Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Boolean Relation Equations in Data Analysis

c Ivan Stankovi , Jelena Ignjatovi and Miroslav Ciri c c


University of Ni, Faculty of Sciences and Mathematics, Ni, Serbia s s ivan.stankovic@pmf.edu.rs, jelena.ignjatovic@pmf.edu.rs, miroslav.ciric@pmf.edu.rs

AbstractIn this paper we discuss certain systems of Boolean relation equations with two unknown relations. We show that these systems have the greatest solutions, which are pairs of quasi-orders or equivalences, we provide efcient algorithms for computing these greatest solutions, and we show how we can use these equivalences and quasi-orders in the reduction of data represented by Boolean data tables.

I. I NTRODUCTION The recent advances in the acquisition, storage and transmission of data have dramatically increased the need for tools that effectively support users in retrieving, understanding and mining the information and knowledge contained in such data. In contrast to the traditional statistical analysis, some contemporary approaches to data analysis, such as concept data analysis and social network analysis, are focused on the recognition and generalization of structural similarities from the data description. A method that is often used in such structural analysis is data reduction, a process of transforming raw data into a more condensed form without losing signicant semantic information. One of the techniques of data reduction is the blockmodeling, an approach widely used in the social network analysis, where large and complex social networks are mapped into simpler structures, called blockmodel images. Blockmodel images are viewed as structural summaries of these large and complex networks. The structure depicted in a blockmodel image can be understood as the fundamental structure of the network and the considered network is an instantiation of the fundamental structure. The key role in blockmodeling play some types of equivalences, like structural and regular equivalences, which in a certain sense simultaneously partition nodes and edges of the network (cf. [1], [4], [12][14]). The data are often represented by a Boolean binary relation R between non-empty sets U and V , where U is understood as a set of objects and V a set of their attributes, as in the concept data analysis and relational database research, or U is viewed as a set of actors and V a set of related events, as when dealing with two-mode networks in the social network analysis. The purpose of this paper is to discuss the reduction of the system (U, V, R) using pairs of relations that are solutions to certain systems of Boolean relation equations. Namely, we study the system of relation equations X R = R and R Y = R, and the equation X R = R Y , where X is an unknown relation on U , Y is an unknown relation on V and is the usual composition of relations. Their solutions are pairs of relations on U and V and can be ordered coordinatewise. We show that both the system and the equation have the greatest solutions, which are pairs of quasi-orders, and we provide algorithms for computing the greatest solutions which satisfy some
Research supported by Ministry Education and Science, Republic of Serbia, Grant No. 174013

pre-specied conditions given by a pair of relations on U and V . We also discuss some related systems of relation equations whose greatest solutions are pairs of equivalences and give algorithms that compute these equivalences. If the system (U, V, R) is viewed as a two-mode social network, we show that pairs of equivalences that are solutions to the discussed systems of relation equations can be understood as pairs of structural and regular equivalences on this network. We study blockmodeling of a two-mode social network using not only structural and regular equivalences, but also related quasi-orders. It is worth noting that the considered systems of relation equations originate from systems of fuzzy relation inequalities and equations which have been studied in [16][18]. They have important applications in the state reduction, simulation and bisimulation of fuzzy and nondeterministic automata (cf. [8] [11]), and in the positional analysis and blockmodeling of onemode social networks (cf. [16], [17]). II. P RELIMINARIES Let U and V be non-empty sets. Any subset R U V is called a relation from U to V , and equality, inclusion, union and intersection of relations from U to V are dened as for subsets of U V . The set of all relations from U to V will be denoted by R(U, V ). The converse (inverse or transpose) of a relation R U V is a relation R1 V U dened by (v, u) R1 if and only if (u, v) R, for all (u, v) U V . If U = V , that is, if R U U , then R is called a relation on U . The set of all relations on U will be denoted by R(U ). As known, R(U, V ) and R(U ) are complete lattices with respect to the inclusion of relations and the operations of union and intersection of relations. By U we denote the equality relation on U . For non-empty sets U , V and W , and relations R U V and S V W , the composition of R and S is a relation R S U W dened by (u, w) (RS) (v V ) (u, v) R & (v, w) S , (1) for all u U and w W . For non-empty sets U , V , W and Z, and arbitrary relations R U V , S, S1 , S2 , Si V W and T W Z, where i I, the following is true: (R S) T = R (S T ), S1 S2 R S1 R S2 & S1 T S2 T, R
iI

(2) (3) (Si T ) (4)


iI

Si =
iI

(R Si ), (R S)1 = S S1 S2

Si T =
iI 1

R1 ,
1 S2 .

(5) (6)

1 S1

Therefore, parentheses in (2) can be omitted. Let U and V be non-empty sets and R U V an arbitrary relation. For each u U , the set uR = {v V | (u, v) R} is called the R-afterset of u, and symmetrically, for every v V , the set Rv = {u U | (u, v) R} is called the R-foreset of v. Note that, despite the notation, the converse relation R1 is not an inverse of the relation R U V in the sense of composition of relations, i.e., R R1 and R1 R are not the equality relations on U and V in general. Note also that if U , V and W are nite sets with |U | = k, |V | = m and |W | = n, then R U V and S V W can be treated as k m and m n Boolean matrices, and R S is their matrix product. In addition, R-aftersets are actually row vectors, and R-foresets are column vectors of the Boolean matrix R. Recall that an equivalence on a set U is any reexive, symmetric and transitive relation on U . Let E be an equivalence on a set U . By Eu we denote the equivalence class of an element u U with respect to E, i.e., Eu = {v U | (u, v) E}. The set of all equivalence classes of E is denoted by U/E and called the quotient or factor set of U with respect to E. Recall also that a quasi-order on a set U is any reexive and transitive (not necessarily symmetric) relation on U . If Q is a quasi-order on U , then the relation EQ = QQ1 is an equivalence on U , which is called the natural equivalence of Q. Denote by U/Q the set of all Q-aftersets, and by U \Q the set of all Q-foresets. For arbitrary u, v U we have that uQ = vQ is equivalent to (u, v) EQ , and also, Qu = Qv is equivalent to (u, v) EQ . This implies that the sets U/Q, U \Q and U/EQ have the same cardinality. Cardinality of these sets is called the index of Q and denoted by ind(Q). If Q is an equivalence, then for every u U we have that the Q-afterset of u equals the Q-foreset of u, and this is precisely the related equivalence class of u. Thus, the index of an equivalence is cardinality of its quotient set. It can be easily shown that if P and Q are two quasi orders, then P Q implies ind(Q) ind(P ). Let Q be a quasi-order on a non-empty set U . An important thing that should also be noted is that each Q-afterset is an up-set, and each Q-foreset is a down-set in the quasi-ordered set (U, Q). In other words, for arbitrary u, v U , by u and (u, v) Q it follows v , and by u and (v, u) Q it follows v . For non-empty sets U and V and a relation R U V we R R dene relations EU R(U ) and EV R(V ) by
R (u1 , u2 ) EU u1 R = u2 R,

est xed point and the greatest post-xed point of . The same theorem also says that the greatest xed point and the greatest post-xed point of are equal. Moreover, the set of all postxed points of is a complete join-sublattice of P . For undened notions and notation we refer to [2], [3], [5], [14], [15], [19]. III. M AIN RESULTS Let U and V be non-empty nite sets and R, S R(U, V ). The left residual of S by R is a relation S/R R(U ) deined by (u1 , u2 ) S/R u2 R u1 S, (9) for all u1 , u2 U , and the right residual of S by R is a relation R\S R(V ) dened by (v1 , v2 ) R\S Rv1 Sv2 , for all v1 , v2 V . We can easily check that X R S X S/R, R Y S Y R\S, (11) (12) (10)

for any X R(U ) and Y R(V ). If X and Y are considered as unknowns, then solutions to relation inequalities X R S and R Y S form principal down-sets of R(U ) and R(V ) generated by S/R and R\S, respectively. In particular, R/R and R\R are the greatest solutions to inequalities X R R and R Y R. It is easy to see that R/R and R\R are quasiorders, and due to reexivity, they are the greatest solutions to relation equations X R = R, R Y = R, (13) (14)

respectively. In practice we often need solutions to the above equations that are equivalences. For this reason we also consider systems X R = R, R Y = R, X 1 R = R, RY
1

(15) (16)

= R.

(7) (8)

(v1 , v2 )

R EV

Rv1 = Rv2 ,

R R for all u1 , u2 U and v1 , v2 V . Evidently, EU and EV are equivalences, and they are respectively called the kernel and the co-kernel of R (cf. [6], [7]). Let P and Q be ordered sets. A function : P Q is said to be an isotone function if x y implies (x) (y), for all x, y P . For an ordered set P , a function : P P , and an element x P , we say that x is a xed point of if x = (x), and that x is a post-xed point of if x (x). If P is a complete lattice and is an isotone function, the well-known Knaster-Tarski xed point theorem (cf. [19, Theorems 12.1 and 12.2]) asserts that the sets of all xed points and all post-xed points of are complete lattices, and hence, there are the great-

The greatest solutions to (15) and (16) are respectively the natural equivalences of the quasi-orders R/R and R\R, i.e., the kernel and the co-kernel of R. Following the terminology from the social network analysis, a pair of quasi-orders which are respectively solutions to (13) and (14) will be called a pair of structural quasi-orders on the system (U, V, R), and a pair of equivalences which are respectively solutions to (15) and (16) will be called a pair of structural equivalences on this system. Evidently, the greatest pair of structural equivalences is the pair consisting of the kernel and co-kernel of R. The main subject of this paper is a more general equation X R = R Y, (17)

where R R(U, V ) is a given relation, and X and Y are unknowns which take values in R(U ) and R(V ), respectively. By a solution to (17) we mean any pair (J, K) R(U )R(V ) such that J R = RK, and J and K are called components of this solution. Note that R(U )R(V ) is a complete lattice with respect to an order dened by (J1 , K1 ) (J2 , K2 ) J1 J2 & K1 K2 ,

for all (J1 , K1 ), (J2 , K2 ) R(U ) R(V ), and thus, solutions to (17) can also be ordered. If (J1 , K1 ) (J2 , K2 ), we will say that (J1 , K1 ) is contained in (J2 , K2 ). In the analysis of equation (17) it is suitable to split this equation into two relation inequalities: X R R Y, R Y X R. (18) (19)

According to (11) and (12), inequalities (18) and (19) are respectively equivalent to X (R Y )/R, Y R\(X R). Dene a function of R(U ) R(V ) into itself by (J, K) = (R K)/R, R\(J R) , for each (J, K) R(U ) R(V ). We have that the following is true. Theorem 3.1: Equation (17) is equivalent to equation Z (Z), (23) (22) (20) (21)

Notice that the set of all solutions to (17) is a complete join-subsemilattice of R(U ) R(V ) (cf. [19, Theorem 12.1]). As we have seen, both components of the greatest solution (X, Y ) to equation (17) are quasi-orders. However, in practice we usually need solutions whose components are equivalences. It is easy to show that a pair consisting of the natural equivalences of X and Y is not necessarily a solution to equation (17). This means that the greatest equivalence which is a solution to equation (17) must be sought in some other way. Consider the system X R = R Y, X 1 R = R Y 1 . (24)

Evidently, a pair (J, K) is a solution to (24) if and only if both (J, K) and (J 1 , K 1 ) are solutions to (17). Dene now a function of R(U ) R(V ) into itself by (J, K) = [(R K)/R] [(R K 1 )/R]1 , [R\(J R)] [R\(J 1 R)]1 , (25) for each (J, K) R(U ) R(V ). If for any (J, K), (J1 , K1 ), (J2 , K2 ) R(U )R(V ) we set (J, K)1 = (J 1 , K 1 ), (J1 , K1 ) (J2 , K2 ) = (J1 J2 , K1 K2 ), then (25) can be written as (J, K) = (J, K) [(J 1 , K 1 )]1 . We have the following. Theorem 3.3: System (24) is equivalent to equation Z (Z), (27) (26)

where Z is an unknown which takes values in R(U ) R(V ). In addition, the function is isotone. Proof: As we have said earlier, equation (17) is equivalent to the conjunction of (18) and (19), i.e., to the conjunction of (20) and (21), and according to (22), the conjunction of (20) and (21) is equivalent to (23). It remains to prove that is an isotone function, which will be of crucial importance in future work. It is wellknown that J J R and K R K are isotone functions from R(U ) and R(V ) to R(U, V ), respectively. Moreover, directly from (9) and (10) it follows that S S/R and S R\S are isotone functions from R(U, V ) to R(U ) and R(V ), respectively. Hence, as compositions of isotone functions, J R\(J R) and K (R K)/R are also isotone functions from R(U ) to R(V ) and vice versa, and by (22) we conclude that is an isotone function. We are now ready to state and prove one of the main results of this paper. Theorem 3.2: The set of all solutions to equation (17) is a complete lattice, and therefore, this equation has the greatest solution. In addition, both components of the greatest solution to (17) are quasi-orders. Proof: According to Theorem 3.1, the set of all solutions to equation (17) is the set of all post-xed points of the function . Since R(U ) R(V ) is a complete lattice and is isotone, by the Knaster-Tarski xed point theorem [19, Theorem 12.2] we obtain that this set is a complete lattice and there is the greatest solution (X, Y ) to (17). Since the pair (U , V ) is a solution to (17), we obtain that (U , V ) (X, Y ), and consequently, X and Y are reexive relations. Furthermore, it is easy to verify that (X X, Y Y ) X, Y Y ) is a solution to (17), so (X (X, Y ), which and Y are transitive relations. This completes means that X the proof of the theorem.

where Z is an unknown which takes values in R(U ) R(V ). In addition, the function is isotone. Proof: According to the above remark and Theorem 3.1, a pair (J, K) R(U )R(V ) is a solution to (24) if and only if (J, K) (J 1 , K 1 ) (J, K) = (R K)/R, R\(J R) , (J 1 , K 1 ) = (R K 1 )/R, R\(J 1 R) , and (29) is obviously equivalent to (J, K) [(R K 1 )/R]1 , [R\(J 1 R)]1 . (30) (28) (29)

Now, the conjunction of (28) and (30) is equivalent to (J, K) [(R K)/R] [(R K 1 )/R]1 , [R\(J R)] [R\(J 1 R)]1 , that is, to (J, K) (J, K). Thus, we have proved that (24) is equivalent to (27). According to (26) and the fact that is isotone we immediately obtain that is also an isotone function. Now we prove a theorem analogous to Theorem 3.2. Theorem 3.4: The set of all solutions to system (24) is a complete lattice, and consequently, this system has the greatest solution.

In addition, both components of the greatest solution to (24) are equivalences. Proof: The same arguments that we have used in the proof of Theorem 3.2 conrm that the set of all solutions to system (24) is a complete lattice. Let (X, Y ) be the greatest solution to (24). As in the proof of Theorem 3.2 we show that X and Y are reexive and tran 1 , Y 1 ) is also a solution sitive. Moreover, we have that (X 1 , Y 1 ) (X, Y ), and we obtain that X to (24), whence (X are symmetric. Hence, X and Y are equivalences. and Y Following the terminology from the social network analysis, a pair of quasi-orders which is a solution to (17) will be called a pair of regular quasi-orders on the system (U, V, R), and a pair of equivalences which is a solution to (17), or to (24), will be called a pair of regular equivalences on this system. The following theorem provides an algorithm for computing the greatest solution to equation (17) contained in a given pair of relations. Theorem 3.5: Let (J, K) R(U ) R(V ) be a given pair of relations, and let {(Xk , Yk )}kN R(U ) R(V ) be a sequence dened as follows (X1 , Y1 ) = (J, K), (Xk+1 , Yk+1 ) = (Xk , Yk ) (Xk , Yk ), (31) (32)

If (J, K) is a pair of quasi-orders and (X, Y ) is the greatest solution to equation (17) contained in (J, K), by the same arguments used in the proof of Theorem 3.2 we can show that (X, Y ) is also a pair of quasi-orders. In the same way we can prove the following theorem, which gives an algorithm for computing the greatest solution to system (24) contained in a given pair of relations. Theorem 3.6: Let (J, K) R(U ) R(V ) be a given pair of relations, and let {(Xk , Yk )}kN R(U ) R(V ) be a sequence dened as follows (X1 , Y1 ) = (J J 1 , K K 1 ), (Xk+1 , Yk+1 ) = (Xk , Yk ) (Xk , Yk ), (33) (34)

for each k N. Then the following is true: (a) the sequence {(Xk , Yk )}kN is nite and descending, and there is the least k N such that (Xk , Yk ) = (Xk+1 , Yk+1 ); (b) the pair (Xk , Yk ) is the greatest solution to system (24) contained in (J, K). If (J, K) is a pair of equivalences and (X, Y ) is the greatest solution to system (24) contained in (J, K), then (X, Y ) is also a pair of equivalences. IV. R EDUCTION OF A RELATION Let U and V be non-empty sets, let R R(U, V ), and let E and F be arbitrary equivalences on U and V , respectively. We dene a relation R R(U/E, V /F ) by (, ) R (u )(v ) (u, v) R, (35)

for each k N. Then the following is true: (a) the sequence {(Xk , Yk )}kN is nite and descending, and there is the least k N such that (Xk , Yk ) = (Xk+1 , Yk+1 ); (b) the pair (Xk , Yk ) is the greatest solution to equation (17) contained in (J, K). Proof: (a) It is clear that the sequence {(Xk , Yk )}kN is descending. According to the starting assumption, U and V are nite sets, so R(U ) R(V ) is also nite and {(Xk , Yk )}kN is a nite sequence. Therefore, there exist k, m N such that (Xk , Yk ) = (Xk+m , Yk+m ), from which one can easily prove that (Xk , Yk ) = (Xk+1 , Yk+1 ). (b) By (Xk , Yk ) = (Xk+1 , Yk+1 ) and (32) it follows that (Xk , Yk ) = (Xk , Yk ) (Xk , Yk ) (Xk , Yk ),

for all U/E and V /F . Equivalently, the relation R can be dened in the following way. Theorem 4.1: In the above notation, the following is true: (, ) R (u )(v ) (u, v) E R F, (36) for all U/E and V /F . Proof: Consider arbitrary U/E and V /F . If (, ) R, then there are u and v such that (u , v ) R. Now, for arbitrary u and v we have that (u, u ) E and (v , v) F , and hence, (u, v) E R F . Conversely, let the right side of the equivalence (36) hold, and consider arbitrary u and v . Then we have that (u, v) E R F , so there are u U and v V such that (u, u ) E, (u , v ) R and (v , v) F . Therefore, u , v and (u , v ) R, whence (, ) R. In the case when (E, F ) is a pair of structural equivalences, the relation R can also be represented in another way. Theorem 4.2: (E, F ) is a pair of structural equivalences on the system (U, V, R) if and only if R can be represented by (, ) R (u )(v ) (u, v) R, for all U/E and V /F . Proof: Let (E, F ) be a pair of structural equivalences. (37)

so (Xk , Yk ) is a solution to inequality (23), and consequently, it is a solution to equation (17). Evidently, (Xk , Yk ) is contained in (J, K). Let (X , Y ) be an arbitrary solution to (17), that is, to (23), contained in (J, K). Then (X , Y ) (X1 , Y1 ), and if we suppose that (X , Y ) (Xm , Ym ), for some m N, then (X , Y ) whence (X , Y ) (Xm , Ym ) (Xm , Ym ) = (Xm+1 , Ym+1 ). (X , Y ) (Xm , Ym ),

Hence, by induction we conclude that (X , Y ) (Xm , Ym ), for every m N, which means that (X , Y ) (Xk , Yk ). Therefore, we have proved that (Xk , Yk ) is the greatest solution to inequality (23) and equation (17).

Let U/E and V /F such that (, ) R. Then by (36), for arbitrary u and v we have that (u, v) E R F = R. Since the converse implication in (37) is obvious, we conclude that (37) holds. Conversely, let (37) hold and let (u, v) E R. Then we have that (u, u ) E and (u , v) R, for some u U , and according to (35) this implies that (Eu , Fv ) R and u Eu . Now by (37) we get that (u, v) R, and hence, we have proved that E R R. Since the opposite inclusion is true because of reexivity of E, we conclude that E R = R. In the same way we show that R F = R. Therefore, we have proved that (E, F ) is a pair of structural equivalences. Finally, when (E, F ) is a pair of regular equivalences, the relation R can be represented as follows. Theorem 4.3: (E, F ) is a pair of regular equivalences on the system (U, V, R) if and only if R can be represented by (, ) R (u )(v ) (u, v) R & (v )(u ) (u, v) R (38)

for all U/E and V \Q. Proof: Let (P, Q) be a pair of regular quasi-orders. Consider arbitrary U/P and V \Q. If (, ) R, then according to (39), for any u and v we have that (u, v) P R Q = P R = R Q. Thus, there are u U and v V so that (u, u ) P , (u , v) R, (u, v ) R and (v , v) Q, so u uP = and v Qv = . Therefore, the right side of the equivalence (40) holds, and the direct implication in (40) is proved. On the other hand, let the right side of the equivalence (40) be true. Consider any u and v . Then there is v such that (u, v ) R, and by = Qv it follows (v , v) Q, which means that (u, v) RQ P RQ. Now, by (39) we get that (, ) R. Hence, we have proved that (40) holds. Conversely, let (40) be true. Suppose that (u, v) P R, for some u U and v V . If we set uP = and Qv = , then u and u . Consider arbitrary u and v . We have that (u , u) P , (u, v) P R and (v , v) Q, so (u , v ) P R Q, and according to (39), we conclude that (, ) R. Now, by (40), we obtain that there is v such that (u, v ) R, and = Qv yields (v , v) Q. Hence, we conclude that (u, v) RQ, and we have proved the inclusion P R R Q. Analogously we prove the opposite inclusion. Accordingly, (P, Q) is a pair of regular quasi-orders, and this completes the proof of the theorem. The following example demonstrates an interesting situation where a relation can not be reduced by means of a pair of regular equivalences, but it can be reduced by means of a pair of regular quasi-orders. Example 4.5: Let U and V be sets such that |U | = 4 and |V | = 3, and let R be a relation between U and V given by 1 0 1 1 1 0 R= 1 0 1 , 0 1 0 and J and K relations on 1 0 0 0 1 0 J = 0 0 1 0 1 0 U and V , respectively, given by 0 1 1 0 1 , K = 1 1 0 . 0 0 0 1 1

for all U/E and V /F . Proof: Let (E, F ) be a pair of regular equivalences. Let U/E and V /F such that (, ) R. Then by (36), for any u and v we have that (u, v) E RF . By the hypothesis, E R F = R F F = R F , and also, E R F = E E R = E R. Accordingly, we have that (u, v ) R and (v , v) F , and (u, u ) E and (u , v) R, for some u U and v V , and clearly, u and v . Therefore, the right side of the equivalence (38) holds, and we have proved the direct implication in (38). Since the converse implication holds according to (35), we get that (38) is true. Conversely, let (38) hold and let (u, v) E R. Then there is u U such that (u, u ) E and (u , v) R, and according to (35) we obtain that (Eu , Fv ) R and u Eu . Now, by (38) it follows that (u, v ) R, for some v Fv , which yields (v , v) F , and hence, (u, v) R F . Thus, we have proved that E R R F . Analogously we prove the opposite inclusion, and hence, (E, F ) is a pair of regular equivalences. In the sequel, let U and V be non-empty sets, and let P and Q be arbitrary quasi-orders on U and V , respectively. For any afterset U/P and foreset V \Q we set = {u U | uP = }, = {v V | Qv = }. It is worth noting that and are equivalence classes with respect to the natural equivalences of P and Q. Further, consider an arbitrary relation R R(U, V ). Then we dene a relation R R(U/P, V \Q) by (, ) R (u )(v ) (u, v) P R Q, (39) for all U/P and V \Q. We have that the following is true. Theorem 4.4: (P, Q) is a pair of regular quasi-orders on the system (U, V, R) if and only if R can be represented by (, ) R (u )(v ) (u, v) R & (v )(u ) (u, v) R (40)

Then the greatest solution to equation (17) contained in (J, K) is the pair (X, Y ) of quasi-orders represented by 1 0 0 0 1 0 0 0 1 0 1 X= 0 0 1 0 , Y = 1 1 0 . 0 0 1 0 1 0 1 The reduced relation R R(U/X, V \Y ), dened according to rule (39), is represented by 1 0 1 R = 1 1 0 . 1 0 1 On the other hand, the greatest solution to system (24) contained in (J, K), i.e., the greatest pair of equivalences which

is a solution to equation (17), is the pair (U , V ) of equality relations on U and V . Therefore, the relation R can not be reduced by means of a pair of regular equivalences, but, as we have shown, it can be reduced by means of a pair of regular quasi-orders. R EFERENCES
[1] V. Batagelj, Large scale social network analysis, in: R. A. Meyers (Ed.), Encyclopedia of Complexity and Systems Science, Springer, New York, 2009, pp. 82458265. [2] G. Birkhoff, Lattice Theory, third ed., American Mathematical Society, Providence, RI, 1973. [3] T. S. Blyth, Lattices and Ordered Algebraic Structures, Springer, London, 2005. [4] S. P. Borgatti, Two-mode concepts in social network analysis, in: R. A. Meyers (Ed.), Encyclopedia of Complexity and Systems Science, Springer, New York, 2009, pp. 82798291. [5] C. Carpineto, G. Romano, Concept Data Analysis: Theory and Applications, Wiley, New York, 2004. c [6] M. Ciri , J. Ignjatovic, M. Bai , I. Jan i , Nondeterministic automata: sc cc Simulation, bisimulation and structural equivalence, submitted to Computers & Mathematics with Applications. c [7] M. Ciri , J. Ignjatovi , S. Bogdanovi , Uniform fuzzy relations and c c fuzzy functions, Fuzzy Sets and Systems 160 (2009) 10541081. c [8] M. Ciri , J. Ignjatovi , N. Damljanovi , M. Bai , Bisimulations for c c sc fuzzy automata, submitted to Fuzzy Sets and Systems.

c [9] M. Ciri , J. Ignjatovi , I. Jan i , N. Damljanovi , Algorithms for c cc c computing the greatest simulations and bisimulations between fuzzy automata, submitted to Fuzzy Sets and Systems. c [10] M. Ciri , A. Stamenkovi , J. Ignjatovi , T. Petkovi , Factorization c c c of fuzzy automata, In: Csuhaj-Varju, E., Esik, Z. (eds.), FCT 2007, Springer, Heidelberg, Lecture Notes in Computer Science 4639 (2007) 213225. c [11] M. Ciri , A. Stamenkovi , J. Ignjatovi , T. Petkovi , Fuzzy relation c c c equations and reduction of fuzzy automata, Journal of Computer and System Sciences 76 (2010) 609633. [12] P. Doreian, Positional analysis and blockmodeling, in: R. A. Meyers (Ed.), Encyclopedia of Complexity and Systems Science, Springer, New York, 2009, pp. 69136927. [13] P. Doreian, V. Batagelj, A. Ferligoj, Generalized blockmodeling of twomode network data, Social Networks 26 (2004) 2953. [14] P. Doreian, V. Batagelj, A. Ferligoj, Generalized Blockmodeling, Cambridge University Press, 2005. [15] B. Ganter, R. Wille, Formal Concept Analysis: Mathematical Foundations, Springer, Berlin 1999. c [16] J. Ignjatovi and M. Ciri , Weakly linear systems of fuzzy relation c inequalities: A survey, submitted to Advances in Fuzzy Systems. c [17] J. Ignjatovi , M. Ciri , S. Bogdanovi , On the greatest solutions to c c weakly linear systems of fuzzy relation inequalities and equations, Fuzzy Sets and Systems 161 (2010) 30813113. c [18] J. Ignjatovi , M. Ciri , N. Damljanovi , I. Jan i , Weakly linear systems c c cc of fuzzy relation inequalities: The heterogeneous case, submitted to Fuzzy Sets and Systems. [19] S. Roman, Lattices and Ordered Sets, Springer, New York, 2008.

You might also like