Elsevier

Information Sciences

Volume 521, June 2020, Pages 398-421
Information Sciences

Generalized textural rough sets: Rough set models over two universes

https://doi.org/10.1016/j.ins.2020.02.044Get rights and content

Highlights

  • An extensive study on rough sets for two-universe approximation spaces are systematically discussed for textures.

  • t-Seriality and t-inverse seriality are defined and almost all basic results on rough sets are formulated in textures.

  • The connection between revised approximation operators, and object oriented and attribute oriented formal concept lattices are given.

Abstract

The aim of this paper is to reveal the useful aspects of textures for generalized approximation spaces in rough set theory. To this end, we present a counterpart of rough sets over two universes for textures. We compare the textural results with the well-known basic properties of rough sets given in the literature. First, we define a t-seriality and a t-inverse seriality in textures. We show that almost all basic results with respect to rough set model over two universes can be formulated using textures. On the way, we give new results and observations due to lower and upper approximation operators which are not taken into consideration by the researchers. Moreover, we also discuss the revised approximation operators and attribute oriented formal concept lattices in a textural framework.

Introduction

Rough set models over two universes ensure a reasonable generalization of classical rough sets of Zdzislaw Pawlak [15]. They provide useful applications with respect to information systems (see e.g., [4], [12], [13], [14], [18], [19], [23]). The studies on the subject were first given by Wong et al. [22], and Yao et al. [24], [25]. An extensive study on two-universe rough sets can be also found in the paper of Pei and Xu [16]. On the other hand, recent studies on textures and rough sets are subjected to single universe rough set models [5], [6], [7], [8], [9]. In this work, we focus on the textural rough sets over two universes with the aim to exhibit the unnoticed connections between rough sets and textures. Note that the seriality of a relation is one of the conditions of a function while the inverse seriality corresponds to surjectivity. In a similar way, the textural seriality and textural inverse seriality given in this paper are in fact the conditions DF1 and SUR properties of a difunction, respectively [1]. Let (U, V, r) be a generalized approximation space. Then (U, V, r, R) is a generalized t-approximation space where U=(U,P(U)), V=(V,P(V)) and R=(U×V)r. This connection provides a systematic approach to the basic properties of rough set models over two universes as follows. Note that the compatibility notion is a necessity in defining the approximation operators for (U, V, r). Let us consider the mapping G:VP(U) defined by G(v)={uU(u,v)r} for all v ∈ V. Then the mapping G:P(V)P(U) given by BP(V),G(B)=vBG(v) leads us to the upper approximation apr¯F(B) of Yao [24] (see Table 1). However, although the existence of the remarkable equalitiesF(A)=apr¯G(A)andF(A)=apr̲G(A),andG(B)=apr¯F(B)andG(B)=apr̲F(B)

for all AU and BV, the researchers did not consider the approximations apr¯G,apr̲G:P(U)P(V) as the mappings F and F. Here, F and G are complementary duals of the mappings F and G, respectively. Under the compatibility condition, Theorem 4 in [16] states that for all BV,

(DL7) apr̲F(B)G(B),

(DH7) G(B)apr¯F(B).

However, without any condition, we already have G(B)=apr¯F(B). Then under the seriality condition, we immediately conclude that apr̲F(B)apr¯F(B). This is in fact one of the equivalent conditions in Theorem 7 in [16]. Further, we easily show thatAU,Aapr̲F(apr¯G(A)),apr¯F(apr̲G(A))A,andBV,apr¯G(apr̲F(B))B,Bapr̲G(apr¯F(B)).

The above inclusions are immediate results of Lemma 2.6 given in this paper. Note that the compatibility of r is not a necessary condition for the above inclusions (see Corollary 3.12). Hence, we see that a textural discussion not only makes Theorem 4 and the related results presented in [16] more understandable, but also removes the unnecessary conditions of the propositions. Another discussion may be given as follows. In [18], Proposition 8(1) asserts that for any generalized approximation space (U, V, r) and for any BV, the inclusion r(r(B)) ⊆ B holds. The operation r is defined for the subsets of V and however, r(B) is a subset of U. Hence, the set r(r(B)) is not defined in [18]. Now let us look at the case from the textural point of view. If (r, R) is a t-inverse serial direlation from (U,U) to (V,V), then we obtain thatBV,R(rB)Br(RB).

For any generalized approximation space (U, V, r), a result of the above inclusion may be given asBV,apr̲G(apr̲F(B))Bapr¯G(apr¯F(B))

where r is an inverse serial relation (Corollary 3.20). Now the statement r(r(B)) ⊆ B should be the same as the first inclusion above. Further, the second inclusion is the proposition (DH7’) in Theorem 4 [16] given under the compatibility condition. Hence, the textural approach removes the deficiency occured in Proposition 8 [18] and Theorem 4 [16] giving a systematic approach for the inclusions from both sides as well. In addition to above inclusions, we also haveAU,apr̲F(apr̲G(A))Aapr¯F(apr¯G(A))where r is a serial relation (Corollaries 3.18 (ii) and 3.33 (i)). For the definability notion, it is worthy to note that if r is compatible and injective, then every subset AU is successor definable. If r1 is compatible and injective, then the dual case also holds, that is every subset BV is predecessor definable. Moreover, for the corresponding conditions, the sets aprG(A) and apr¯G(A), and aprF(B) and apr¯F(B) are successor definable and predecessor definable sets, respectively. Another significant case in this work is the formulation given for the revised aproximations in textures asBP(V),Apr̲F(B)=rrB,Apr¯F(B)=RRB,andAP(U),Apr̲G(A)=RRA,Apr¯G(A)=rrAwhere (r, (U × V)∖r) is the direlation from (U,P(U)) to (V,P(V)) with (U×V)r=R. This gives all basic properties with respect to revised approximations through the textures. Here, we may also observe that for all AU, AprG(A) and Apr¯G(A) are successor definable. Similarly, for all BV, Apr¯F(B) and AprF(B) are predecessor definable. What is more, for any subset of B of V, the sets aprF(B) and apr¯F(B) are successor definable sets with respect to revised approximations on U. Moreover, in view of a textural result given in Theorem 6 in [9], the revised lower and upper approximations can be considered as natural generalizations of the subset approximations of Yao. Indeed, if we take U=V, then the revised approximation operators Apr̲F,Apr¯F,Apr̲G and Apr¯G correspond to the subset approximation operators apr̲,apr¯ given by Yao in [26]. For instance, if (U, r) is an approximation space, then for all AU, we conclude thatApr̲F(A)=apr̲F(A)andApr¯F(A)=apr¯F(A).

In particular, for any two domain of discourses U and V with U ⊂ V, this argument allows us a comparison between the approximation operators of the spaces (U, r) and (U, V, r). Indeed, one of the approximations of subsets of (U, r) may change asAU,apr¯FV(A)Apr¯F(A)even if the relation r does not consider the objects of the set VU. Consequently, this paper reveals the useful aspects of textures for generalized approximation spaces in rough set theory.

This work is structured as follows. Section 2 is devoted to the basic concepts and results on textures. In Section 3, seriality and inverse seriality are discussed in textures and some basic formulations for approximations are presented. In particular, we present a comparison about the formulations with respect to mappings and approximations in the framework of textures and rough sets (see Table 1). In Section 4, the revised approximations given by Pei and Xu in [16] are considered in textures. It is known that the formal concept analysis [21] and rough set theory [15] are two complementary fields from the point view of data analysis. In the appendix, we also present a systematic discussion on the connections between the revised approximation operators and the object oriented formal concept lattices of Yao [28].

Section snippets

Textures

Definition 2.1

[3]

Let U be a set. Then UP(U) is called a texturing of U, and (U,U) is called a texture space, or in brief a texture, if

  • (i)

    (U,) is a complete lattice containing U and ∅, which has the property that arbitrary meets coincide with intersections, and finite joins coincide with unions, that is, for all index sets K,kKAk=kKAkand for all finite index sets KkKAk=kKAkwhere {AkkK}U.

  • (ii)

    U is completely distributive, that is, for all index set K, and for all k ∈ K, if Jk is an index set and if AkjU,

Generalized textural approximation spaces

It is known that a generalized rough set with respect to two universes can be defined using the notion of interval structure given in [22]. Note that this approach may be considered in view of the compatibility relation. If U and V are two domain of discourses and r be a binary relation from U to V, then r is said to be [17]

  • (i)

    serial, if for any u ∈ U, there exists a v ∈ V such that (u, v) ∈ r,

  • (ii)

    inverse serial, if for any v ∈ V, there exists a u ∈ U such that (u, v) ∈ r,

  • (iii)

    a compatibility relation if

Revised two-universe rough set models

Note that for a two-universe rough set model, an approximation of a subset of one of the domain of discourses is a subset of the other one (see, e.g., [16]). However, this is mostly not the case since the evaluations may have no meaning from the point of view of knowledge discovery for information systems. Here, we consider this case in the framework of textures using sections and presections of a direlation. Note that if (r, R) is a direlation from (U,U) to (V,V), then in view of the

Conclusion and appendix

The studies on rough sets and textures are mostly subjected to approximation spaces with a single universe. This paper is a first step for an extensive study on textures and two-universe approximation spaces. In view of the representation theorem of textures (Theorem 2.1 in [2]), this work also provides a base for the studies on textures, rough sets, fuzzy sets in the framework of concept lattices. For instance, recall that for any two domain of discourses U and V, and for all AU, BV, the

Declaration of Competing Interest

We wish to confirm that there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome.

Acknowledgement

The authors would like to express their sincere thanks to the anonymous reviewers for their valuable comments improving the presentation of this paper.

Cited by (0)

View full text