and B = A. A is the extent —nd B is the intent of the ™on™ept (A, B)F „he ™on™eptsD ordered ˜y (A1, B1) ≥ (A2, B2) ⇐⇒ A1 ⊇ A2 @B2 ⊇ B1A form — ™omplete l—tti™eD ™—lled the ™on™ept l—tti™e B(G, M, I)F (·) is — ™losure oper—tor @idempotentD extensiveD monotoneA 5 / 24
  • 6. Road Map 1 Formal Concept Analysis (FCA) 2 Triadic FCA 3 Towards a closure operator for 3-adic case 6 / 24
  • 7. Triadic Formal Concept Analysis ‘pFvehm—n 8 ‚F‡illeD IWWS“ Denition 1 Triadic context K = (G, M, B, I) consists of set G (objects), M (attributes), B (conditions) and ternary relations Y ⊆ G × M × B. Triple (g, m, b) ∈ Y means that the object g has the attribute m under the condition b. Denition 2 (Formal) triconcept of K is a triple (X, Y, Z) which is maximal w.r.t. its components inclusion, i.e. X ⊆ G, Y ⊆ M, Z ⊆ B è X × Y × Z ⊆ Y 7 / 24
  • 8. Triadic Formal Concept Analysis ‘pFvehm—n 8 ‚F‡illeD IWWS“ For convenience, a triadic context is denoted by K = (X1, X2, X3, Y ). A triadic context K = (X1, X2, X3, Y ) gives rise to the following dyadic contexts K(1) = (X1, X2 × X3, Y (1)), K(2) = (X2, X1 × X3, Y (2)), and K(3) = (X3, X1 × X2, Y (3)) where gY (1)(m, b) :⇔ mY (2)(g, b) :⇔ bY (3)(g, m) :⇔ (g, m, b) ∈ Y . The derivation operators (primes or concept-forming operators) induced by K(i) are denoted by (.)(i). For each induced dyadic context we have two kinds of such derivation operators. 8 / 24
  • 9. Triadic Formal Concept Analysis ‘pFvehm—n 8 ‚F‡illeD IWWS“ That is, for {i, j, k} = {1, 2, 3} with j k and for Z ⊆ Xi and W ⊆ Xj × Xk, the (i)-derivation operators are dened by: Z → Z(i) = {(xj, xk) ∈ Xj × Xk|xi, xj, xk —re rel—ted ˜y ‰ for —ll xi ∈ Z}, W → W(i) = {xi ∈ Xi|xi, xj, xk —re rel—ted ˜y ‰ for —ll (xj, xk) ∈ W}. Formally, a triadic concept of a triadic context K = (X1, X2, X3, Y ) is a triple (A1, A2, A3) of A1 ⊆ X1, A2 ⊆ X2, A3 ⊆ X3 such that for every {i, j, k} = {1, 2, 3} with j k we have (Aj × Ak)(i) = Ai. 9 / 24
  • 10. Triadic Formal Concept Analysis ‘pFvehm—n 8 ‚F‡illeD IWWS“ The set of all triadic concepts of K = (X1, X2, X3, Y ) is denoted by T(K). There is a quasiorder i for each i ∈ {1, 2, 3} and its corresponding equivalence relation ∼i is dened by (A1, A2, A3) i (B1, B2, B3) :⇐⇒ Ai ⊆ Bi and (A1, A2, A3) ∼i (B1, B2, B3) :⇐⇒ Ai = Bi. These quasiorders satisfy the antiordinal dependencies: For {i, j, k} = {1, 2, 3} and all triconcepts (A1, A2, A3) and (B1, B2, B3) from T(K) it holds that (A1, A2, A3) i (B1, B2, B3) and (A1, A2, A3) j (B1, B2, B3) imply (A1, A2, A3) k (B1, B2, B3). 10 / 24
  • 11. Folksonomy ‘„F v—n der ‡—lD PHHR“ he(nition e qu—druple (U, T, R, Y ) is ™—lled folksonomyD where U is — set of usersD T is — set of t—gsD R is — set of resour™esD —nd Y ⊆ U × T × R. e triple (u, t, r) ∈ Y denotes th—t the user u —ssigns the t—g t to the resour™e rF 11 / 24
  • 12. Folksonomy example st is inspired ˜y fi˜sonomy @httpXGG˜i˜sonomyForgAF 12 / 24
  • 13. TRIAS ‘t¤—s™hke et —lFD PHHT“ Trias is a method for nding triadic formal concepts, that are closed 3-sets. Triadic formal concepts can be interpreted as absolutely dense triclusters. NextClosure algorithm enumerates all formal concepts of the dyadic context in their lexicographical order Trias is a NextClosure extension to the triadic case Minimal support constraints are added (triclusters with too small extent, intent or modus are skipped) 13 / 24
  • 14. TRIAS sgn—tov et —lFX „ri—di™ porm—l gon™ept en—lysis —nd tri™lusteringX se—r™hing for optim—l p—tternsF w—™hine ve—rning @PHISA righ el—psed time „oo l—rge num˜er of sm—ll wellEinterpreted tri™lusters @tri™on™eptsA ix—mples of the tri™on™epts for the swhf ™ontextX 1 {„he €rin™ess fride @IWVUAD €ir—tes of the g—ri˜˜e—nX „he gurse of the fl—™k €e—rl @PHHQA}D {€ir—te}D {p—nt—syD edventure} 2 {€l—toon @IWVTAD vetters from swo tim— @PHHTA}D {f—ttle}D {hr—m—D‡—r} 3 {† for †endett— @PHHSA}D {p—s™istD „erroristD qovernmentD ƒe™ret €oli™e D pight}D {e™tionD ƒ™iEpiD „hriller} 14 / 24
  • 15. Trisets Denition (Trabelsi et al., 2012) Let K = (G, M, B, I) be a formal tricontext. A triple (X, Y, Z) is called a triset of K i X × Y × Z ⊆ I. Note that Cerf et al.(2009) dene a triset of K dierently: X × Y × Z ∈ 2G × 2M × 2B. For trisets t1 = (A1, B1, C1) and t2 = (A2, B2, C2), t1 t2 means that A1 × B1 × C1 ⊆ A2 × B2 × C2, i.e. every triple (a, b, c) ∈ (A1, B1, C1) is in (A2, B2, C2). 15 / 24
  • 16. Triset-based concept forming operator Denition (Trabelsi et al., 2012) Let S = (X, Y, Z) be a tri-set of K = (G, M, B, I ⊆ G × M × B). The mapping h : 2G × 2M × 2B ∩ 2I → 2G × 2M × 2B is dened as follows: h(S) = {(U, V, W) | U = {g ∈ G | ∀m ∈ Y, ∀b ∈ Z : (g, m, b) ∈ I}, V = {m ∈ M | ∀g ∈ U, ∀b ∈ Z : (g, m, b) ∈ Y }, W = {b ∈ B | ∀g ∈ U, ∀m ∈ V : (g, m, b) ∈ Y } Note that every triconcept is a maximal or closed triset, i.e. a triset that cannot be extended by triples from I being a triset. 16 / 24
  • 17. Triset-based concept forming operator Proposition h(·) is extensive and idempotent by on T = {t | t is — triset of K} = {(X, Y, Z) ∈ 2G × 2M × 2B | (X, Y, Z) ⊆ I} and every xpoint f of h (i.e. h(f) = f) is a triconcept of K. Proof. One can nd the proof of extensivity and idempotency in (Trabelsi et al, 2012). It is easy to see that every formal triconcept is a xpoint of h(·) and every triset (X, Y, Z) is transformed by h(·) to the triconcept ((Y ×Z)(1), ((Y ×Z)(1)×Z)(2), ((Y ×Z)(1)×((Y ×Z)(1)×Z)(2))(3). Indeed, all formal triconcepts should be listed since a triset is allowed to be a triple with at least one component being ∅. 17 / 24
  • 18. Triset-based concept forming operator Theorem For a given tricontext K = (G, M, B, I ⊆ G × M × B) and its associated triset system T = {(X, Y, Z) ∈ 2G × 2M × 2B | (X, Y, Z) ⊆ I} operator h is not monotone w.r.t. . Proof. To construct a violating example, one needs two dierent triconcepts with the same extent, c1 = (X, Y1, Z1) and c2 = (X, Y2, Z2) of K such that Y1 ⊂ Y2 and Z1 ⊃ Z2. Consider the tri-set s = (X, Y1, Z2): s c1 ⇒ h(s) = c2 h(c1) = c1. 18 / 24
  • 19. Triset-based concept forming operator Example For the tricontext in Figure 2, the violating example for monotonicity of h(·) is as follows: x = ({u1, u2}, {t1}, {r1}) y = ({u1, u2}, {t1}, {r1, r2}) ⇒ h(x)) = ({u1, u2}, {t1, t2}, {r1}) h(y) = ({u1, u2}, {t1}, {r1, r2}). t1 t2 t3 u1 × × u2 × × u3 r1 t1 t2 t3 u1 × u2 × u3 r2 Figure: A small example with Bibsonomy data 19 / 24
  • 20. Switching generators Denition Let K = (K1, K2, K3, I) be a triadic formal context. A triset S is called a (maximal) switching generator of the context K i this is a non-empty component-wise intersection of c1 and c2 , where c1 and c2 are concepts of K. 20 / 24
  • 21. Triset-based concept forming operators For tricontext K = (K1, K2, K3, I) we consider a family of operators {σijk|σijk : 2K1 × 2K2 × 2K3 → 2K1 × 2K2 × 2K3 su™h th—t σijk : (X1, X2, X3) → (Y1, Y2, Y3), where Yi = (Xj×Xk)(i) , Yj = (Yi×Xk)(j) , Yk = (Yi×Yj)(k) D —nd {i, j, k} = {1, 2, 3}}. The cardinality of the family is 3! = 6 and n! for its n-ary case generalisation. 21 / 24
  • 22. Triset-based concept forming operators Proposition Operators σijk(·) are not commutative, i.e. σijk(σlmn(·)) = σlmn(σijk(·)), where (i, j, k) = (l, m, n) and {i, j, k} = {1, 2, 3}. €roofF gonsider — tri™ontext given ˜elowF t1 t2 t3 t4 u1 × × × × u2 × × × u3 × × × u4 × r1 t1 t2 t3 t4 u1 × × × u2 × × × u3 × × u4 r2 t1 t2 t3 t4 u1 × u2 u3 u4 r3 „he system of —ll swit™hing gener—tors S ™ont—ins s1 = {u1, t4, r1} —nd s2 = {u1, u2, t3, t4, r1}F s1 proves th—t σi__(·) = σj__(·) = σk__(·) —nd s2 proves th—t σijk(·) = σikj(·) for {i, j, k} = {1, 2, 3}F „he f—™t th—t σlmnσijk(·) = σijk(·) proves the propositionF 22 / 24
  • 23. Triset-based concept forming operators Theorem For K = (K1, K2, K3, I) and the associated triset system T there is no an associated closure operator in case there exist at least two concepts c1 = (X1, Y1, Z1) and c2 = (X1, Y2, Z2) such that they have the common non-empty maximal switching generator s, i.e. the intersection of the unions of their componets, respectively, is not an empty triset. Proof. Let σ be a closure operator for K. Since s c1 and s c2 then σ(s) should result in ci which is either c1 or c2 (or one of other concepts ck with s ck if any exist). So let σ(s) = ci and consider s cj; it implies that σ(s) = ci σ(cj) = cj for i = j, and {i, j} = {1, 2}. Contradiction. 23 / 24
  • 24. For more details see https://arxiv.org/abs/1602.07267 Thank you! 24 / 24