The MindSpaces Knowledge Graph: Applied Logic and Semantics on

Indoor and Urban Adaptive Design

Evangelos A. Stathopoulos

1 a

, Alexandros Vassiliades

1,2 b

, Sotiris Diplaris

1 c

Stefanos Vrochidis

1 d

, Nick Bassiliades

2 e

and Ioannis Kompatsiaris

1 f

Information Technologies Institute, Center for Research and Technology Hellas, Thessaloniki, Greece

School of Informatics, Aristotle University of Thessaloniki, Greece

Keywords:

Knowledge Graph, Ontology, Reasoning Mechanism, Indoor/Urban Adaptive Design, Semantic Web.

Abstract:

The evolution of Knowledge Graphs (KGs), during the last two decades, has encouraged developers to create

more and more context related KGs. This advance is extremely important because Artiﬁcial Intelligence (AI)

applications can access open domain speciﬁc information in a semantically rich, machine understandable for-

mat. In this paper, we present the MindSpaces KG, a KG that can represent emotions-relevant and functional

design for the indoor and urban adaptive design. The MindSpaces KG can integrate emotional, physiologi-

cal, visual, and textual measurements, for the development of online adapting environments. Moreover, we

present a reasoning mechanism that extracts crucial knowledge from the MindSpaces KG, which can help

users in real-life scenarios. The scenarios were provided by experts.

1 INTRODUCTION

The evolution of Knowledge Graphs (KGs) in the last

20 years allowed developers to construct context re-

lated KGs (i.e., KGs that can be used only in speciﬁc

environments). The creation of context related KGs

seems to be the next step for allowing KGs to be-

come the main knowledge representation format for

the Web (Berners-Lee et al., 2001). Our focus is on

representing emotions-relevant and functional design

for the indoor and urban adaptive design. Emotions-

relevant refer to emotions created to individuals when

experiencing an indoor or urban area, such as stress,

calmness, happiness, among others. The functional

design of an indoor or urban location imply the practi-

cality of the location, for example movability of work-

ers in a workspace in the case of indoor environments,

or redesign-relocate of a non helpful bus station in the

case of urban environments.

The MindSpaces KG was developed in or-

der to work as the knowledge representation of

https://orcid.org/0000-0003-3713-5833

https://orcid.org/0000-0003-4569-503X

https://orcid.org/0000-0002-9969-6436

https://orcid.org/0000-0002-2505-9178

https://orcid.org/0000-0001-6035-1038

https://orcid.org/0000-0001-6447-9020

the MindSpaces project

. The motivation for the

MindSpaces KG stands in 2 different key points. The

ﬁrst is to improve urban design in a rapidly expanding

city by addressing new challenges that may arise re-

lated to its functionality, mobility attractiveness, pro-

tection of culture and environment. MindSpaces KG

will increase sensitivity and awareness towards the

cultural signiﬁcance and current issues of a city, re-

lated to the environment and mobility. While the

second, is to increase opportunities for positive so-

cial interaction in work environments which leads to

improved productivity and creativity across depart-

ments and teams, by helping to readjust workspaces

to achieve better aesthetics and functionality.

The MindSpaces KG is mostly oriented for

artists, designers and architects for redesigning indoor

workspaces in order for the workers to feel more com-

fortable, and to improve functionality, mobility and

overall attractiveness of areas. But other users can

also utilize the MindSpaces KG, as for instance the

ideas of citizens of a city can be very useful in re-

designing urban areas. Moreover, we provide a rea-

soning mechanism that aids users in real-life scenar-

ios which were provided by experts.

The problem we are addressing is the construc-

tion of a general KG, which can represent emotions-

https://mindspaces.eu/

334

Stathopoulos, E., Vassiliades, A., Diplaris, S., Vrochidis, S., Bassiliades, N. and Kompatsiaris, I.

The MindSpaces Knowledge Graph: Applied Logic and Semantics on Indoor and Urban Adaptive Design.

DOI: 10.5220/0011666600003393

In Proceedings of the 15th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2023) - Volume 3, pages 334-341

ISBN: 978-989-758-623-1; ISSN: 2184-433X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

relevant and functional design for indoor and urban

adaptive design, by mapping emotional, physiologi-

cal, visual, and textual measurements. Moreover, we

address the fact that a user in order to access the in-

formation inside the KG in a convenient way, must be

provided with a reasoning mechanism based on real-

life scenarios. The real-life scenarios ideally should

be formulated by domain experts.

The scalability of the MindSpaces KG is wide, as

it can represent information for any urban area and

indoor workspace. More speciﬁcally, for the indoor

workspaces the MindSpaces KG is mostly oriented

for ofﬁces, but also other indoor workspaces, such as a

cottage industry, and can also exploit the information

inside the MindSpaces KG. A natural extension of it

could be to extend representing emotions relevant to

outdoor workspaces, and recommend tasks to citizens

on how they can increase the functionality, mobility,

and attractiveness of an urban area with respect to the

cultural signiﬁcance and environment.

The problem tackled with this semantic frame-

work has to do with the different nature of data

present in the system. Several approaches can be

found that address the knowledge representation with

ontologies in each domain distinctly. The motivation

behind this work is that unique semantic requirements

needed to be addressed altogether and combine inter-

disciplinary domains in one unique ontology, tied to-

gether carefully with precise and meaningful custom

interconnections to serve multimodal knowledge rep-

resentation and enable smart reasoning mechanisms.

Our contribution with this paper, is on one hand

the MindSpaces KG which can represent multi-modal

measurements, which in turn help artists with the em-

pirical and pragmatic perception of actual occupants,

so as to drive the development of unconventional so-

lutions in the design of spaces. On the other hand,

is the reasoning mechanism for the MindSpaces KG,

which proves to be helpful in real-life scenarios.

The rest of this paper is organized as follows. Sec-

tion 2, contains the related work. Next, in Section 3

we present the MindSpaces KG, the reasoning mech-

anism, and the data upon we constructed the KG. Sec-

tion 4, contains the evaluation of the KG and the rea-

soning mechanism. We conclude our paper with Sec-

tion 5.

2 RELATED WORK

The study has two main directions, the MindSpaces

KG, and the reasoning mechanism which extracts

knowledge from the MindSpaces KG. For this reason,

we will separate our related work into two main sub-

sections, one for similar knowledge graphs, and one

for the domain speciﬁc reasoning mechanisms from

knowledge graphs.

Knowledge Graphs: The ﬁrst category of knowl-

edge graphs that can be considered close to the

MindSpaces KG, are KGs for arts and artists. These

KGs fall into the category of KGs for cultural her-

itage (Hyv

onen, 2012; Schneider, 2020). Moreover,

the area of KGs about arts and artists is not so rich,

as there are not many studies that could be classiﬁed

clearly in this domain. One exception, is the study of

Raven et al. (Raven et al., 2020), where the authors

present a KG that can represent the steps of speciﬁca-

tion, conceptualization, integration, implementation

and evaluation in a case study concerning ceramic-

glass. The difference with MindSpaces KG, can be

noticed, as we offer a KG focused on the knowl-

edge representation of emotions-relevant and func-

tional design for the indoor and urban adaptive de-

sign.

Thereinafter, the area of KGs for architecture and

architects seems to be quite richer. (Lopes, 2007),

showcases the main notions that must be represented

in a KG about architectural concepts. On the other

hand, KGs like (Wagner and R

uppel, 2019; Kumar

et al., 2019), which are considered in the area of ar-

chitecture mostly contain information about materials

of objects and their uses. For the former, we noticed

it is a theoretical study, while we offer a constructed

KG. For the latter three KGs, our KG does not contain

similar type of information, as we offer knowledge

about different interdisciplinary distinct domains.

Reasoning: The area of reasoning over KGs, is

quite rich and it was enhanced in the last decade with

the constant evolution of KGs. The reason why KGs

are very helpful in retrieving information, is because

they have a predeﬁned format, which is easily under-

standable by the machine, and their terminology and

assertion components can be deﬁned based on a set of

rules which can represent commonsense knowledge.

Even though many general techniques have been pre-

sented (Munir and Anjum, 2018; Asim et al., 2019;

Yu, 2019), there is lack of studies for reasoning over

KGs restricted to a speciﬁc domain. The reason why

is hard to create a reasoning mechanism over KGs

which are restricted to a speciﬁc domain, is because

domain experts are needed in order to create real-life

scenarios, based on which the reasoning mechanism

should be constructed (as mentioned in (Chi et al.,

2019)). (Vassiliades et al., 2020) is an exception as a

reasoning mechanism for the household environment

is presented, but in contrast to our study the scenarios

presented are not a result of knowledge provided by

domain experts.

The MindSpaces Knowledge Graph: Applied Logic and Semantics on Indoor and Urban Adaptive Design

335

3 THE MindSpaces KNOWLEDGE

GRAPH

The MindSpaces KG is part of the MindSpaces

project. Therefore, the MindSpaces KG communi-

cates with other components of the project. We men-

tion this because some parts of the mapping mech-

anism, which receives messages, in JSON format,

from the visual, textual, and stress analysis compo-

nent will not be analyzed in detail in this paper. But

the source code of the mapping mechanism can be

found here

. The idea of the pipeline is that after the

mapping mechanism has received the message from a

component, it will map the information into the KG.

Then, any user can hand-pick from a set of predeﬁned

SPARQL queries, in order to access the information

in the KG. Notice that all SPARQL queries were de-

ﬁned with the help of domain experts, and some of

them also require an image as an input in order to cast

the SPARQL query. Figure 1, shows an outline of the

pipeline, each number in the circles shows the order

of steps.

Figure 1: Pipeline of the MindSpaces KG.

3.1 Nature of Data

The multi-modality and variety of data ﬂowing in the

system and the necessity of homogenization and fu-

sion mandated the adoption of a semantic knowledge

graph to address the requirements of the project. The

knowledge graph is not responsible for archiving and

storing raw data ﬁles, since there is an underlying

data storage facility for that purpose. Instead, the KG

hosts metadata of raw data, analyses results and mis-

cellaneous information with semantic value among

other candidates for being mapped and fused into the

knowledge base. One can ﬁnd a blueprint of the mes-

sages fused in the KG here

. The KG is accompanied

with novel ontological models to achieve proper se-

mantic annotation of the raw data.

The main categories of data which needed to be

captured in the knowledge graph were: physiological

analysis results from galvanic skin responses (GSR),

https://github.com/valexande/MindSpacesPUC1-2

visual analysis results from images, textual analysis

results derived from online retrieved content, and gen-

eral information about VR experiments. Some of the

aforementioned analyses results share some common

relations such as in the case of the imageability met-

ric.

The physiological signals (GSR) were captured

during hot-spot or navigation VR experiments where

multiple users conducted stress-induced tasks inside

virtual environments containing multiple conﬁgura-

tions of work ofﬁces. The ultimate goal was to obtain

and assess stress indicators directly from the skin of

the subjects while experiencing different setups of the

space.

The visual analysis component consists of several

machine learning models either trained from scratch

or ﬁne-tuned from other efforts. The model perform-

ing semantic segmentation (Qiu et al., 2021) on im-

ages was trained to extract semantic labels and per-

centages per pixel on images while the Verge clas-

siﬁer (Andreadis et al., 2020) was deployed to clas-

sify the images to one or more classes based on con-

text. The valence-arousal model for exterior design

was trained on a newly collected dataset annotated by

experts to deliver conﬁdence and values on a happy-

unhappy and calm-excited scale, while a visual im-

ageability score is generated (Pistola et al., 2022).

3.2 The MindSpaces Knowledge Graph

The MindSpaces KG is separated into two big sub

graphs. One for the purpose of representing knowl-

edge that aims at improving urban design in a rapidly

expanding city by addressing new challenges that

may arise related to its functionality, mobility, attrac-

tiveness, protection of culture and environment. The

other aims at representing knowledge that involves in-

creasing opportunities for positive social interaction

in work environments which leads to improved pro-

ductivity and creativity across departments and teams,

by helping to readjust workspaces to achieve better

aesthetics and functionality.

The subgraph for the ﬁrst case was constructed

based on the information we received from the var-

ious components. In more detail, we analysed the

messages that we received from the textual and vi-

sual analysis components, and we developed a KG

based on those. In the corresponding part of the on-

tology we represent the important concepts of these

messages through classes and relations among them.

Moreover, we analysed the requirements of the rea-

soning mechanism, meaning that we took into consid-

eration the competency questions given by users and

experts, and deﬁned classes and relations in such a

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

336

way that would help the reasoning mechanism return-

ing the crucial information. Notice that a competency

question, is a question which as a user we would like

to be answered by the KG, with the information that

it contains. In Figure 2 one can see the classes and the

object type properties of the ﬁrst subgraph.

Figure 2: The schema of the ﬁrst subgraph of the

MindSpaces KG.

The namespace mind1 is used to indicate the

classes and relations for the ﬁrst subgraph of the

MindSpaces KG. Next, we will give a detailed analy-

sis of the classes and the relations between them.

• The Sentence class contains information about

the sentences that compose a textual description

of an image. For each sentence it has informa-

tion about the emotional tag that was given by the

user, and the conﬁdence of each emotional tag.

The emotional tag is the sentiment label that was

given by the user to a sentence, such as positive,

negative, and neutral.

• The Text class contains information about the tex-

tual description of an image. The Text class is

connected through the property hasSentence with

the class Sentence, in order to give further infor-

mation about the sentences that compose the tex-

tual description. Moreover, it has information for

the language of the textual description, and the

textual description itself. Finally, the Text class

is connected through the property hasNer with the

Ner class, which has information about named en-

tities found in the textual description. Named en-

tities can be words that refer to real life objects,

actions, or activities.

• The Ner class contains information about the

named entities found in the textual description of

an image. The Ner class gives information for

the category of the named entity relation (the cat-

egory of a named entity is a classiﬁcation that

was given by the domain experts), the imageabil-

ity score of the named entity, and is connected

through the property hasURI with the class URI.

The imageability score is a conﬁdence score that

is composed by the visual analysis component.

Imageability according to the urban planner Kevin

Lynch is the quality of a physical object to evoke

a strong image in any observer, thus being memo-

rable (Lynch et al., 1960).

• The URI class contains information about the

URIs of the named entities. Currently, the URIs

point only to DBpedia entities. The URI class has

information about the URI link and the conﬁdence

that a named entity should be related with a spe-

ciﬁc URI.

• The VergeLabel class contains information about

the labels of the verge classiﬁcations found in an

image.

• The VergeContainer class contains information

about the imageability scores of the verge classi-

ﬁcations found in an image. Moreover, it is con-

nected though the property isVergeLabel with the

class VergeLabel, in order to indicate the label of

a verge classiﬁcation.

• The SemSegLabel class contains information

about the labels of the segmented objects found

in an image.

• The SemSegContainer class contains informa-

tion about the imageability scores of the seg-

mented objects found in an image, the percent-

age of space they capture in the image, and the

conﬁdence that they are part of the image. More-

over, it is connected though the property hasSem-

SegLabel with the class SemSegLabel, in order

to indicate the label of a segmented object.

• The Arousal class contains information about the

arousal score that was given by a user for an im-

age. Arousal is a conﬁdence score given by the

users.

• TheValence class contains information about the

valence score that was given by the user for an

The MindSpaces Knowledge Graph: Applied Logic and Semantics on Indoor and Urban Adaptive Design

337

image. Valence is a conﬁdence score given by the

users.

• The Image class is the most important class, as

it contains a lot of metadata information about

the characteristics of the image, such as the lati-

tude, longitude, the pitch, the zoom, and others.

But apart from these it is connected: (i) with the

Arousal class through the property hasArousal

to indicate its arousal, (ii) with the Valence

class through the property hasValence to indi-

cate its valence, (iii) with the VergeContainer

class through the property isVergeContainer to

give information about the verge classiﬁcations

that it has, (iv) with the SemSegContainer class

through the property isSemSegContainer to give

information about the segmented objects it con-

tains, and (v) with the Text class through the prop-

erty hasText to give information about the textual

description that it has.

In Figure 3 one can see the classes and the object

type properties of the second subgraph.

Figure 3: The schema of the second subgraph of the

MindSpaces KG.

The namespace mind2 is used for the second sub-

graph. Next, follows a detailed analysis of the classes

and the relations of the second subgraph.

• The Collection class contains information about

the experiment status data change, meaning that it

indicates when the user run has started, stopped,

and if it goes from state ON to state OFF.

• The ExperimentType class contains information

about the experiment type data changes, meaning

that it indicates if the experiment type is a naviga-

tion task, a navigation selection, hot spot experi-

ment, or it does not have a type.

• The Navigation class contains information about

the navigation conﬁguration, meaning that it con-

tains information about the type of the conﬁg-

uration, and the timestamp that the node was

captured. Moreover, it is related through the

property hasNode with the class NavigationNode

that contains information about the navigation in-

stances (i.e., navigation nodes).

• The NavigationNode class contains information

about the navigation nodes, such as the x, y, z co-

ordinates of the node, the fusion score, and the

GSR score. The fusion and the GSR score, are

some conﬁdence score given by the visual and

stress level analysis components.

• The ItemKey class contains information about

the item key information, such as item key label,

the collaboration, focus work, overall design, pri-

vacy, and stress scores for each item key. All of

the aforementioned scores, are some conﬁdence

scores given by the users.

• Finally, the UserRun class is the most impor-

tant class, as it connects the information from the

aforementioned classes with a user run. More

speciﬁcally the UserRun class is connected: (i)

with the ExperimentType class through the prop-

erty hasExperimentType to indicate the experi-

ment types that it contains, (ii) with the ItemKey

class through the property hasItemKey to indi-

cate the item keys that it contains, (iii) with the

Collection class through the property hasCollec-

tion to give information about the collections that

it contains, and (iv) with the Navigation class

through the property hasNavigation to give infor-

mation for navigation nodes that it contains.

Notice that the two subgraphs eventhough they re-

fer to different notions, i.e., the ﬁrst subgraph is for

the outdoor adaptive design and the second for the

redesign of internal workspaces, they are related be-

tween them with the property hasImage. The prop-

erty hasImage has domain the class Collection and

range the class Image. The reason for that is because

a collection is a set of images.

3.3 Reasoning and Logic

The reasoning mechanisms take advantage of the

MindSpaces KG which was created to support the use

cases of urban environments and indoor workspaces,

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

338

as well as of the population of the knowledge base

with content and metadata deriving from both artists

and users.

The main idea, for the reasoning over the ﬁrst sub-

graph (see subsection 3.2), was to feed on demand an

interactive google map with geolocated 2D points in

the form of CSV ﬁles corresponding to image entries

inside the knowledge base through application pro-

gramming interfaces (APIs) developed in node.js and

Javascript. The graphdh.js library was used to estab-

lish the connection and transactions towards and from

the GraphDB repository

with authorization and au-

thentication ensured. The ﬁle delivered in response

follows a scalable and dynamic approach, meaning it

is being generated on demand based on live requests,

thus ensuring always up-to-date data delivery as the

knowledge base supports a continuous online popula-

tion and new entries may arrive anytime. The format

of the ﬁle consists of as many lines as the images fulﬁl

the SPARQL queries and 5 columns:

• latitude (the latitude when the corresponding im-

age was captured)

• longitude (the longitude when the corresponding

image was captured)

• point size (the size of the circle to be depicted on

the map)

• point opacity (the opacity of the circle to be de-

picted on the map)

• point colour (the colour of the circle to be depicted

on the map)

In total, 16 SPARQL queries were formulated,

each satisfying a different user requirement, followed

by an additional multipurpose sparse function. But

due to space restrictions only two Scenario A and

Scenario B, which were deﬁned by domain experts

as real-life scenarios, will be analyzed.

Scenario A: Given a snap image, return a list of

images, where: (i) the Top 3 segmentation labels of

the snap (based on the coverage percentage) exist in

the images, (ii) the images must have imageability ≥

imageability of snap + 0.05, (iii) the Top 3 segmen-

tation labels must exist in the images with the same

coverage percentage, or a 20% difference, and (iv) the

results must be limited to 8 images, if there exists as

many, sorted based on their imageability.

Scenario B: Given a snap image, return a list of

images, where: (i) the Top 3 segmentation labels of

the snap (based on the colorfulness percentage) ex-

ist in the images, and their colorfulness is above 1,

and (ii) for each one of the 3 segmentation classes

http://160.40.52.169:6161

bring the Top 5 images with the highest colorfulness

for each segmentation classes.

Notice that the queries which are refered in this

subsection are also Competency Questions (CQs), ex-

cept from Scenario A and Scenario B which are a

combination of CQs.

4 EVALUATION

The evaluation of the MindSpaces KG was twofold.

On the one hand, we evaluated the consistency and

completeness of the MindSpaces KG; we did this

with two different evaluation methods. Firstly, we

evaluated the completeness of the MindSpaces KG,

by deﬁning a set of CQs that the KG must be able

to answer with the information it contains (subsec-

tion 4.1). Secondly, we evaluated the consistency of

the MindSpaces KG by testing if it obeys a set of

SHACL constrains (subsection 4.1). On the other

hand, the evaluation of the reasoning mechanism

was performed by computing the precision-recall-F1

scores used for reasoning systems (subsection 4.2).

4.1 Competence and Consistency of the

Knowledge Graph

The completeness of the MindSpaces KG was evalu-

ated through a set of CQs assembled during the for-

mation of the ofﬁcial ontology requirements speci-

ﬁcation document (ORSD) (Su

arez-Figueroa et al.,

2009). For this reason, before constructing the

MindSpaces KG, we asked from users to deﬁne

a set of questions that they would like from the

MindSpaces KG to contain as knowledge and be able

to answer. The users were architects from the School

of Architects of the Aristotle University of Thessa-

loniki

, either undergraduate-master students or pro-

fessors, and architects or designers from Zaha Hadid

Architects

. In total a number of 83 CQs was col-

lected, the complete list of CQs can be found here

Based on the fact that we constructed the

MindSpaces KG on the aforementioned set of CQs,

this means that if any of the CQ is translated into

a SPARQL counterpart, our KG would answer the

question with the information it contains. For this

reason, we translated each CQ into a SPARQL coun-

terpart and we expected to return the desired infor-

mation. The completeness of the MindSpaces KG

was found adequate, as each CQ when translated into

https://www.auth.gr/school/arch/

https://www.zaha-hadid.com/

The MindSpaces Knowledge Graph: Applied Logic and Semantics on Indoor and Urban Adaptive Design

339

a SPARQL counterpart returned the desired informa-

tion.

Additionally to the CQs, we performed a valida-

tion procedure in order to inspect the syntactic and

structural quality of the metadata in the KB and to

check the consistency of them. The consistency of

the MindSpaces KG was found adequate, as out of 12

SHACL rules, from which 4 referred to object type

properties and 8 to data type properties, none of them

returned any invalidation of the rule. Moreover, we

checked if instances exist which belong to intersec-

tion of classes, as we did not desire such a case, and

there were not any.

4.2 Knowledge Retrieval Metrics

The evaluation of the reasoning mechanism was con-

ducted using the precision, recall and F1-score used

for reasoning systems (Equations 1, 2 and 3), over the

two real-life scenarios presented in 3.3 (i.e., Scenario

A and Scenario B).

precision =

|{RelevantInstance} ∩ {RetrievedInstance}|

|{RetrievedInstance}|

(1)

recall =

|{RelevantInstance} ∩ {RetrievedInstance}|

|{RelevantInstance}|

(2)

F1 = 2 ∗

recall ∗ precision

recall + precision

(3)

Retrieved Instances are considered all the images

for which the reasoning mechanism, did not return an

error when we casted a question to retrieve informa-

tion for them.

Relevant Instances are considered all the images

for which the reasoning mechanism, managed to re-

turn some information, when we casted a question to

retrieve information for them.

We denote by Retrieved

, Retrieved

the number

of retrieved images for Scenario A and Scenario B,

respectively. Relevant

, Relevant

are the numbers

of relevant images for Scenario A and Scenario B, re-

spectively. Next, precision

, precision

are the pre-

cision scores for Scenario A and Scenario B, recall

recall

are the recall scores for Scenario A and Sce-

nario B, and F1

, F1

are the F1 scores for Scenario

A and Scenario B, respectively.

The dataset on which we evaluated our reasoning

mechanism contains a set of 1200 images, and can be

found here

. All images were considered Retrieved,

meaning our reasoning mechanism did not return any

error. Thus, Retrieved

= Retrieved

= 1200. The

same does not hold for the relevant images, for both

scenarios, as for Scenario A the Relevant

images

were 1157, and for Scenario B the Relevant

images

were 1142.

Based on the aforementioned numbers the preci-

sion, recall and F1-scores for both scenarios can be

found in Table 1. Notice, the results are rounded to

four decimals.

Table 1: Precision, recall and F1-scores for Scenario A and

Scenario B.

Precision Recall F1

Scenario A 0.9649 1.0 0.9821

Scenario B 0.9632 1.0 0.9812

5 DISCUSSION & CONCLUSION

In this paper, we presented the MindSpaces KG,

a KG for representing emotions-relevant and func-

tional design for indoor and urban adaptive design.

The MindSpaces KG is populated with emotional,

physiological, visual, and textual measurements, for

the development of adapting environments. It is

mostly oriented for artists, designers and architects,

and its purpose on one hand is for redesigning indoor

workspaces for the workers to feel more comfortable,

and on the other hand to improve functionality, mo-

bility, attractiveness of spaces by taking into respect

the cultural heritage and environment. But other users

can also utilize the MindSpaces KG, for instance the

ideas of citizens of a city can be very useful in re-

designing urban areas. Moreover, the MindSpaces

KG offers a reasoning mechanism to access the infor-

mation in the KG in a convenient way. For this reason,

it retrieves helpful information for real-life scenarios.

The scenarios that were used to develop the reasoning

mechanism, were provided by experts.

The ﬁnal step was to evaluate the completeness,

the consistency and the reasoning mechanism of the

MindSpaces KG. The completeness of the KG (Sec-

tion 4.1) was evaluated with CQs, which were col-

lected by domain experts. Then, we translated each

CQ into a SPARQL counterpart, and checked each

one’s results, proving that our KG is able to provide

information to all users. The consistency of the KG

(Sections 4.1) was evaluated with a set of custom con-

straint rules created and enforced upon the KG where

no violation warnings were detected.

The high F1 scores achieved both for Scenario A

(98.21%) and Scenario B (98.12%), show that it can

be used as an individual mechanism for helping users,

by providing insightful information. Additionally, for

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

340

the relevant instances that were missed, both for Sce-

nario A and Scenario B, we can comment that this

happened because these two scenarios require an im-

age as an input, which is analyzed and the Top-3 seg-

mented labels (i.e., Top-3 object labels that are most

likely contained in the image) are considered in order

to ﬁnd similar images from the KG (see subsection

3.3). Therefore, the missed relevant images contained

Top-3 segmented labels which did not exist in any im-

age in our KG simultaneously.

As for future work, our plan is to enrich the

MindSpaces KG with domain-knowledge from rele-

vant Semantic Web KGs, such as ConceptNet (Speer

et al., 2017) and WordNet (Fellbaum, 2010), and

compare the knowledge in the MindSpaces KG with

other existing KG related to the indoor and urban

adaptive design. Moreover, we will investigate and

expand the quantity of real-life scenarios that the rea-

soning mechanism can support. Finally, we also plan

to create a more friendly user interface, as at the mo-

ment the queries are formulated through SPARQL.

ACKNOWLEDGEMENTS

This work has been supported by the EC-funded

project MindSpaces (H2020-825079)

REFERENCES

Andreadis, S., Moumtzidou, A., Apostolidis, K., Gkoun-

takos, K., Galanopoulos, D., Michail, E., Gialam-

poukidis, I., Vrochidis, S., Mezaris, V., and Kompat-

siaris, I. (2020). Verge in vbs 2020. In International

Conference on Multimedia Modeling, pages 778–783.

Springer.

Asim, M. N., Wasim, M., Khan, M. U. G., Mahmood, N.,

and Mahmood, W. (2019). The use of ontology in

retrieval: a study on textual, multilingual, and multi-

media retrieval. IEEE Access, 7:21662–21686.

Berners-Lee, T., Hendler, J., and Lassila, O. (2001). The

semantic web. Scientiﬁc american, 284(5):34–43.

Chi, N.-W., Jin, Y.-H., and Hsieh, S.-H. (2019). Developing

base domain ontology from a reference collection to

aid information retrieval. Automation in Construction,

100:180–189.

Fellbaum, C. (2010). Wordnet. In Theory and applications

of ontology: computer applications, pages 231–243.

Springer.

Hyv

onen, E. (2012). Publishing and using cultural heritage

linked data on the semantic web. Synthesis lectures on

the semantic web: theory and technology, 2(1):1–159.

Kumar, V. R. S., Khamis, A., Fiorini, S., Carbonera, J. L.,

Alarcos, A. O., Habib, M., Goncalves, P., Li, H., and

Olszewska, J. I. (2019). Ontologies for industry 4.0.

The Knowledge Engineering Review, 34.

Lopes, D. M. (2007). Shikinen sengu and the ontology of

architecture in japan. The Journal of aesthetics and

art criticism, 65(1):77–84.

Lynch, K. et al. (1960). The image of the city (vol. 11).

Munir, K. and Anjum, M. S. (2018). The use of ontolo-

gies for effective knowledge modelling and informa-

tion retrieval. Applied Computing and Informatics,

14(2):116–126.

Pistola, T., Georgakopoulou, N., Shvets, A., Chatzis-

tavros, K., Xefteris, V.-R., Garc

ıa, A. T., Koulalis, I.,

Diplaris, S., Wanner, L., Vrochidis, S., et al. (2022).

Imageability-based multi-modal analysis of urban en-

vironments for architects and artists. In International

Conference on Image Analysis and Processing, pages

198–209. Springer.

Qiu, W., Li, W., Liu, X., and Huang, X. (2021). Subjective

street scene perceptions for shanghai with large-scale

application of computer vision and machine learning.

Technical report, EasyChair.

Raven, D., de Boer, V., Esmeijer, E., and Oomen, J. (2020).

Modeling ontologies for individual artists. Vrije Uni-

versiteit Amsterdam.

Schneider, A. (2020). Alternatives: World ontologies and

dialogues between contemporary arts and anthropolo-

gies. In Alternative Art and Anthropology, pages 1–

26. Routledge.

Speer, R., Chin, J., and Havasi, C. (2017). Conceptnet 5.5:

An open multilingual graph of general knowledge. In

Thirty-ﬁrst AAAI conference on artiﬁcial intelligence.

arez-Figueroa, M. C., G

omez-P

erez, A., and Villaz

on-

Terrazas, B. (2009). How to write and use the

ontology requirements speciﬁcation document. In

OTM Confederated International Conferences” On

the Move to Meaningful Internet Systems”, pages

966–982. Springer.

Vassiliades, A., Bassiliades, N., Gouidis, F., and Patkos, T.

(2020). A knowledge retrieval framework for house-

hold objects and actions with external knowledge. In

International Conference on Semantic Systems, pages

36–52. Springer, Cham.

Wagner, A. and R

uppel, U. (2019). Bpo: The building

product ontology for assembled products. In Proceed-

ings of the 7th Linked Data in Architecture and Con-

struction workshop (LDAC 2019)’, Lisbon, Portugal,

page 12.

Yu, B. (2019). Research on information retrieval model

based on ontology. EURASIP Journal on Wireless

Communications and Networking, 2019(1):1–8.

The MindSpaces Knowledge Graph: Applied Logic and Semantics on Indoor and Urban Adaptive Design

341