You are on page 1of 10

A Topology-based Semantic Location Model for Indoor

Applications

Dandan Li Dik Lun Lee


Department of Computer Science and Department of Computer Science and
Engineering Engineering
Hong Kong University of Science and Hong Kong University of Science and
Technology Technology
Clear Water Bay, Hong Kong Clear Water Bay, Hong Kong
ddli@cse.ust.hk dlee@cse.ust.hk

ABSTRACT 1. INTRODUCTION
Location-based services (LBSs) play more and more impor- Location based services (LBSs) are quickly growing in
tant roles in our daily life with the prevalence of mobile prevalence with the advances of wireless communication and
devices and the internet. Location modeling is a signif- mobile devices. Facing sophisticated applications, such as
icant research topic in LBSs, which is needed to provide location-based instant messaging and location-based alerts,
a well-defined representation of location knowledge for lo- LBSs are expected to provide more intelligent services for
cation browsing, navigation and query processing. In this large user community. Location model is an essential part
paper, we propose that a topological structure can be at- of location aware applications, which requires the represen-
tached to an exit-location space model, which can preserve tation of location knowledge to be understandable and ca-
the topology and distance semantics between locations (ex- pable of capturing the rich semantics of the physical world.
its). The Q-analysis developed by R. H. Atkin is used to Much research has been demanded and carried out in de-
analyze the semantic information of the model. Compared veloping location models to describe the physical space and
with those existing models which only reveal the relation- properties of the objects contained in it.
ships between two entities, this novel model can provide the In this paper, we focus on designing a semantic loca-
analysis of n-ary relationships (i.e., the relations among n en- tion model for indoor applications. Compared with out-
tities) from both local and global viewpoints. Moreover, by door environments, the indoor environments are inherently
using the rich structures obtained from the topological anal- diverse and much more highly structured. In general, an
ysis, we define a semantic distance which can support more outdoor environment can be clearly described by the univer-
meaningful navigation and queries on complicated indoor sal longitude-latitude-altitude coordinate system. However,
environments. Examples are described in detail to demon- this coordinate system is not really suitable for indoor envi-
strate the effectiveness of our model. ronments since each indoor environment has its own way to
describe its locations which have various contexts. There-
fore, the location model with rich structures is needed for
Categories and Subject Descriptors indoor applications which can capture complicated relation-
H.2 [Information Systems]: Database Management; H.3.2 ships among locations, such as the type of a location (room
[Information Storage and Retrieval]: Content Analy- or corridor), the connective structure of n locations, etc.
sis and Indexing; H.3.3 [Information Storage and Re- We propose in this paper the use of Q-analysis devised by
trieval]: Information Search and Retrieval. R. H. Atkin to semantic location model for indoor applica-
tions. The indoor space is modeled by two types of entities:
locations and exits, where a location is a bounded geometric
General Terms area with one or more exits and an exit is a boundary point
Algorithms and Theory. through which a user can leave or enter a location [28][20].
Based on this model, we construct an exit-location matrix
A, of which the elements aij are only one of the two val-
Keywords ues 1 and 0, with 1 (respectively 0) indicating the presence
Semantic Location Model, Topology, Q-analysis, Connec- (respectively absence) of the ith exit in the jth location.
tiveness, Simplicial Family. Hence this matrix is an incidence matrix. A simplex, which
is a basic object in the study of algebraic topology, is at-
tached to each column of the incidence matrix, and then
the collection of these simplexes forms a simplicial family.
Permission to make digital or hard copies of all or part of this work for The Q-analysis is applied to analyze the topological struc-
personal or classroom use is granted without fee provided that copies are ture of the location model, which is a technique of multidi-
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
mensional data analysis that makes us observe the semantic
republish, to post on servers or to redistribute to lists, requires prior specific structure of the indoor space from both local and global per-
permission and/or a fee. spectives. In comparison with existing symbolic models, the
ACM GIS ’08 , November 5-7, 2008. Irvine, CA, USA. Q-analysis method is like a microscope and therefore makes
(c) 2008 ACM ISBN 978-1-60558-323-5/08/11...$5.00.
us do much deeper investigations into structures of an in- tary benefits and drawbacks [24]. To overcome their weak-
door environment. Our topology-based model can reveal nesses, hybrid location models are proposed by [29][24][12],
the n-ary relationships (i.e., the relations among n entities) which are combinations of symbolic and geometric models.
such as hole, star and hub, as well as basic relationships be- The common basic idea under these hybrid models is that
tween two entities, such as containment and connectedness. objects are organized in a hierarchy in which every level
Furthermore, by using the rich structures obtained from the is a refinement of its previous and then each object in the
topological analysis, a semantic distance can be defined to hierarchy is given its own coordinate system [26]. More-
support more meaningful navigation and queries on compli- over, coordinate systems of subspaces and superspaces can
cated indoor environments. be transformed between two spaces [24].
The rest of the paper is organized as follows. Section 2 Pradhan [34] introduced a semantic representation of ob-
discusses the related work on modeling locations. Section 3 jects, which is totally different from symbolic and geometric
reviews the basic concepts of algebraic topology which are representations. The semantic representation provides the
the foundations of the core subjects of the paper. In Sec- contextual information of an object, which can characterize
tion 4, we first represent the indoor space as an exit-location the situation around the object [36]. Another semantic loca-
matrix, and then build the topological location model by at- tion model for indoor navigation was developed by Hu and
taching the mathematical structure of simplex to each col- Lee [19][20]. The model consists of two types of entities: lo-
umn/row of the matrix. Based on this model, some topolog- cation and exit. It is actually a graph-based symbolic model,
ical characteristics are derived to reveal semantic relation- but it can be derived from the geometric representation of
ships among locations. By using the information obtained the indoor space [38][28]. The hierarchical structure of the
from the topological analysis, we explore the potential ap- model is consistent with human cognition during navigation
plications of the model and give a comparison analysis of the and therefore it is more suitable for display on mobile de-
new model and existing location models in Section 5. The vices than a plain floor plan. However, this model can not
last section summarizes the results obtained in this paper fully support object moving and can deal with only one type
and discusses directions for future work. of relation “reachability” [28].
Further related work in this direction includes Ye et al.
2. RELATED WORK [38] who proposed a unified space model for more complex
environment based on lattice and graph models, as well as
This section will give an overview of existing location mod-
O’Connell et al. [32], Pederson [33] and Schlieder et al. [35]
els, which can be mainly classified into symbolic and geomet-
who build models for mobile objects or users.
ric models [29][9].
In a symbolic location model, all objects are represented
as symbols and referred to by names. Based on the inclusion 3. MATHEMATICAL FOUNDATIONS
relation between two sets, these symbols can be divided into Q-analysis proposed by R. H. Atkin (standard references
sets and subsets, and then form a hierarchical structure, are [3][5][6][1][2][4]) is a technique of multidimensional data
such as a tree or a lattice [20][26][12][25]. A representa- analysis, which makes us observe the space from both lo-
tive example of symbolic models is the semantic location cal and global perspectives. Data structures are revealed
model presented by Brumitt and Shafer [10]. The model by techniques of algebraic topology, which impose noth-
is not a geometric model since it can represent containment ing on the data and “let the data speak for themselves”
and connectedness relationships within a space and not with [8][16]. Compared with the theories of graph and network,
any specific geometric position of the object in that space. Q-analysis can explore n-ary relationships (i.e., the relation-
Moreover, it has a lattice structure and a friendly naming ships among n entities) as well as the relationships between
system, which allow a person to perform queries according two things, which are more important for some practical
to semantic information of symbols. The prime advantage problems [30][13]. Q-analysis has been successfully applied
of symbolic models is that they have clear representation to many fields, such as geographical studies [8], social and
of spatial relationships with semantic information which is planning sciences [1][2][4][7][17], mass media [18], artificial
easily understandable for humans. However, it is a heavy intelligence [21][37], information retrieval [11][27][30][31], data-
cost to construct and maintain symbolic models manually. base interaction [22], spatial databases [23][14], and so forth.
In a geometric location model, the physical space is rep- The followings are the basic concepts of algebraic topol-
resented as the Euclidean space and the objects therein are ogy, including simplex, simplicial family, and Q-analysis which
represented as points, lines, regions or volumes in the Eu- are foundations of the core subjects of the paper [31][30].
clidean space. Each object in the geometric model is de- A simplex is a basic element of algebraic topology, and
scribed by the set of coordinates defined in the Euclidean every simplex has a dimension. A 0-dimensional simplex is
space. Distance between two objects can be calculated based a point in the Euclidean space IRN , a 1-simplex is a line
on their coordinates. There are no set inclusion relation- segment joining two points and a 2-simplex is a triangle.
ships between two objects like the above symbolic models. Figure 1 shows four simplexes of dimension k = 0, 1, 2, and
A well known application based on the geometric model is 3, respectively. The formal definition is given as follows.
the GPS coordinate system, where locations are described
by the universal longitude-latitude-altitude coordinate sys- Definition 1. Let v0 , v1 , . . . , vk be k + 1 points of IRN
tem [26][24][29]. Geometric models have the following ad- in general position (i.e., the vectors − v− → −−→ −−→
0 v1 , v0 v2 , . . . , v0 vk are
vantages: they provide accurate position representation for linearly independent). We denote by σ k = (v0 , v1 , . . . , vk )
objects and support more flexible ways to retrieve location the smallest convex set spanned by these points. We call σ k
information. However, they have weak structures and can a simplex of dimension k (or a k-simplex) and v0 , v1 , . . . , vk
not reflect well semantic relationships among objects. the vertices of the simplex. A q-simplex spanned by a subset
In fact, symbolic and geometric models have complemen- of these vertices is called a q-face of σ k . We use the notation
maximum value of q = q ∗ the maximal L-chain, denoted by

Cab (L). We say that σa and σb are q ∗ -connected if they are
connected by a maximal chain.
The relation “is q-connected to” on a simplicial family Σ
obeys the following three conditions: “reflexivity”, “symme-
try” and “transitivity”, therefore it is an equivalence rela-
tion, which is represented as rq . We have the following “Q-
analysis” structure on a simplicial family.
Definition 4. The relation “is q-connected to” on a sim-
Figure 1: From left to right—simplexes of respective plicial family Σ, is an equivalence relation which is de-
dimensions 0, 1, 2, and 3. noted rq . Let Σq be the set of simplexes in Σ with dimension
greater than or equal to q, where q = 0, 1, . . . , dimΣ. Then
rq partitions Σq into equivalence classes of q-connected sim-
plexes. These equivalence classes are called the q-connected
components of Σ. Qq denotes the number of q-connected
components in Σ. The determination of the components and
Qq for each value of q is termed a Q-analysis of Σ.

4. A TOPOLOGY-BASED SEMANTIC LO-


CATION MODEL
In this section, we build a topology-based semantic lo-
(i) (ii) cation model for indoor applications and analyze the char-
acteristics of its topological structure to reveal the latent
Figure 2: (i) A simplicial family. (ii) Not a simplicial semantic relationships among objects.
family.
4.1 Modeling the Indoor Space
The indoor environment we consider is composed of two
type of entities, including locations and exits. Here, a loca-
. to denote the face relation, which is a partial order on any tion is defined as a bounded geometric area with one or more
set of simplexes. exits and an exit is a boundary point through which a user
can leave or enter a location [28][20]. We then construct
an exit-location matrix A = [aij ], where the elements aij
If we splice some simplexes together, then they can form are only one of the two values 1 and 0, with 1 (respectively
a larger object. 0) indicating the presence (respectively absence) of the ith
exit in the jth location. Based on the above mathematical
Definition 2. Two simplexes σ1m and σ2n in IRN are prop- foundations, we can attach each location vector with a sim-
erly joined if either they do not intersect or their intersec- plex, then the collection of locations form a simplicial fam-
p
tion σ1∩2 = σ1m ∩ σ2n , p ≤ m ≤ n, is a face of both σ1m and ily. Therefore semantic information among locations may
n
σ2 . A simplicial family is a finite collection Σ of sim- be investigated comprehensively in terms of this topologi-
plexes which are properly joined. The dimension of Σ is cal model. The following simple example gives a detailed
the highest dimension of the simplexes in Σ and is denoted explanation for the modeling process.
dimΣ. Example 1. Figure 3 shows a small area of a building,
which consists of 4 rooms(R), 1 corridor(C), and 6 exits(e).
Figure 2 shows two collections of simplexes, one is a sim-
R1 , R2 , R3 , R4 and C are locations, and e1 , e2 , · · · , e6 are
plicial family and the other is not.
exits. Table 1 gives the 6×5 exit-location matrix correspond-
ing to Figure 3.
“Nearness” and “connectiveness” are important relation-
ships in a simplicial family, which are defined as follows:
Observing the column R1 of the exit-location matrix, the
Definition 3. Let Σ be a simplicial family and d is the row e1 contains a “1” and the other rows contain “0”. We
0
highest dimension of the simplexes in Σ. Let 0 ≤ q ≤ d be associate with R1 a 0-simplex σ(R 1)
= (e1 ). In a similar way
an integer. We call two simplexes σa and σb in Σ q-near if we obtain the following simplexes for the remaining columns:
they have a common q-face. We call σa and σb q-connected 1
σ(R = (e2 , e3 ),
2)
if there exists a sequence 1
σ(R 3)
= (e5 , e6 ),
σ1 , σ2 , . . . , σj (1) 1
(2)
σ(R 4)
= (e4 , e5 ),
of distinct simplexes of Σ, such that σ1 = σa , σj = σb , and 3
σ(C) = (e1 , e2 , e4 , e6 ).
σi is qi -near to σi+1 for all 1 ≤ i ≤ j − 1, 0 ≤ qi ≤ d an
integer, and q = min{qi }. We call the sequence (1) a q- From Figure 4, we can see that exits are modeled by ver-
chain Cab from σa to σb and the number (j − 1) the length of tices, locations are represented as simplexes and the five
Cab , denoted by l(Cab ). For all possible q-chains connecting location simplexes form a simplicial family.
σa to σb with the same length L, we call the chain with the
Strength Length Chain
R1 to R2 1 2 R1 → C → R2
R3 to R2 1 2 R3 → C → R2
R4 to C 1 1 R4 → C
R1 R2 e3
Table 2: Connective strength and length in Example
e1 e2 1.
C
e6 e4
4.2 Mining Semantic Relationships within the
R3 e5 Topological Model
R4
The semantic information of an entity is any information
that can be used to characterize the situation around the
Figure 3: A small area of a building. entity. In general, the types of entities in an indoor envi-
ronment are more diverse than those in an outdoor envi-
ronment. Therefore, an indoor environment has more abun-
dant semantic information. In this paper, we consider two
fundamental semantics: semantic relationships and seman-
tic distance between entities. In the following part, we ex-
plore semantic relationships among locations based on our
topological location model. Compared with a graph-based
model, the topological model can reveal not only the binary
relations between two entities but also the relations among
R1 R2 R3 R4 C n entities by the topological characteristics given as below.
e1 1 0 0 0 1 Moreover, each location in the graph model is only repre-
e2 0 1 0 0 1 sented by a node and there is no other information about
e3 0 1 0 0 0 the location, but in the topological model, each location is
e4 0 0 0 1 1 represented by a simplex with a dimension, and therefore
e5 0 0 1 1 0 the latter can provide more elaborate information. For ex-
e6 0 0 1 0 1 ample, the graph model can only tell us that a location is
connected to the other location, but the topological model
can describe further how they are connected to each other.
Table 1: The exit-location matrix corresponding to In our model, each location is represented as a simplex
Figure 3. while each exit as a vertex. The dimension of the sim-
plex plus 1 is equal to the number of exits that a location
has. The connective relationship between two locations is
described by the following characteristics of the chain con-
necting the two location simplexes.

4.2.1 Connective Strength and Connective Length


We assume that the weight of each edge in a simplex fam-
0 ily is equal and is set to 1 in the whole paper. In other
σR 1 cases positive real number may be assigned to each edge to
e1 reflect the distance between two exits. Suppose two location
simplexes σa and σb are connected by a q-chain with length
e3 L in terms of the definition 3, then the connective strength
of the two locations is defined as q + 1 and the connective
length is L.
e6 e2
1
σR 2
Example 2. Referring to Figures 3 and 4 in Example 1,
we have the following results in Table 2. Since R1 is con-
hole
3 nected to R2 by the chain {R1 , C, R2 }, the connective length
1 e4 σC
σR 3
is equal to 2. Furthermore, R1 and C are 0-connected and
C and R2 are also 0-connected, and therefore R1 and R2 are
e5 0-connected and the connective strength is equal to 1. Other
1
σR results in Table 2 can be obtained in a similar way.
4
These results tell us that a person must pass through the
Figure 4: The simplexes generated by columns of
location C (a corridor) if he or she wants to arrive at the
the matrix 1.
location R2 from the location R1 or R3 , but he or she can
get to the corridor C directly from the location R1 , R3 or
R4 .
The complete pseudo-code for calculating “connective stre-
ngth” and “connective length” between location a and other
locations is given as below. The basic idea is to find those
locations that are connected to location a at the first level
(with length 1), and then find connections at higher levels
step by step.

Algorithm 1 Algorithm for Computing Connective L4 L2


Strength and Length. e2
Input: An m-dimensional vector a, which is the represen- e3 e4
tation of a location with respect to m exits; e5
An m × n matrix A, which is the exit-location matrix of
other n location column vectors. L3 e1 L1
Output: A vector Q, which records the connective strength
between location a and other locations;
A vector L recording the connective length between lo-
cation a and other locations. Figure 5: An exit-location area in a building.
1: C = a, Ā = A, k = 1; {k is the connection length}
2: while C T Ā is not a zero matrix do
3: B = C T Ā;
4: I = i, ∃j, such thatB(j, i) >= 1; {find location i in Ā
that connects to location j in C.}
5: L(i) = k, ∀i ∈ I; {mark the connection length for the
found locations in L.}
6: Q(i) = mink {maxj B(j, i)}, ∀i ∈ I; {mark the connec-
tive strength for the found locations in Q.}
7: C = {Ā(:, i), i ∈ I}; Ā ← Ā \ C; k = k + 1 {seper-
ate the locations in Ā into two sets; and increase the
connection length.} L1 L2 L3 L4
8: end while e1 1 0 1 0
e2 0 1 0 1
e3 1 1 0 0
e4 1 1 0 0
4.2.2 Hole
0 1 3
e5 0 0 1 1
Observing Figure 4, we see that σR 1
, σR 2
and σC form
3 1 1
a tightly knitted group whereas σC , σR3 and σR4 are rela-
tively isolated. The reason is that there is an empty region Table 3: The exit-location matrix corresponding to
3 1 1
within the structure constructed by σC , σR 3
and σR 4
. An Figure 5.
empty region in a simplicial family is termed a hole. The
hole structure is formed in Figure 4 because the exits e4 , e5
and e6 do not appear simultaneously in one location. The
hole structure in an indoor environment tends to restrict
the walking route of a mobile person. The following exam-
ple gives a good demonstration of the fact.

Example 3. Figure 5 is another small area of a building.


The corresponding exit-location matrix is given in Table 3.
We attach each location column vector with a simplex then
we get the simplicial family in Figure 6. Since there does e3
2
not exist a location which includes all four exits e1 , e2 , e4 2 σL
σL 2
1

and e5 , they form a hole structure. Under this obstacle, a


person can not get to the exit e2 from the exit e1 directly and e4
must go through two locations {L3 , L4 } or {L1 , L2 }.
e2 hole
e1

1 e5
The computation of a hole structure is related to the σL 4
1
σL 3
mathematical concept “homology group”, which is beyond
the scope of the knowledge of this paper. Figure 6: Location simplicial family generated by
columns of the Table 3.
4.2.3 Q-analysis of a Simplicial Family
Based on Definition 4, we know that Q-analysis is a tech-
nique of multidimensional data analysis and makes us ob-
serve a simplicial family from both local and global angles.
The following example gives a detailed illustration.
Example 4. Table 4 gives the Q-analysis result of the
simplicial family in Example 3 according to Definition 4 and e2
Algorithm 1. Since the highest dimension of location sim- e3 e5
plexes is 2, the Q-analysis of the simplicial family has 3
levels, q = 0, 1, 2 respectively. The level q = 2 consists of L1 e1
L2
those simplexes with dimension greater than or equal to 2, e4
2 2 2
so this level contains two simplexes σL and σL . Since σL
2
1 2 1
L3
and σL2 are not 2-connected to each other but 2-connected
2 2
to itself respectively by a chain of length zero, σL 1
and σL 2
forms two 2-connected components. Next, at the level q = 1,
1 1 2
two more location simplexes σL3 and σL4 come in. σL1 and Figure 7: An exit-location area in a building.
2
σL 2
are 1-connected by a chain of length 1 because they have
two common exits e3 and e4 , therefore they are merged to e3
1
form a 1-connected component at the level q = 1. σL 3
and
1
σL4 both are 1-connected to itself and thus form two one- e5
member components. All the four location simplexes form a
larger 0-connected component at the level q = 0 since σL 2
is e2
1
1
2 2 1 1
1-connected to σL2 , σL2 is 0-connected to σL4 , and σL4 is 0- σL 3

connected to σL 1
. Moreover, the value Qq denotes the num- e1 e4
3
ber of q-connected components at each q-level. This structure 2
vector provides a measure of some of the global structure of a σL 2
2
simplicial family while each q-connected component captures σL 1
the local structure information of a simplicial family.
Figure 8: The simplexes corresponding to Figure 7.
Potential applications of the local and global information
are discussed in Section 5.
8, we have
4.2.4 Eccentricity ImpL1 (e) =
1
= 1;
In a building, an exit is a connector between two locations 1
or a boundary point opening into the outside of the build-
ing. For each location, the importance of its exits is different. 2 1
ImpL2 (e) = = ;
For example, if a location has only one exit, then this exit 4 2
is very important for the location since it is the only entity
2
connecting this location to others, while if a location has ImpL3 (e) =
.
many exits and some of them can open simultaneously into 3
the other location, then the importance of its exits is small. We can see that the importance of exit e1 for location L1
This information is important for a mobile user. For exam- gets the maximum value since it is the only exit leaving or
ple, a person may estimate how many ways he/she has to entering location L1 .
leave a location through its exits based on this information.
We will define a measure to quantify the above information.
Given a simplicial family, eccentricity (ecc) of a simplex We give the complete pseudo-code for computing “impor-
has been devised to measure the extent to which the simplex tance” of exits as below.
is integrated into the simplicial family [3]. The formula for
calculating eccentricity of a simplex σ is given by
4.2.5 Star and Hub
In our location model, rooms and corridors are all repre-
q̂ − q̌ sented as location simplexes. It is valuable to differentiate
ecc(σ) = , (3)
q̂ + 1 types of locations in an indoor environment since it may
where q̂ is called top-q and denotes the dimension of simplex aid in limiting queries about locations [10]. The following
σ, and q̌ is called bottom-q and is the largest number of definitions † and proposition can help us solve this problem.
vertices minus 1 that σ shares with any simplex. Definition 5. Let the star of a simplex σ in a simplicial
Based on Equation 3, we can calculate the eccentricity of family K be defined as
each location simplex. Let e be the exit of location L, then
0 0 0
the importance of exit e for location L (denoted by ImpL (e)) star(σ) = {σ |σ is q-near to σ, σ ∈ K}; (5)
is defined as
that is,
q̌ + 1
ImpL (e) = 1 − ecc(σL ) = . (4) 0 0 0
q̂ + 1 star(σ) = {σ |σ ∩ σ . σ, σ ∈ K}. (6)
The range of importance measure is at the interval [0, 1]. Definition 6. Let fi , i = 1, 2, . . . , s be the subsets of the
The larger the value is, the more important the exit for its simplexes of K. Let F be a set of these subsets, that is,
location is.

We have tailored the original definition of “star” in Q-
Example 5. Considering the exits in Figure 7 and Figure analysis to fit our present application.
Dimension Number of components
at level q List of components
q Qq
2 2 1 1
0 1 {σL 1
, σL 2
, σL 3
, σL 4
}
2 2 1 1
1 3 {σL1 , σL2 } {σL3 } {σL 4
}
2 2
2 2 {σL 1
} {σL 2
}

Table 4: Q-analysis of the simplicial family in Example 3.

Algorithm 2 Algorithm for computing the importance of σe12


exits. L4
Input: an m × n exit-location matrix A.
L2 σe15
Output: an n-dimensional vector Imp recording the im-
hole
portance of the exits for each location.
1: L = AT A; σe13 , σe14 L3
2: Imp = 0n×1 ;
3: q̌ = maxj=2 to n L1,j − 1; L1
4: q̂ = L1,1 − 1; σe11
5: Imp1 = (q̌ + 1)/(q̂ + 1);
6: for i = 2 to n-1 do Figure 9: Exit simplicial family generated by rows
7: q̌ = max(maxj=1 to i-1 Li,j , maxj=i+1 to n Li,j ) − 1; of the Table 3.
8: q̂ = Li,i − 1;
9: Impi = (q̌ + 1)/(q̂ + 1);
10: end for
11: for i = dim:dim do also associate each exit row vector with a simplex, then they
12: q̌ = maxj=1 to i-1 Li,j − 1; form an exit simplicial family. We call this characteristic the
13: q̂ = Li,i − 1; duality of the simplicial family. The exit simplicial family
14: Impi = (q̌ + 1)/(q̂ + 1); of Example 3 is given by Figure 9. We discuss its possible
15: end for functions in processing queries in the following section.

F = {f1 , f2 , . . . , fs }. Then we define hub(F) to be the in-


tersection of all elements in F :
\
s 5. POTENTIAL APPLICATIONS AND COM-
hub(F ) = fi . (7)
i=1,fi ∈F
PARISON ANALYSIS
We have given a detailed description of the topology-based
Based on the definitions of “star” and “hub” in a simpli- location model in the above section. Now, two natural ques-
cial family, we can show that a corridor simplex could be tions arise: what are the potential applications that this new
identified by its relation with other room simplexes. model can support? And what are the main differences be-
tween this model and the existing location models? In this
Proposition 1. A corridor simplex is one element in the section, we will answer these questions preliminarily with
hub of the stars of more than one room simplexes. some small examples.
We can get an intuitive feeling about the proposition from
the following validation process for Figure 3. In Example 5.1 Potential Applications
1, the stars of four room simplexes are listed respectively as The topology-based location model can reveal some hid-
follows: den structures among locations, such as, connective strength,
star(σR1 ) = {σR1 , σC }, connective length, hole structure, Q-analysis and dual struc-
star(σR2 ) = {σR2 , σC }, ture. Their potential applications are discussed as follows.
Application Example 1: Semantic Distance. For the gen-
star(σR3 ) = {σR3 , σR4 , σC }, eral navigation and nearest neighbor query tasks, we are
star(σR4 ) = {σR3 , σR4 , σC }. only interested in the shortest path and the shortest distance
between two objects. However, indoor environments are in-
Let F = {star(σR1 ), star(σR2 ), star(σR3 ), star(σR4 )}. Then
trinsically complicated. Both the connective strength and
we have
connective length between two locations may affect a per-
hub(F ) = star(σR1 ) ∩ star(σR2 ) ∩ star(σR3 ) ∩ star(σR4 ) son’s travel between them. Therefore, the semantic distance
= {σC }. should be considered before navigation, which can capture
not only the real distance information but also the structure
4.2.6 The Dual Structure information between two locations. Based on the topologi-
In previous examples, we attach each location column vec- cal model, we define the Connective Degree (CD) between
tor of an exit-location matrix with a simplex. In fact, we can two locations as follows, which is used to measure how easy
e3
a person can get from one location to another location, 2
2 σL
σL 1
2 ∗ Strength(Loi , Loj ) 2
CD(Loi , Loj ) = ∗ βL,
(dim(Loi ) + 1) + (dim(Loj ) + 1) e4
(8)
where Strength(·, ·) and L represent the connective strength e2 hole
e1
and length between two locations respectively; dim(·) de-
notes the dimension of a location; β is a parameter at the
interval [0, 1] which regulates the function of the chain length 1 e5
σL 1
σL
L in the measure. This idea is partially inspired by the 4
3
work of Prasanna Ganesan et.al [15] and Dandan Li et. al
Figure 10: The simplexes generated by columns of
[31][30]. According to this definition, we see that the larger
the Table 3.
the connective degree is, the easier the locations can be gone
through. This conclusion is consistent with our intuition
since it meets the following objective facts: the location
with high dimension means that it has many exits, which
building, and in the first floor the simplicial family forms a
makes it difficult for a person to choose a right exit lead-
single connected component at a relatively high dimensional
ing to the destination. That is why we consider this factor
level whereas in the second floor the structure is composed of
as the denominator of Equation 8. The large connective
several connected component even at the lower level. Then
strength between two locations implies that the number of
we know that the connective structure in the first floor is
the exits connecting them is great, and thus provides more
more complicated than that in the second floor and a per-
paths for a person to arrive at the destination and makes
son may cost more energy to find the optimal route in the
the travel easier. Furthermore, since the larger the length is
first floor.
the more difficult the travel is, we set the parameter β at the
Application Example 3: Different queries on Dual struc-
interval [0, 1]. The following small example gives a further
tures. The exit simplicial family in Figure 9 and the location
explanation of Equation 8.
simplicial family in Figure 6 represent a dual structure of the
Example 6. Observing Figure 10 that is the simplicial exit-location matrix (Table 3). For the exit simplicial family,
family corresponding to the area in Figure 5. Now a person each location is represented by a vertex and each exit is a
is standing at exit e1 and he wants to get to exit e2 . The simplex, whereas for the location simplicial family, each exit
question is which path is the optimal one for him. is a vertex and each location is denoted by a simplex. The
dual simplicial families reveal the connective structure infor-
From Figure 10, we know that there are two possible navi- mation in the same indoor environment from two different
gation routes from e1 to e2 , one is through the locations σL2 angles and thus can be used to deal with different queries.
1
2 1 1 Some example questions that arise might be:
and σL2 , the other one is through the locations σL3 and σL4 .
On the location simplicial family (Figure 6)
The lengths of the two paths are equal, but their difference
• Which exit of location 2 is nearest to the exit 5 of location
exists in the connective strength. If we set the parameter
3?
β = 21 , then we can get
• Is it possible for Joe to arrive at exit 2 directly from exit
2∗2 1 1 1?
CD(Lo1 , Lo2 ) = ∗ ( )1 = , (9)
3+3 2 3 On the exit simplicial family (Figure 9)
• Which location might Joe pass through if he walks to lo-
and
cation 2 from location 3?
2∗1 1 1 These three questions could be answered by both simplicial
CD(Lo3 , Lo4 ) = ∗ ( )1 = . (10)
2+2 2 4 families, but their efficiency is different.
According to this results, if the distances of these paths are
comparable, then the optimal path we think should be that 5.2 Comparison Analysis
2 2
with σL 1
and σL 2
since the larger connective degree means Existing location models can be classified into geometric
that the person can be easier to pass through the boundary or symbolic model. A geometric model represents the phys-
of two locations. In particular, this kind of decision is more ical space as a Euclidean space and the objects therein as
suitable for mobile environment. For example, suppose both points, lines, shapes and volumes in the Euclidean space. In
paths have huge flowing crowds, the person in the path with our model, the exit and the location (room or corridor) com-
2 2
σL 1
and σL 2
can choose an appropriate exit going through posed of line, shape, etc are replaced by vertex or simplex.
the location boundary according to the temporal condition, Therefore the topology-based model is symbolic in nature.
however the other path provides the only one choice. The set-based and graph-based models are two main types
of symbolic location models. The basic functions of location
Application Example 2: Structure Analysis. From the models include position queries (P), nearest neighbor queries
multilevel structure of Q-analysis, we can understand an (N), range queries and navigation (R). Based on the repre-
indoor environment comprehensively. Each connected com- sentation of the location simplex and the nearness relation
ponent at different levels provides us with the local connec- given in Definition 3, the containment relation is supported
tion information between two locations, including connective and therefore we can use the topology-based model to deal
strength and connective length. Moreover, the structure vec- with range queries, which is similar to the set-based model.
tor Qq in Q-analysis indicates the complexity of the global Since the graph-based model supports the definition of the
structure of a space. Suppose that there are two floors in a topological relation “connected to ” as well as the quanti-
Supported queries environments. We also plan to apply the model to large
scale location-based applications.
P R N
Set Good Good Basic
Graph Good Basic Good
Topology Good Good Good(Semantic)
7. ACKNOWLEDGEMENT
This work was supported by grants 616005 and CA05/
06.EG03 from Hong Kong Research Grants Council. We
Table 6: Comparison II of symbolic location models. would like to thank the anonymous reviewers for their in-
sightful comments and helpful suggestions for our prepara-
tion of the manuscript.
tative distance between locations, it is well-suited for near-
est neighbor queries as well as navigation. However, range
queries is not well supported by the graph-based model with 8. REFERENCES
standard graph-based algorithms. On the other hand, with [1] R. Atkin. An approach to structure in architectural
our topology-based model, which mines the deep topolog- and urban design: 1. introduction and mathematical
ical properties of the graph, not only the ”connected to ” theory. Environment and Planning B, 1:56–67, 1974.
relation but also the semantic distance can be captured, in
[2] R. Atkin. An approach to structure in architectural
which the semantic distance comprises the structure infor-
and urban design: 2. algebraic representation and
mation as well as the real distance information between two
local structure. Environment and Planning B,
locations. Therefore, it could be more suitable for nearest
1:173–191, 1974.
neighbor queries and navigation on the complicated indoor
[3] R. Atkin. Mathematical Structure in Human Affairs.
environment than the graph-based model. Table 5 and Ta-
Heinemann, London, 1974.
ble 6 summarize the properties of the set-based, graph-based
and topology-based location models. Detailed discussions [4] R. Atkin. An approach to structure in architectural
about the existing location models are given in [9]. and urban design: 3. illustrative examples.
We describe briefly the structures of graph theory and Q- Environment and Planning B, 2:21–57, 1975.
analysis here. A more comprehensive discussion is given in [5] R. Atkin. Combinatorial Connectivities in Social
[13] by Earl and Johnson. In essence, graph theory reveals Systems. Birkhauser, Basel, 1977.
the binary relation between two things, whereas Q-analysis [6] R. Atkin. Multidimensional Man. Penguin Books,
method generalizes this idea to n-ary relations (i.e., the re- Harmondsworth, Middx, 1981.
lations among n things ), which is more important for many [7] J. Beaumont. A description of structural change in a
practical problems. The graphical method has the advan- central place system: a speculation using q-analysis.
tage of enabling us to take in the whole structure at a glance, International Journal of Man-Machine Studies,
while the Q-analysis method provides us with a series of fil- 20:567–594, 1984.
ters for observing the structures at various dimensional levels [8] J.R. Beaumont and A.C. Gatrell. An Introduction to
[11]. The Q-analysis method is like a microscope and there- Q-Analysis. Geo Abstracts, Norwich, 1982.
fore makes us do much deeper investigations into structures [9] C. Becker and F. Dürr. On location models for
than the graph method. ubiquitous computing. Personal and Ubiquitous
Computing, 9:20–31, 2005.
[10] B. Brumitt and S. Shafer. Topological world modeling
using semantic spaces. In Proceedings of Location
6. CONCLUSION AND FUTURE WORK Modeling Workshop at Ubicomp, 2001.
[11] R. Davies. Q-analysis: a methodology for librarianship
In this paper, we build a topological semantic location
and information science. Journal of Documentation,
model for indoor applications based on an exit-location ma-
41(4):221–246, 1985.
trix. Semantic relationships and semantic distance within
an indoor environment can be revealed by this novel model. [12] F. Dürr and K. Rothermel. On a location model for
The Q-analysis method is used to analyze the semantic in- fine-grained geocast. In Proceedings of the 5th
formation of the model. Topological characteristics, such as International Conference on Ubiquitous Computing
connectiveness, hole, eccentricity and star-hub, are defined (UbiComp’03), pages 18–35, 2003.
to model n-ary relationships among entities as well as basic [13] C.F. Earl and J.H. Johnson. Graph theory and
relationships between two entities. The connected compo- q-analysis. Environment and Planning B: Planning
nents and the structure vector in Q-analysis of a simplicial and Design, 8(4):367–C391, 1981.
family provide us with local and global angles to observe [14] M.J. Egenhofer, A.U. Frank, and J.P. Jackson. A
the indoor space. Based on the information obtained by the topological data model for spatial databases. In
topological analysis, we explore the potential applications of Lecture Notes in Computer Science, volume 409, pages
the model and give a comparison analysis of the new model 271–286, 1989.
and existing location models. [15] P. Ganesan, H. Garcia-Molina, and J.Widom.
In this paper, we apply algebraic topology to semantic lo- Exploiting hierarchical domain structure to compute
cation model. The richness results in algebraic topology may similarity. ACM Transactions on Information
cast new light to developing location models for location- Systems, 21:64–93, 2003.
based services. In the future, we are interested in further [16] P. Gould. Letting the data speak for themselves.
exploring more useful information of the model that can Annals of the Association of American Geographers,
support the navigation and queries on complicated indoor 71:166–176, 1981.
Model Distance “Connected to” “Containment”
Set Limited Yes Good
Graph Good to very good Yes Limited
Topology Semantic Yes Good

Table 5: Comparison I of symbolic location models.

[17] P. Gould. A structural language of relations. Future 41–44, 2001.


Trends in Geomathematics, 1981. [33] T. Pederson. Object location modeling in office
[18] P. Gould, J. Johnson, and G. Chapman. The Structure environments - first steps. In Proceedings of Location
of Television. London: Pion Limited, 1984. Modeling Workshop at Ubicomp, 2001.
[19] H. Hu and D.L. Lee. Semantic location modeling for [34] S. Pradhan. Semantic location. Personal Ubiquitous
location-aware application in mobile and ubiquitous Computing, 4:213–216, 2000.
computing. Technical Report, Hong Kong University of [35] C. Schlieder, T. Vogele, and A. Werner. Location
Science and Technology, 2003. modeling for intentional behaviour in spatial
[20] H. Hu and D.L. Lee. Semantic location modeling for partonomies. In Proceedings of Location Modeling
location navigation in mobile environment. In Workshop at Ubicomp, 2001.
Proceedings of the 5th International Conference on [36] S. Steiniger, M. Neun, and A. Edwardes. Foundations
Mobile Data Management (MDM’04), pages 52–61, of Location Based Services. Lecture Notes on LBS,
2004. Department of Geography, University of Zürich, 2006.
[21] P. Iravani. Discovering relevant sensor data by [37] E. Valencia and J. L. Giavitto. Algebraic topology for
q-analysis. In Robot Soccer World Cup IX, pages knowledge representation in analogy solving. In
81–92, 2005. European Conference on Artificial Intelligence
[22] T. L. Jacobson, D.S. Fusani, and W. Yan. Q-analysis (ECAI98), pages 23–28, 1998.
of user-database interaction. International Journal of [38] J. Ye, L. Coyle, S. Dobson, and P. Nixon. A unified
Man-Machine Studies, 38(5):787 – 803, 1993. semantic space model. In Lecture Notes in Computer
[23] B. Jiang and I. Omer. Spatial topology and its Science, volume 4718, pages 103–120, 2007.
structure analysis based on the concept of simplicial
complex. to appear in Transactions in GIS, 2007.
[24] C. Jiang and P. Steenkiste. A hybrid location model
with a computable location identifier for ubiquitous
computing. In Lecture Notes in Computer Science,
volume 2498, pages 246–263, 2002.
[25] W. Kainz, M. Egenhofer, and I. Greasley. Modeling
spatial relations and operations with partially ordered
sets. International Journal of Geographical
Information Systems, 7:215–229, 1993.
[26] B.S. Karlsen. Enabling a Ubiquitous Location Based
Service on Campus. Master Thesis, Norwegian
University of Science and Technology, 2006.
[27] C.P. Kwong. Simplexes, Q-analysis, and Information
Retrieval. Technical Report CUHK-ACE-03-03, The
Chinese University of Hong Kong, 2003.
[28] D.L. Lee, M. Zhu, and H. Hu. When location-based
services meet databases. Mobile Information Systems,
1:81–90, 2005.
[29] U. Leonhardt. Supporting Location-Awareness in Open
Distributed Systems. PhD Thesis, Department of
Computing, Imperial College London, 1998.
[30] D.D. Li. Geometric and Topological Approaches to
Semantic Text Retrieval. PhD Thesis, The Chinese
University of Hong Kong, 2007.
[31] D.D. Li and C.P. Kwong. Understanding latent
semantic indexing : A topological structure analysis
using q-analysis method. In Proceedings of the 2007
International Conference on Information and
Knowledge Engineering, pages 110–116, 2007.
[32] T. O’Connell, P. Jensen, A. Dey, and G. Abowd.
Location in the aware home. In Proceedings of
Location Modeling Workshop at Ubicomp, pages

You might also like