International Conference on Cultural Informatics, Communication & Media

Studies

Vol 1, No 1 (2020)

Proceedings of The International Conference on Cultural Informatics, Communication & Media 2019-CICMS2019

Towards implementing an AI chatbot platform for

museums

Savvas Varitimiadis, Konstantinos Kotis, Andreas

Skamagis, Alexandros Tzortzakakis, George Tsekouras,
Dimitris Spiliotopoulos

doi: 10.12681/cicms.2732

https://epublishing.ekt.gr | e-Publisher: EKT | Downloaded at: 09/05/2024 18:16:44

Powered by TCPDF (www.tcpdf.org)

2nd International Conference on Cultural Informatics, Communication & Media Studies

Towards implementing an AI chatbot platform for museums

Varitimiadis S.1, Kotis K.2, Skamagis A1. Tzortzakakis A.1, Tsekouras G.2, Spiliotopoulos D. 3
1
Aegean Solutions SA
Milatou 14, Iraklio 712 02, Greece
{svaritimiadis, atzortzakakis, askamagis}@aegeansolutions.com
2
University of the Aegean,
Dept. of Cultural Technology and Communication
Intelligent Systems Lab
University Hill, Mytilene, 81100 Greece
{kotis, gtsek}@aegean.gr
3
University of the Peloponnese
Department of Informatics and Telecommunications
Tripoli, Greece
[email protected]
Abstract

Recently, understanding their unique role in storytelling and aiming to attract more visitors, several museums have
integrated modern ICT technologies. The problem with these technologies however is that gradually tend to be of
no real interest to visitors, lack of significant interaction, cannot be continuously updated, and eventually distract
visitors from experiencing the exhibits. Museum visitors do not need to be impressed by a technological application
but need to learn about the stories of the exhibits in a creative, human-centered and interactive manner. This paper
presents an ongoing work towards implementing a new interactive technological trend for museums, i.e., a museum
chatbot platform, namely MuBot. The MuBot platform aims to provide museums the opportunity to create simple,
interactive and human-friendly apps for their visitors. Such apps will integrate an intelligent chatbot that uses some
of the most advanced AI technologies of Machine Learning, Natural Language Processing/Generation, and the
Semantic Web. Museum visitors will be able to use a chatbot application that will be created through the MuBot
platform, to chat with a ‘smart’ exhibit. They will be able to ask questions through text or voice (in natural
language) and receive audible or written answers. The more the visitors ask, the more MuBot will learn and store
new knowledge in its knowledge base. The paper presents a preliminary design of the proposed MuBot platform,
experimenting with first prototype implementations using the well-known Dialogflow framework, as well as using
a Knowledge Graph-based approach.

Keywords:
Museums, Chatbots, Artificial Intelligence, Machine Learning, Conversational Systems

1. Introduction
Museums are complex organisations as they carry the responsibility to gather, document,
protect and preserve artefacts from our past, artefacts that carry ideas and practices that prove
our intelligence. Also have the responsibility to present and carry this knowledge to all the
members of the community in a creative, human-centered and interactive manner.
Recently, several museums, understanding their unique role in storytelling and at the same
time aiming to attract more visitors, have integrated modern ICT technologies. Doing so, their
visitors are now able to walk around exhibits using not only the assistance of a human or audio
guide, but also using high-end technologies such as digital guides, smart mobile apps, virtual
and augmented reality devices/glasses, smart tags, and more. The problem with these modern
ICT technologies however is that, after the first impression and use, they tend to be of no real
interest to visitors, since they either distract them from the exhibits or their application lack of
significant human-like/natural interaction. In addition, these technologies cannot be

1
continuously and promptly updated, and become less functional and more disturbing over time.
It is conjectured that museum visitors do not need to be impressed by a technological
application, but rather need to learn about the stories of the exhibits in a creative, human-
centered, human-like and interactive manner.
The aim of our ongoing research work is to use AI technologies to develop intelligent
applications for supporting museum visitors before, during and after their visit to a museum. In
this paper our preliminary efforts are presented towards implementing a new interactive
technological AI trend for museums, i.e., a Museum chatbot platform, namely MuBot.
The motivation of our work is the real need to go beyond the current status in museums’
technological advances by introducing a new way of attracting visitors to exhibitions, by
making technologies friendly and useful, and by turning museums into an interactive learning
and social economic development environment. To do this, the proposed MuBot technological
initiative will bridge the following gaps:
 the lack of communication and continuous human-centered/human-like interaction of
modern museums with their visitors,
 the limited visibility of museum content to the public through high technological
applications, and
 the lack of exploiting the content of museums as a key of economic development for
both the local community and the global community.
The main contribution of our work is to enable museum curators and administrators to create
simple, interactive and human-friendly chatbot apps for their visitors with the support of the
MuBot platform. Intelligent chatbots that use some of the most advanced Artificial Intelligence
(AI) technologies of today such as Machine Learning (ML), Natural Language Processing
(NLP) and Semantic Web (SW), will be designed and implemented. The visitors will be able to
use a chatbot application that will be created through the MuBot platform, to chat with a ‘smart’
exhibit when they are in front of (or close to) it. They will be able to ask questions through text
or voice (in natural language) and receive audible or written answers with the use of AI
technology. The more the visitors ask, the more MuBot learns and stores new knowledge in
platform’s knowledge base.
The MuBot platform is currently in the design phase, however, an experimental ML-based
chatbot using Google’s Dialogflow has been implemented for evaluation of the related
technology1. In addition, we are already experimenting with an alternative approach (as another
MuBot instantiation) based on SW technology (Ontologies and Knowledge Graphs). Our vision
is to combine the advantages of both technologies with NLP/NLG technology, towards a state-
of-the-art chatbot platform for museums.
The rest of the paper is structured as follows: Section 2 presents the related work concerning
the use of AI chatbots technologies with specific examples from museums. Also, this section
provides a brief introduction of the basic terminology of the AI technologies that are under
research for integrating in the MuBot platform. The section concludes with the recording of the
requirements of a museum chatbot application and platform. Section 3 presents the MuBot
platform, its purpose, the special features, and the architectural design. Also, in this section, the
Cretan MuBot is presented, the first experimental chatbot of the MuBot Platform, which will
be evaluated in the Heracleum Archaeological Museum. Section 4 discusses future steps in
MuBot platform development, Finally, section 5 concludes the paper.

1
http://tiny.cc/d56a6y

2
2. Related Work and Technologies
In this section Artificial Intelligence technologies that have been used in the development of
chatbots are briefly presented. Also, this section includes a short review on the chatbot
applications that have been implemented for museums.
2.1. Artificial Intelligence technologies and chatbots
Chatbots are rapidly evolving AI applications used by several companies and organizations all
over the world in order to engage and serve the users’ needs by exploiting the most common
skill of them, which is chatting. A short selected (from the literature) definition of a chatbot is:
“Chatbot (or chatterbot, Bot, smartbot, Conversational interface or Artificial Conversational
Entity) is a computer program or an artificial intelligence which conducts a conversation via
auditory or textual methods” (Bilange, 1991; Vassos, Malliaraki, Falco, Maggio, Massimetti,
Nocentini & Testa, 2016; Følstad & Brandtzaeg, 2017; Valtolina, Barricelli, Gaetano &
Diliberto, 2018). Their major advantage is that they can serve their users anytime and if they
get smart enough, they can provide a satisfactory human-like assistance. Chatbot applications
are using advanced AI technologies, as presented in the following subsections.
2.1.1 Machine Learning (ML)
Machine Learning is the field of AI science that focuses on getting machines to “learn” and to
continually develop autonomously. ML utilizes supervised or unsupervised algorithms, such as
decision trees, neural networks, deep learning and others, that enable computer systems to
optimize the ways of problem solving (Sukhbaatar, Weston & Fergus, 2015; Serban, Sordoni,
Bengio, Courville & Pineau, 2016, Wehle, 2017). For chatbots, ML is used to understand what
users are saying, learning from examples provided for training the system on what a user might
say when interacting with the chatbot, analyzing and understanding user intent in real-time.
2.1.2 Natural Language Processing (NLP)
Understanding natural language is the most difficult task of an AI chatbot system, as language
is an evolving entity with different levels of complexity. NLP allows computers to segment,
assign meaning, and analyze human communication in its natural forms. NLP is a process that
uses several AI techniques such as entity recognition, relationship extraction, language parsing,
sentiment analysis, speech recognition and others in order to become effective and natural
(Orth, 2017; Rivero, 2018).
2.1.3 Semantic Web and Ontologies
As Tim Berners-Lee wrote in the May 2001 issue of Scientific American, “The Semantic Web
is an extension of the current web in which information is given well-defined meaning, better
enabling computers and people to work in cooperation.”. In the Semantic Web, data/information
must have/convey meaning and for this to happen new technologies have been introduced. As
W3C describes, Semantic Web technologies enable people to create data/information stores on
the Web, build commonly agreed and shared vocabularies, and write formal and explicit rules
for handling linked data. Linked (and open) data are empowered by technologies such as RDF,
SPARQL, OWL (W3C.org, 2015; Cahn, 2017). Such technologies play a crucial role in the
formal representation and structuring of information and knowledge that chatbots must be fed
in order to be able to support accurate machine-processable and machine-understandable
conversations/dialogues (in the form of human-like questions/answers).
2.1.4 Knowledge graphs (KGs)
KGs can be viewed as a new AI technological trend that has its origin to the Semantic Web. As
it is still evolving, several definitions can be found. A selected from recent literature definition

3
is: “A knowledge graph (i) mainly describes real world entities and their interrelations,
organized in a directed graph, (ii) defines possible classes and relations of entities in a schema,
(iii) allows for potentially interrelating arbitrary entities with each other and (iv) covers various
topical domains”. KGs are very flexible and could potentially give to a chatbot unlimited access
to stored and structured knowledge (Riedl & Bulitko,2013; Stichbury, 2017; Bonatti, Decker,
Polleres & Presutti, 2019).
All the above technologies, with all their detailed specifications, extensions and modules,
are important and necessary components for developing an AI chatbot. Programmers have a
choice to create their own custom-made solution for the development of a chatbot, which will
give them full freedom and flexibility but a lot of coding time. On the other hand, today, IT
companies and individual consortiums have created AI Chatbot platforms (as discussed in
following section) that can be used by developers to create their chatbots fast and with low
effort.
2.2. AI Chatbot Platforms
Before using a readymade chatbot, a developer must define the reasons and the goals that the
chatbot is designed for. There are mainly three types of chatbot platforms: No-programming
platforms, conversation-oriented platforms, and platforms backed by the industry (Couto,
2017).
1. The no-programming platforms are simple task-oriented platforms that do not have NLP
or ML capabilities. There are easy in use and in construction of chatbots and can answer
to simple questions concerning simple tasks such as order a pizza or buy a ticket. The
generated chatbots can be easily integrated in social media and webpages. The most
famous platforms are Chatfuel (2019), Octane.ai (2019) and Motion.ai (2019).
2. The conversation-oriented platforms on the other hand are not built to serve a task but
are designed to make a conversation. Most of them use the Artificial Intelligence
Modelling Language (AIML) (2019) in order to model the conversation with a user and
manage to conduct good discussions. The don’t use NLP or ML techniques so the
developers must keep them manually updated. The most famous conversational bots’
platform is PandoraBots (2019).
3. The platforms backed by the industry are advanced chatbot platforms that are using ML
and/or NLP technology, and may integrate large external sources of
data/information/knowledge such as Web pages, ontologies, and Knowledge Graphs.
The most famous of these platforms are Google Dialogflow (2019), Facebook Wit.Ai.
(2019), KITT.AI (2019), ΙΒΜ Watson Assistant (2019), Microsoft Luis (2019) and
Amazon Lex (2019).

2.3. Chatbots and the Museum

Museums have been experimenting with chatbots for more than a decade. The first chatbot
applications were conversational bots that were interacting with the audience through screens
as avatars or through phone or text services (Boiano, Gaia & Caldarini, 2003; Boiano, Borda,
Gaia, Rossi, & Cuomo, 2018).
Gradually with the rise of new advanced AI technologies and the use of social media, new
more advanced chatbot apps were introduced and used in several sectors. Chatbots in museums
are providing a new way of increasing visitors engagement by functioning as museum guides
or as simple question/answer (QA) info agents or in a more sophisticated way as visit planners
or by using, for example, gamification techniques (Schaffer, Gustke, Oldemeier & Reithinger,
2018; Roussou, Perry, Katifori, Vassos, Tzouganatou, & McKinney 2019). A brief review of
recent efforts in creating bots for museums is provided in the following paragraphs.

4
 At the National Art Museum of the Republic of Belarus, a simple QA Facebook messenger
bot has been created. This bot can answer simple questions about the artefacts of the museum,
and it can be called as a simple Infobot. However, this bot lacks quality conversational skills
(Boiano, Borda, Cuomo, Gaia & Rossi, 2018).
At Anne Frank House in Amsterdam, an off-site Facebook messenger bot is developed and
implemented. The content of this bot is well curated as the user is guided to follow a
predetermined route. The bot is in control of the conversation and gives to the user’s different
ways to move through the life of Anne Frank and the museums exhibits. The bot is not designed
to answer free questions and it gets confused when this happens. The designed scenario is
helpful and safe for the museum as it provides protection from bad usage (Tzouganatou, 2018).
The Maxxi’s Chatbot is developed at the National Museum of the 21st Century Arts in Rome.
The Facebook messenger bot uses predefined content and leads the visitor through selections
to plan or follow a certain route path. The bot is lacking conversational skills but provides
engagement through rewards (Tzouganatou, 2018).
At the House Museum of Milan an advanced Facebook messenger bot was created, named
Di Casa in casa adventour bot. The bot uses gamification techniques as sole users or groups of
visitors play a search tour game, trying to find clues and learn new things. The bot is providing
strong visitors’ engagement, though there is not a lot of conversation involved (Boiano, Borda,
Gaia, Rossi, & Cuomo, 2018).
At Catal Hoyuk Neolithic site an advanced FB Messenger bot was developed and
implemented by a group of experts. The bot had the skill not only to inform through a Q/A
dialogue but also to provoke the user about a certain issue. The bot surprises its users, takes
control of the conversation, and awakes emotions and thoughts. It is designed to make
conversations that have a meaning. For example, the bot can start a conversation about the
concept of death in the Neolithic period. The final scope of the bot is to alter the perception of
the users about an issue and engage them (Roussou, Perry, Katifori, Vassos, Tzouganatou, &
McKinney 2019).
Considering all the above examples about bots implemented in museums, we can
distinguish and categorize four different types of bots in the museum’s application domain:
 Simple Q/A infobots
 Predefined conversation routes chatbots
 Gamification and reward bots
 Provoking bots
The main challenge of the newly designed museum bots is for the developers and the museum
stakeholders to decide whether the bots that they implement will stay simple and predefined or
will follow the recent trends, become “smarter” using advanced AI technologies and methods,
and engage audience through meaningful human-like (near human-intelligence) conversations
(Thies, Menon, Magapu, Subramony, & O’Neill, 2017; Schlesinger, O’Hara & Taylor, 2018;
Tzouganatou, 2018).

2.4. Requirements and motivation

A chatbot application in order to be effective and useful for the museum, must combine the
recent AI technological advances with the museum needs and purposes. A museum chatbot
must fulfill a large percentage (if not all) of the following requirements and characteristics
(Sameera, A. Abdul-Kader, & Woods, J., 2015; AbuShawar & Atwell, 2015; Radziwill &
Benton, 2017; Nahdatul, 2018).

5
2.4.1 Non-functional Requirements
A museum chatbot must be simple, informative, accurate and precise. It must have strong
conversational skills and provide meaningful content. The museum bot should be entertaining
and should engage the audience in the whole experience/tour duration. Furthermore, it must be
able to avoid prejudice and misuse by its users and overcome unexpected input. In addition, it
could be positive if there is a capability by the bot to provoke users to find and learn more, but
at the same time to be sensitive and understanding on human emotions.

2.4.2 Functional Requirements

The developers of a museum chatbot must be able program/configure it to have the ability of
NLP/NLG and to achieve continuous learning with ML technologies. Also, it must be
programmed to respond to text messages and to voice input with the use of text or speech
recognition technologies. Furthermore, it must have the infrastructure to be available anytime,
to have an attractive graphical design and to be easy to use, without profound technical
parameters and problems. In addition, considering the possibility to be used during a museum
visit, its location should be pointed out with GPS or other techniques and it should be enriched
with multimedia content, so it could provide to the visitors a complete museum tour. Finally,
the developers should consider all the security threats and take all the necessary measures to
develop a secure app for the museum visitors and the museum content.
The challenge of our work is to realize the usefulness of these requirements and create
chatbots that can serve the complex needs of the visitors and the museums curators.

3. The MuBot Platform

3.1. Aim, Technologies and Architecture

The proposed MuBot platform will provide museums the opportunity to create simple,
interactive and human-friendly chatbots for their visitors. The visitors will be able to use a
chatbot application that will be created through the MuBot platform, to chat with a ‘smart’
exhibit when they are in front of (or close to) it. The users will be able to ask questions through
text or voice (in natural language) and receive audible or written answers.
The more the visitors ask, the more MuBot will “learn” and store new knowledge in their
knowledge base. Visitors will chat with the exhibits in real-time, and on the other hand, exhibits
(through chatbot and the use of AI technology) will be able to provide the right answers to their
visitors, in different forms (text, voice, image, video, etc.).
The MuBot Platform will use advanced ML, NLP and Semantic Web technology. Also, will
be able to connect with external resources like domain-specific Webpages, AIML libraries,
ontologies, knowledge bases and knowledge graphs. In Figure 1, we present the early
architectural design of the proposed MuBot platform:
This preliminary architecture that is proposed relies on three aspects (Hallili, 2014): i) A
Knowledge Base that relies on a Semantic Web technology (RDF, SPARQL query language,
and OWL ontologies), for data extraction, representation, linking, reasoning and querying, ii)
NLP for interpreting users input, and iii) NLG for creating the proper well-defined answers.
The proposed platform will be evaluated in the domain of museums. The MuBot domain-
specific instantiation of the proposed platform will be able to conduct dialogues/conversations
such as the one designed for our experimental chatbot, the Cretan MuBot:
- User: Where was Snake Goddess figurine found?
- MuBot: The Snake Goddess was found at Knossos Palace.
- User: And where was Knossos Palace located?
- MuBot: Knossos Palace was in Crete.

6
 Fig.1. Museum Bot (MuBot) platform early design architecture.

3.2. Use Case Scenario: Cretan chatbot in Dialogflow

As MuBot platform is currently in the design phase, a demo chatbot was developed for the
Archaeological Museum of Heraklion with the use of Google’s Dialogflow Platform
(Orth,2017). The demo chatbot is called Cretan (i.e., from Crete) MuBot and allows users to
chat with a famous exhibit of the museum, the “Snake Goddess” figurine. The Cretan MuBot
is online for demonstration at http://tiny.cc/d56a6y.

Fig. 2. The Snake Goddess figurine in Cretan MuBot experimental application environment.

7
For the creation of the Cretan MuBot, specific guidelines suggested by the Dialogflow platform
(2019) were followed:
1. A small scale archaeological scientific research was carried out with the scope of
defining the purpose of the chatbot, conducting a detailed documentation of the exhibit
and researching of external sources and further content.
2. Considering the internal architecture and functions of the Dialogflow platform (2019),
certain categories, types of content, relations, possible questions and answers for the
Cretan MuBot experiment exhibit were engineered. For the Snake Goddess figurine, the
categories of chronology, material and purpose of the figurine were defined. Also, some
questions and answers were designed for each content category. For example, some of
the questions that were designed for the type of content Chronology were: “When is the
figurine dated?”, “What is the chronology of the figurine?”. Possible answers were also
defined and engineered.
3. The next step was the use of Dialogflow platform for the creation of the infrastructure
of the bot. Intents, entities and relationships were created and through QA inputs the
Cretan MuBot was trained. The Entities that were defined were: chronology, material,
location and purpose. Some of the intents were: What is the material of the figurine?
Why does she hold snakes? Where was the figurine found? Dialogflow platform’s ML
and NLP technology assisted in making the Cretan MuBot “smarter”.
4. Afterwards, the API was created and the connection of the Cretan MuBot with the
experimental Web page was established. At that point the developers considered the
graphic design parameters and the usefulness of the bot, so it could be attractive and
easy in use.
5. Finally, the Cretan MuBot got into a testing period by expert users which were making
certain questions to the bot. In specific, 3 experienced Web and mobile app
programmers and 5 cultural and museum informatics specialists tested with their
questions the Cretan MuBot.
The first results were that the bot provided the correct answers only when the predefined or a
highly similar question was made. Each time the bot could not answer, a fixed predefined
answer was provided. After each test the developers were evaluating the results and tried to
train the Cretan MuBot by creating new intents, entities and relations. The Dialogflow had the
technical ability to use ML in order to identify similar questions and train the bot, and also could
store all the questions that were made, so developers could train the bot by providing the right
answers, give alternatives to the intents and define new entities. The Cretan MuBot
conversational skills were not fascinating and there is a lot of work to be done in order to be
classified as “smart”.
As for the abovementioned methodology it must be pointed out that for the creation of a
chatbot every Chatbot platform provides basic guidelines that follow the basic principles of
Web apps development methodologies: requirements analysis, design, development and testing.
This methodology has different extensions and variables according to the domain that is
implemented. The specification of a more detailed methodology for the development (as well
as for the evaluation) of museum chatbots must (and will) be researched and defined in our
future work.

8
3.3. Use Case Scenario: Cretan chatbot in MuBot proposed KG-based architecture
In addition to the ML-based implementation of our Cretan MuBot (using Dialogflow), we have
experimenting with the proposed knowledge graph-based preliminary architectural design, as
depicted in Figure 3 and described in the following subsections. In the following paragraphs,
we demonstrate an example knowledge base and related queries that have been engineered for
this purpose.

Fig. 3. The Cretan MuBot KG-oriented instantiation.

3.3.1. Knowledge Base

The knowledge base uses an example MuBot ontology engineered for the specific scenario,
with IRI: http://i-lab.aegean.gr/ontologies/mubotOnto/ and prefix “mbo”. Also, uses ontologies
from external sources such as the DBpedia ontology, with IRI: http://dbpedia.org/ontology/ and
prefix “dbo”.
The data of the MuBot experimental (RDF) model is represented in the triples form: <subject>,
<predicate>, <object>. In our example, the statement “Snake Goddess figurine was discovered
at Knossos Palace” is expressed in RDF with the following triples:

a) <mbo:SnakeGoddess> <rdf:type> <mbo:Figurine>

b) <mbo:SnakeGoddess> <mbo:discoveredAt> <mbo:KnossosPalace>

By extending and enriching the model with external semantic data derived by DBpedia and
the MuBot ontology, the following triples can be added:

9
 c) <mbo:Figurine> <mbo:discoveredAt> <dbo:Place>
d) <mbo:KnossosPalace> <dbo:location> <dbo:Crete>
e) <dbo:Crete> <rdf:type> <dbo:Place>
f) <mbo:KnossosPalace> <rdf:type> <dbo:Place>

The classes that are used for the example instantiation are: mbo:Figurine and dbo:Place, and
the properties are: mbo:discoveredAt, rdf:type, and dbo:location. Both classes and properties
may be enriched with synonyms so the system would be able to understand different questions
with the same meaning. The connection/linkage of the MuBot platform with external KGs such
as DBpedia (2019), as well as the reasoning engine (inferencing mechanism) will provide/add
a range of RDF triples that will make the Cretan MuBot more ‘intelligent’ and effective in
understanding users’ questions (as presented in the following lines).

3.3.2. Question understanding with NLP

MuBot platform supports textual or vocal input of user-made questions. This input can be
interpreted using several NLP techniques. Related work (Hallili,2014; Cabrio, Cojan, Aprosio,
Magnini, Lavelli & Gandon, 2012) describes three aspects of identification: i) Expected Answer
Type (EAT) which in our architecture is named as Content Type Recognizer (CTR) and is what
the input is looking for, ii) the property that relates the question with the possible answer, and
iii) the NE (Named Entity) which in our architecture is named Content Entity Recognizer (CER)
and represents the subject of the question made. In the presented example the question made is
“Where Snake Goddess figurine was discovered?”. The CTR is mbo:location, the property is
mbo:discoveredAt and the CER is mbo:SnakeGoddess and mbo:Figurine. Up to this point, NLP
techniques were used in the example in order to identify the entities, its properties and the
expected answers. These NLP techniques are using pattern matching, entities relevance, scoring
strategy and others.
When all the above elements are processed the system is generating a query that produces
results from the KB. The example query (in SPARQL query language) looks as follows:

PREFIX mbo: <http:/i-lab.aegean.gr/ontologies/mubotOnto/>

PREFIX dbo: <http://dbpedia.org/ontology/>
SELECT? v? z
WHERE {
mbo: SnakeGoddess mbo:discoveredAt ?v.
?v dbo:location ?z.
}

After query execution and pattern matching, the following triples are matched and returned:
a) <mbo:SnakeGoddess> <mbo:discoveredAt> <mbo:KnossosPalace>
b) <mbo:KnossosPalace> <dbo:location> <dbo:Crete>
In a more elaborated example engineered to demonstrate the inferencing/reasoning
capabilities of the proposed approach, the following statement is defined (in terms of
properties and property hierarchy):
c) <mbo:discoveredAt> <rdfs:subPropertyOf> <mbo:foundAt>
The restriction of mbo:foundAt object property has as domain the class mbo:Figure, and as
range the class dbo:Place. Inferencing will add the following inferred triple in the model:
d) <mbo:SnakeGoddess> <mbo:foundAt> <mbo:KnossosPalace>

10
This allows for querying the inferred model (tested in Snap SPARQL plugin of Protégé 5.5)
with the following queries also:

I. “Where Snake Goddess figurine was found?”:

PREFIX mbo: <http:/i-lab.aegean.gr/ontologies/mubotOnto/>
PREFIX dbo: <http://dbpedia.org/ontology/>
SELECT ?v ?z
WHERE {
mbo:SnakeGoddess mbo:foundAt ?v.
?v dbo:location ?z.
}

The returned data in variables are: ?v = mbo: KnossosPalace, z? = mbo:Crete.

II. “What was found in Crete?”:

PREFIX mbo: <http:/i-lab.aegean.gr/ontologies/mubotOnto/>
PREFIX dbo: <http://dbpedia.org/ontology/>
SELECT ?s ?v
WHERE {
?s mbo:foundAt ?v.
?v dbo:location mbo:Crete.
}

The returned data in variables are: ?s = mbo:SnakeGoddess, v? = mbo: KnossosPalace,

allowing for NLG module (described in next section) to provide answers such as “Snake
Goddess was found in Knossos Palace, in Crete”.
The example Cretan Mubot model (in .owl and .ttl serializations) and the related queries
in SPARQL can be accessed from https://github.com/KotisK/muBotOnto-example.

3.3.3. NLG responses

The last step of the question/answer processing of the MuBot is the creation/formation of the
appropriate response/answer to the given question(s). The system uses NLG techniques in order
to pick a response pattern that matches with the query-provided triples and procced to the final
response formatting.
In the presented example the NLG module will be able to generate the following answers:
a) “Snake Goddess was found at Knossos Palace”
b) “Knossos Palace is located in Crete”
Or better, by synthesizing the two statements above:
c) “Snake Goddess was found at Knossos Palace, in Crete” (replacing ‘is located in’ with
‘, in’).
The generation processes could be more creative and exploit other semantic properties that, for
example, could return as a response multimedia files like photographs of the figurine.

4. Future work
The Cretan MuBot is the first step for the creation of the MuBot chatbot platform for museums.
The Cretan MuBot will be fully developed, enriched with more exhibits, tested with different
AI technologies and types of display (webpage bot, FB messenger bot, real time museum app

11
bot) in the museum and evaluated with a certain evaluation strategy. Radziwill and Benton
(2017) describes a chatbot quality evaluation strategy that tries through a literature review to
derive the best quality attributes that a chatbot must have.
In the following design and development stages of our ongoing work, all the available AI
technologies that can be integrated in a chatbot platform will be researched and recorded,
especially those related to ontologies, Knowledge Graphs, and Linked Open Data. Also, one
critical aspect for us to put in the context of a chatbot development is the identification of the
special requirements of a museum for the integration of intelligent bots that can deliver
meaningful conversations and change the way users think of the exhibits/exhibitions and their
hidden ideas. Finally, the definition of a specific development methodology as well as an
evaluation strategy for museum chatbots are left for future work.

5. Conclusions

The integration of the AI chatbot technology in the museum domain is an intriguing ongoing
project for the museum’s stakeholders and the AI developers. The design and implementation
of such bots must consider several variables. The museums are not a commercial organization
and has different scopes to fulfill. A museum bot must provide meaningful content and to adjust
to the visitor’s choices and habits. The bot must be simple and use the AI technologies in a way
that visitors will not be abstracted from the content. The museum bot must have strong AI
conversational skills, but developers always must consider the limitations of the content and to
care for the privacy and security issues. A full understanding of all the aspects of the
implementation of an AI chatbot for museums is the main purpose of the MuBot Platform
ongoing project. This paper has presented the first steps and preliminary research outcomes of
such a project.

References

AbuShawar, B., Atwell, E. (2015). ALICE Chatbot: Trials and Outputs, Computación y
Sistemas, Vol. 19, No. 4, 2015, pp. 625–632

Amazon Lex. (2019). Conversational interfaces for your applications,

https://aws.amazon.com/lex/, Last accessed: 20/10/2019

Artificial Intelligence Markup Language. (2019). The open standard scripting language for
chatbots, http://www.aiml.foundation/, Last accessed: 20/10/2019

Bilange, E. (1991). A task independent oral dialogue model. In Proceedings of the Fifth
Conference of the European Chapter of the Association for Computational Linguistics,
pages 83–88, Berlin, Germany. Association for Computational Linguistics.

Boiano, S., Borda, A., Gaia, G., Rossi, S., & Cuomo, P. (2018). Chatbots and New Audience
Opportunities for Museums and Heritage Organisations. (September).
https://doi.org/10.14236/ewic/eva2018.33

Boiano, S., Gaia, G., & Caldarini, M. (2003). Make Your Museum Talk: Natural Language
Interfaces for Cultural Institutions. Museums and the Web 2003, Retrieved from
https://www.museumsandtheweb.com/mw2003/papers/gaia/gaia.html

12
Boiano, St., Borda, A., Cuomo, P., Gaia, G., Rossi, S. (2018). Chatbots in Museums: Hype or
Opportunity?, MuseWeb, in https://www.museweb.net/chatbots-in-museums-hype-or-
opportunity-%E2%80%A8/

Bonatti, P. A., Decker, S., Polleres, A., & Presutti, V. (2019). Knowledge graphs: new
directions for knowledge representation on the semantic web. Report from Dagstuhl
Seminar, 8(9), 29–111. https://doi.org/10.4230/DagRep.8.9.29

Cabrio, E., Cojan J., Aprosio A. P., Magnini B., Lavelli A., and Gandon F. (2012). Qakis: an
open domain QA system based on relational patterns. In International Semantic Web
Conference (Posters & Demos)

Cahn, B. J. (2017). [Thesis] Chatbot Literature Review. Retrieved from

https://static1.squarespace.com/static/569293741c1210fdda37b429/t/59160b6bff7c50104
e601a85/1494616940469/CHATBOT_thesis_final.pdf

Chatfuel, chatbot platform for Facebook Messenger. (2019) Retrieved from

https://chatfuel.com

Couto, J. (2017). Building a Chatbot: Analysis and Limitations of Modern Platforms - DZone
AI. Retrieved from https://dzone.com/articles/building-a-chatbot-analysis-amp-
limitations-of-mod

DBpedia. (2019). Structured Wikipedia data, Retrieved from https://wiki.dbpedia.org/

Dialogflow. (2019). Build natural and rich conversational experiences, Retrieved from
https://dialogflow.com

Følstad, A., & Brandtzaeg, P. B. (2017). Chatbots and the New World of HCI. Interactions,
24(4), 38–42. https://doi.org/10.1145/3085558

Hallili, A. (2014). Toward an Ontology-Based Chatbot Endowed with Natural Language

Processing and Generation. 26th European Summer School in Logic, Language &
Information, Aug 2014, Tübingen, Germany. hal-01089102.

IBM Watson Assistant. (2019). More than a chatbot, Retrieved from

https://www.ibm.com/cloud/watson-assistant/

Kitt.ai. (2019). Conversational Understanding as a Service, Retrieved from http://kitt.ai/

Language Understanding (LUIS). (2019). A machine learning-based service to build natural

language into apps, bots, and IoT devices. Retrieved from https://www.luis.ai

Medhi Thies, I., Menon, N., Magapu, S., Subramony, M., & O’Neill, J. (2017). How do you
want your chatbot? An exploratory Wizard-of-Oz study with young, Urban Indians.
Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial
Intelligence and Lecture Notes in Bioinformatics), 10513 LNCS, 441–459.
https://doi.org/10.1007/978-3-319-67744-6_28

Motion.Ai. (2019). A visual chatbot builder, Retrieved from https://www.motion.ai/

13
Nahdatul, Α. (2018). Review of Chatbots Design Techniques, International Journal of
Computer Applications (0975 – 8887) Volume 181 – No. 8,
https://www.researchgate.net/publication/327097910_Review_of_Chatbots_Design_Tech
niques

Octane.ai. (2019). Commercial and social media bots, Retrieved from https://octaneai.com/

Orth, A. (2017). Building chatbots with Dialogflow and Grank.AI in Grank.ai, Retrieved from
https://blog.grakn.ai/chatbots-and-grakn-ai-67563c64cfde

PandoraBots. (2019). Intelligent conversational agents, Retrieved from

https://home.pandorabots.com/home.html

Radziwill, N. M., & Benton, M. C. (2017). Evaluating Quality of Chatbots and Intelligent
Conversational Agents. Retrieved from http://arxiv.org/abs/1704.04579

Resource Description Framework (RDF). (2019). Concepts and Abstract Syntax, Retrieved
from https://www.w3.org/TR/rdf-concepts/

Riedl, M.O., Bulitko, V. (2013). Interactive narrative: An intelligent systems approach. AI Mag.
34 (1), 67–77, https://www.aaai.org/ojs/index.php/aimagazine/article/view/2449

Rivero, A. (2018). Next generation of chatbots with NLP services and Graphs in Chatbots
Life, Retrieved from https://chatbotslife.com/next-generation-of-chatbots-with-nlp-
services-and-graphs-cd811a8165d7

Roussou, M., Perry, S., Katifori, A., Vassos, S., Tzouganatou, A., & McKinney, S. (2019).
Transformation through Provocation? (627), 1–13.
https://doi.org/10.1145/3290605.3300857

Sameera, A. Abdul-Kader, & Woods, J. (2015). Survey on Chatbot Design Techniques in

Speech Conversation Systems. International Journal of Advanced Computer Science and
Applications, 6(7), 72–80. https://doi.org/10.14569/ijacsa.2015.060712

Schaffer, S., Gustke, O., Oldemeier, J., & Reithinger, N. (2018). Towards chatbots in the
museum. CEUR Workshop Proceedings, 2176, 1–7.

Schlesinger, A., O’Hara, K. P., & Taylor, A. S. (2018). Let’s talk about race: Identity, chatbots,
and AI. Conference on Human Factors in Computing Systems - Proceedings, 2018-April.
https://doi.org/10.1145/3173574.3173889

Stichbury, J. (2017). WTF is a Knowledge graph? in Hackernoon.com:

https://hackernoon.com/wtf-is-a-knowledge-graph-a16603a1a25f

Tzouganatou, A. (2018). Can Heritage Bots Thrive? Toward Future Engagement in Cultural
Heritage. Advances in Archaeological Practice, 6(4), 377–383.
https://doi.org/10.1017/aap.2018.32

14
 Valtolina, S., Barricelli, B. R., Gaetano, S. Di, & Diliberto, P. (2018). Chatbots and
conversational interfaces: Three domains of use. CEUR Workshop Proceedings, 2101, 62–
70.

Vassos, S., Malliaraki, E., Federica dal Falco, Jessica Di Maggio, Massimetti, M., Nocentini,
M., Testa, A. (2016). Art-Bots: Toward Chat-Based Conversational Experiences in
Museums, in https://link.springer.com/chapter/10.1007/978-3-319-48279-8_43

W3C.org. (2015). Semantic Web, Retrieved from https://www.w3.org/standards/semanticweb/

Wehle Hans-D. (2017). Machine Learning, Deep Learning and AI: What’s the Difference?,
Retrieved from
https://www.researchgate.net/publication/318900216_Machine_Learning_Deep_Learning
_and_AI_What's_the_Difference

Wit.ai. (2019). Natural Language for Developers, Retrieved from https://wit.ai/

15

Powered by TCPDF (www.tcpdf.org)