Retail Platform with Natural Interaction: LOAD's Vision
Pedro Colarejo
1
, Davide Ricardo
1
, André Fernandes
1
, Miguel Fonseca
1
, Pedro Oliveira
1
,
Nidhal Cherni
1
, João Abrantes
1
, Hélio Guilherme
1
and António Teixeira
2
1
LOAD Interactive, Lda., Aveiro, Portugal
2
IEETA, Dep. Telecommunications and Information Technology Electronics, University of Aveiro, Aveiro, Portugal
{pedro.colarejo, davide.ricardo, andre.fernandes, miguel.fonseca, pedro.oliveira, nidhal.cherni, joao.abrantes,
Keywords: Retail, Information for Consumers, Virtual Assistants, Dialogue, Natural Language, Image Recognition,
Blockchain.
Abstract: In a context of indirect sales channels, where the product reaches the final consumer through intermediaries, it
is very difficult for consumers to know the origin of the product and identify the origin of problems. In this
paper an innovative technological platform is proposed, oriented to the retail market, and capable of providing
information from the entire product distribution chain to its various stakeholders. The platform is based on 3
pillars: a decentralized blockchain-based information network covering the path of a product from its origin to
the end consumer; extraction of information about users/customers and products from images and video;
interaction using natural language. A first instantiation of the platform is also presented as well as the first
results. In its development, recent technologies were used in the areas of image recognition and dialogue
systems.
1 INTRODUCTION
From the point of view of the global process of selling
products/services by an organization to its customers,
the customer service has always had a great
importance in the ability of these organizations to
generate revenue. This assistance to customers can
consist of a set of services that help customers to
know more information about a product/service they
intend to purchase and help their choice. In this sense,
these organizations have a higher cost to improve
these services, investing in the training of employees,
to understand by interacting with the clients their
opinions and needs regarding the services provided.
A bad provision of these services increases to
probability of losing customers to the competition.
Examples of poor practices in the provision of this
type of services are well known, including excessive
duration in problem solving, rude treatment, lack of
transparency in resolution processes, not being able
to give a concrete answer on the problem to be solved,
promises that have not been fulfilled, etc.
In a context of indirect sales channels, where there
is a value chain, and where the product reaches the
final consumer from intermediaries, the customer
service, during the sales process, becomes more
complicated. The manufacturer who produces the
product or provides the service does not reach the
final consumer directly. There is a set of distributors
and retailers that complete this value chain, where
distributors buy the product and sell it to retailers,
who will make the product available and sell to final
consumers. There are advantages to this type of
distribution channels, but there are also some
limitations. On the one hand, the manufacturer loses
control over the message that is transmitted to the end
customer, as this is done by retailers and resellers, nor
can it control the importance of the product facing its
competitors in retail stores. Furthermore, in situations
where there is a problem with the product (delivery
delay, defects, etc.), it is very difficult for end
consumers to understand the origin of the product and
whether the problem is in the manufacturer, or
elsewhere in the chain.
In recent years, organizations have invested in
automated methods to provide these support services.
Virtual Customer Assistants (VCA), which include
chatbots, are the latest of these methods. VCAs are
applications that are typically available online
capable of establishing a conversation with customers
via voice and/or text. VCA's use has grown rapidly,
as there is great potential in improving the provision
Colarejo, P., Ricardo, D., Fernandes, A., Fonseca, M., Oliveira, P., Cherni, N., Abrantes, J., Guilherme, H. and Teixeira, A.
Retail Platform with Natural Interaction: LOAD’s Vision.
DOI: 10.5220/0010996000003179
In Proceedings of the 24th International Conference on Enterprise Information Systems (ICEIS 2022) - Volume 2, pages 293-300
ISBN: 978-989-758-569-2; ISSN: 2184-4992
Copyright
c
2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
293
of customer service, both on the organization side and
on the client side. Part of this is justified by the
growing tendency among customers to use text and
instant messaging as their first choices when
communicating with an organization, rather than
using the traditional voice channel (the phone). In
addition, there is a growing tendency for customers to
want to solve their own problems in a direct way,
rather than asking someone else to do so for them.
From the point of view of organizations, the
implementation of VCA, in addition to its greater
efficiency, can result in a considerable reduction in
costs, since the traditional voice channels, using the
phone, are quite expensive, being much cheaper to
keep this service fully available 24 hours a day, than
the same availability with human resources.
However, there are several technological
challenges to implement such assistants, capable of
interacting as if they were human. The interaction
platform has to be able to improve the way it interacts
with customers, not only from the point of view of
usability, but also from the user experience, that is,
the perception with which customers are when
interacting with the system, or the level of
satisfaction/frustration that the customer experiences.
This implementation tends to be sustained by
technological advances in artificial intelligence,
namely Machine Learning (ML), with applications in
Natural Language Processing (NLP) and Image
Recognition.
Thus, the main objective defined for this work,
was the development of an innovative technological
platform, oriented to the retail market, applicable in
different business areas, to provide information of the
entire product distribution chain to its various
stakeholders. It must be based on a decentralized
information network, covering the complete path of a
product from its origin until it reaches the final
consumer.
This article is structured as follows: in section
2.
the work related to the areas most directly related to
the implementation of the platform (Blockchain,
image/video recognition and natural language
dialogue system) is presented. In the third section is
defined the main scenario to be considered, in order
to realize the implementation of the prototype. In
section 4 the general architecture of the system and its
main components is described; in section 5 how the
first prototype was implemented; and, in section 6,
the first results of the integration of the various
modules. The main conclusions from the work
performed and some future paths are presented in
section 7.
2 RELATED WORK
The development of a solution such as the one
proposed in this paper requires the integration of
results from various areas. In the following
subsections is very briefly presented the work related
to the implementation of a blockchain for the retail
sector, object recognition technologies and
technologies associated with conversational
assistants (chatbots). No platforms or Assistants
directly related to what is proposed in this paper were
found.
2.2 Blockchain (for Retail)
Blockchain technology (Yaga, Mell, Roby, &
Scarfone, 2019) has been attracting increasing
attention in a wide range of industries due to its ability
to reliably manage transaction-based applications
without the need for a centralized authority. This was
not possible before blockchain emerged. This
technology ensures trust in transactions in a network
between untrusted nodes, since all of these nodes can
trade, even if they do not trust each other. Blockchain
was initially developed to support secure digital
currency transaction (Bitcoins) in order to prevent
transactions from being improperly duplicated.
However, its potential extends to several domains
using peer-to-peer (P2P) architectures, where
interconnected nodes make transactions with each
other without using a centralized management
system, such as electronic voting, purchase and sale
of intellectual property, distribution of confidential
medical information, or intelligent value chains
applying Internet of Things (IoT), called Smart
Logistics (Uckelmann, 2008; Kawa, 2012).
Great strides have been made in the use of
Blockchain for application areas not related to crypto-
currencies, such as Slock.it, a company that works in
smart electronic lock (Slocks), which can be unlocked
with smartphones that have the appropriate token, and
that can be purchased on Ethereum using Ethers. A
Slock owner can rent his house or car at a price for
timed access, as the door lock is automatic. Anyone
interested in renting can identify the Slock, pay the
requested amount and unlock the door. Another
example are transactive energy companies with
blockchain-based solutions, allowing you to buy and
sell energy (generated by solar panels) automatically,
according to the criteria defined by the user.
In relation to value chains, there is a huge
potential for application of Blockchain technology.
With a network based on Blockchain technology,
product tracking and control is complete, since the
ICEIS 2022 - 24th International Conference on Enterprise Information Systems
294
same information in the database is shared by all
entities, where all updates are cryptographically
verified and propagated over the network
automatically, thus creating a record of the
information history that can be validated by all
entities in the network. There are several initiatives in
the retail areas in pilot phases, some more advanced
than others, as, for example, jewelry (e.g., De Beers
1
and TrustChain Jewelry
2
) and food industry (e.g.,
IBM Food Trust
3
and Norwegian Sea Food
4
).
2.2 Object Recognition
A state-of-the-art survey of image recognition was
carried out, and some technologies available on the
market with the potential to integrate a first
implementation of the platform were identified and
characterized, such as:
● Watson - Visual Recognition
5
- Developed by
IBM, the Watson system offers - among many
other things - the Visual RecognitionAPI, which
allows you to: sort content according to scenarios;
recognize objects, faces, colors, food, text, and
inappropriate content; train models for specific
cases.
● Amazon Rekognition
6
- The AWS cloud
computing services platform has in its spectrum
this Image Recognition service that allows you to
identify objects, people, text, scenarios, activities,
and explicit/inappropriate content in both image
and video.
● Google Mobile Vision
7
- Part of the Google ML
Kit, in addition to identifying faces in both
photographs and video, allows to find and observe
facial landmarks (such as eyes, mouth, nose, etc.)
providing information about whether, for
example, the subject is laughing or has his eyes
open. It also allows you to recognize text and
present it according to its structure, thus
maintaining paragraphs, lines, punctuation, etc...
● Google Cloud Vision API
8
- Is part of the Google
Cloud AIplatform, providing detection of objects,
faces, emotions, scenarios, tags, explicit or
inappropriate content, text and colors.
1
https://www.jckonline.com/editorial-article/de-beers-
blockchain-platform/
2
https://www.jckonline.com/editorial-article/richline-
blockchain-trace-gold/
3
https://www.ledgerinsights.com/blockchain-food-
traceability-gs1-foodlogiq-ibm-food-trust-ripe-io-sap/
4
https://cointelegraph.com/news/from-sea-to-table-
norways-seafood-industry-hooks-into-ibm-blockchain
5
https://www.ibm.com/watson/services/visual-
recognition/
● Vize.ai
9
- This company provides the cloud
solution (Vize Custom Image Recognition API)
and an On-device solution (Vize MobileModels)
to recognize and classify images. Unlike most,
this solution requires initial training according to
categories defined by the user in order to cover
specific problems.
● Microsoft Azure Cognitive Services Vision
10
- It
is one of the 5 sets of services that Microsoft
provides through its Cognitive Services platform.
● Cloudsight AI
11
- Provides an API that allows you
to filter and categorize images as well as monitor
explicit/inappropriate content.
● Clarifai
12
- Clarifai provides a cloud solution that
allows you to tag images and videos, create your
own template that is applied to a specific use case,
search content through visual or tag similarities,
and moderate content by identifying whether the
content is explicit/inappropriate or not.
● Imagga
13
- It's a PaaS that provides image
recognition APIs that allow companies to build
and monetize image-intensive applications in the
cloud.
● Kairos
14
- It’s one of the leading companies in the
field of Facial Recognition, providing an API and
SDK that allow you to detect multiple faces,
identify faces (answer the question, "who is
this?"), check faces (answers the question, "this is
..."), detect emotions, detect age groups, detect
facial landmarks, detect levels of attention and
finally detect ethnicities.
A comparison of these solutions is presented in
Table 1. The most promising solutions considering
the objectives were those of IBM and Microsoft.
However, we wanted to ensure full flexibility to
change and adapt to our idea, which was not
guaranteed by these platforms. It was then considered
beneficial to develop solutions tailored to the project,
to gain more knowledge in the area and have more
flexibility to adapt the functionalities. The libraries
Tensorflow, OpenCV and PyZbar were adopted.
6
https://aws.amazon.com/rekognition/
7
https://developers.google.com/vision
8
https://cloud.google.com/vision/
9
https://vize.ai/
10
https://azure.microsoft.com/en-us/services/cognitive-
services/directory/vision/
11
https://cloudsight.ai/
12
https://www.clarifai.com/
13
https://imagga.com/
14
https://www.kairos.com/
Retail Platform with Natural Interaction: LOAD’s Vision
295
Table 1: Comparison of evaluated solutions for image recognition.
2.3 Conversational Assistants
Dialogue systems, Chabots and Conversational
Assistants are increasingly popular due to their use-
fulness: they provide easier and more versatile access
to large and diverse sets of information. Conversation
assistants perform chatbot-like interactions and allow
speech as input and outputs. Google Assistant and
Amazon Alexa are popular examples.
To allow interaction in a dialog format, a typical
conversation system integrates several modules: (1)
Automatic Speech Recognition (ASR) that converts
speech into text; (2) Natural Language Understanding
(NLU) processes ASR word sequences to identify
important information, such as intentions and entities;
(3) The Dialog Manager (DM) manages dialogue and
context information, taking into account the intention,
entities and previous conversations; (4) Natural
Language Generation (NLG) generates phrases; (5)
Speech synthesis uses text-to-speech (TTS) to
produce synthetic speech.
In recent years, there has been impressive progress
in NLP and NLU technologies, particularly accelerated
by the increased popularity of technologies such as the
Internet of Things (IoT) and artificial intelligence. As
a result, several platforms have emerged providing
sophisticated NLU capabilities, one of the main
components of a dialog system, and tools aimed at
developing dialog/chatbot systems, of which are
representative examples: TrindiKit, OpenDial, IBM's
Watson Conversation Service, RASA, (Bocklisch,
Faulkner, Pawlowski, & Nichol, 2017)PyDial, Dialog
Flow, Alexa Skills, ICECAPS, Emora and Plato. One
example to highlight is Plato (Papangelis, et al., 2020),
a tool available by Uber in 2020 that can be used to
create conversational agents, supporting interactions
through speech, text, or dialogue. An application in
Plato consists of four main components: the dialogue
that defines and implements acts of dialogue and states;
the domain that includes the ontology of the dialogue
and the database that the dialogue system consults; the
controller who orchestrates the conversations; and the
agent that implements different components of each
conversation agent.
3 SCENARIO
The following scenario was defined, depending on
which the prototype was designed, developed:
A consumer in a supermarket approaches a shelf
with a particular product. The chatbot/Assistant
detects the presence and starts a conversation,
greeting him/her and asking what it can do to help
(using the right gender context).
ICEIS 2022 - 24th International Conference on Enterprise Information Systems
296
Figure 1: General Architecture of the proposed System/Platform, with its 3 main modules.
The consumer requests information about the
product, such as storage conditions, temperature,
humidity, and also some location.
The Conversational Assistant interprets the
questions and responds, querying the blockchain
to obtain the information needed.
The system complements the speech/text output
with a map showing the product path and all
geographic points that the product has passed
through.
4 PROPOSED SYSTEM
Figure 1 schematically presents the architecture
proposed for the system, consisting of 3 main
modules:
• Interaction with the user.
• Image, Speech and Language Processing.
• Blockchain Infrastructure.
The User interaction module is the frontend
responsible for the user's information, from questions
by voice, to his/her image, or even possibly text, but
also for informing the user, through
questions/answers by voice, image (e.g., map) and
written information displayed on the screen.
The Image, Speech, and Language Processing
Module is a system backend module, which includes
the following components:
Image recognition, responsible for image
processing for facial and product codes
recognition.
NLP/NLU and Dialogue Management, supporting
interaction by natural language through the
identification of intentions and entities in user
interactions (NLU), natural language generation
and dialogue management.
The third main component of the system is
dedicated to the entire Blockchain Infrastructure,
implemented in EOS.IO. This is where all the
information added to the product (transactions, state
changes, value) is stored throughout its progression in
the distribution chain.
5 PROTOTYPE DEVELOPMENT
5.1 Speech and Language Processing
The simplest way to communicate with the system is
through natural language, whether the text is from the
user or Speech Recognition.
To recognize speech and translate it into text a
speech recognition service is used (from Google). To
transform system’s response to speech a module was
developed in C# using Microsoft Speech SDK.
To implement the determination of users'
intentions, extraction of entities (e.g., topic, day),
language generation and interaction management, the
Plato Research Dialogue System, previously
mentioned in this article, was used.
USER INTERACTION
PROCESSING
BLOCKCHAIN
Retail Platform with Natural Interaction: LOAD’s Vision
297
5.2 Image Processing
The integration of image processing with the user
interface, by integrating the web interface through
Websockets services, is a challenge, because it is
where the harmony between the machine and the
human needs to occur. For this prototype, image
recognition focuses on three actions: face recognition,
user gender recognition, and barcode scanner. Face
recognition informs the assistant regarding the start of
the conversation; gender recognition is needed to
shape the conversation; the barcode scanner allows
product identification, so that the system can obtain
information from the blockchain and answer the
questions posed by the customer.
Specific solutions have been created in Python
using libraries such as Tensorflow, OpenCV and
PyZbar. Tensorflow was used to detect client's gender
and emotions; OpenCV was used to detect a person's
face; and PyZbar was used to read barcodes (Abadi,
et al., 2016).
5.2 Blockchain and Communications
The code base of the EOS blockchain has been
identified as being the most suitable for the targeted
prototype. We worked on the integration of external
libraries (AI) in smart contracts. Focusing on
applying proof of concept to a real solution to real
problems (blockchain for intelligent logistics), the
development of a DAPP (Decentralized App) was
started, to store different states of a supply chain,
from the creation of raw material to assembly, and
until reaching the final consumer.
Support has been developed for a set of
interactions with blockchain, such as:
Deployment of the initial contracts for basic use of
the blockchain.
Creation of the default wallet.
Creation of accounts to deploy contracts.
Deployment of smart contracts.
Interaction with the smart contract through
transactions.
Automatic accounts were then created, multi-
indexed tables and the available API (JSON RPC)
were used to communicate with the blockchain and
obtain data from the recorded tables. During this
process, it was concluded that files (such as images)
should be saved off-chain, and then their reference
saved in the blockchain. The IPFS was selected as the
most promising approach.
To integrate the blockchain with the dialog, an
API was developed, which also intervenes in the
interaction with the IoT component (temperature,
humidity, and location sensors). In this way, it is
possible to read and write data to a tag, using the
RFID-RC522 shield, as well as allow the sending of
images captured the assistant to be stored in IPFS (the
distributed system for storing and accessing
applications and data). The API between blockchain
and dialog allows to:
Register users, companies, devices, and products.
Relate each device with different types of sensors
and with a company.
Relate each product to different IDs to be able to
accept different barcodes and RFID.
Send measurement messages from the IoT
component, which will be stored on the blockchain
(and it was possible to save location, time,
temperature, and humidity for a specific product).
Get all the values of each registered entity.
Receive a request for characteristics of a product
(for example: send barcode and receive general
information about a product).
Receive specific questions about a product for the
different sensors (e.g., product x part at
temperature y on day/hour z).
6 INITIAL RESULTS
This section presents information about results for the
first prototype of the platform, starting with the
Blockchain-based module and ending with the
interaction with the system Assistant, visible face of
the system and, in a certain way. integrator of the
whole system.
Once the blockchain was implemented, creating
the initial contracts to record the transactions to be
recorded along the distribution chain, it was
integrated with a set of sensors used to simulate
temperature, humidity, and location data. After this
definition, it was then possible to extract the
information needed (topic, a date, and a location),
making a call to the blockchain API in which the topic
(temperature / humidity), date and location are passed
as parameters. If the blockchain response is
successful, a response is constructed by Plato with the
entities received and the quantity returned. The
following is an example of using blockchain:
-data
'{"table":"product","scope":"loadtestac
c1","code": "loadtestacc1", "json":
"true"}'
{"rows":[{"key":0,"user_id":"0,""nam
e":"lettuce,""description":"organic
lettuce,""from_product":",""sub_product
ICEIS 2022 - 24th International Conference on Enterprise Information Systems
298
s_list":[]},
{"key":1,"user_id":"0,""name":"artichok
e,""description":"artichoke,""from_prod
uct":",""sub_products_list":[]},{"key":
2,""user_id":"0,""name":"carrot,""descr
iption":"organic-growing
carrot,""from_product":",""sub_products
_list":[]},
more":true,""next_key":"10,""next_key_b
ytes":""}
With the following blockchain blocks produced:
info 2021-05-24T12:44:27.000 nodeos
producer_plugin.cpp:1965 produce_block
] Produced block 08cd1f92adcfd669...
#234949 @ 2021-05-24T12:44:27.000 signed
by eosio [trxs: 0, lib: 234948,
confirmed: 0]
info 2021-05-24T12:44:27.500 nodeos
producer_plugin.cpp:1965 produce_block
] Produced block 77107e01d7396056...
#234950 @ 2021-05-24T12:44:27.500 signed
by eosio [trxs: 0, lib: 234949,
confirmed: 0]
info 2021-05-24T12:44:28.002 nodeos
producer_plugin.cpp:1965 produce_block
] Produced block 45e8c73f7ea32c3...
#234951 @ 2021-05-24T12:44:28.00signed
by eosio [trxs: 0, lib: 234950,
confirmed: 0]
As mentioned in the previous section, the
interaction with the user uses image to sense a human
face approaching, its gender (male, female) and
feeling (positive, negative, neutral). The same camera
is also used for reading the product barcode. The
barcode reading process allows to read most barcode
formats, as shown in Figure 2.
Figure 2: Examples of bar code reading by the developed
system.
Figure 3 presents an example of an interaction
with the system, using the implemented interface.
In the example presented in Figure 3, we can
observe the system locating a user who approaches
the system, evaluating whether it should refer to
verbal forms in the male or female, and his/her feeling
at that moment, to better decide how to greet and
answer the customer.
After the desired information collected, a map
with a selection of the product route is displayed
according to the question that was asked.
Figure 3: Graphical part of the user interface of the
prototype: (top left) dialogue that represents in writing
what is said by voice; (right) images of the client with
Information relating to Image Recognition; (bottom left)
graphic information (map with path and other Information);
The reproduction of two other real interaction
examples is presented next, following the internal flow
of data from the user: NLU processing, request to the
blockchain API, generation of the response, and trans-
mission of the system's global response to the user:
Interaction Example 1
User:
What is the humidity in Aveiro on
March 25, 2021?
NLU:
Topic = humidity
Day = 25
Month = March
Year = 2021
City = Aveiro
Blockchain:
API request:
product/info/0?
sensors=humidity&time=1616630400.0
&location=Aveiro
API Response:
{'error': False, 'code': 9000,
'description': 'Product info sent
with success', 'date': {'inTime': 0,
'date': [{'sensor': 'humidity',
'value': 4, 'time': 1610983085}]}}
System Response:
Humidity on March 25, 2021 in Aveiro
was 4%
Retail Platform with Natural Interaction: LOAD’s Vision
299
Interaction Example 2
User:
What is the temperature in Aveiro on
March 25, 2021
NLU:
Topic = temperature
Day = 25
Month = March
Year = 2021
City = Aveiro
Blockchain:
API request:
product/info/0?
sensors=temperature&time=1616630400.
0&location=Aveiro
API Response:
{'error': False, 'code': 9000,
'description': 'Product info sent
with success', 'date': {'inTime': 0,
'date': [{'sensor':
'temperature','value': 3, 'time':
1610983085}]}}
System Response:
The temperature on March 25, 2021 in
Aveiro was 3ºC
In the first interaction example the user asks a
question regarding the storage temperature of the
product that was recognized through its barcode, the
language processing system identifies the topic
(temperature), which is a question as well as date and
location, uses this information to query the
Blockchain and, based on the API response of this
(value=4), generates a sentence to be transmitted as a
response to the user. The second example is similar,
changing the topic detected in the user's sentence,
now "temperature". Although not integrating the
examples presented, due to space limitations, the
system is equipped with the ability to question the
user if he/she does not immediately provide all the
information necessary to have a complete query to the
blockchain. For example, if you the customer doesn’t
inform the local, the system will ask for it.
7 CONCLUSIONS
This article presents the vision for a platform that
facilitates consumer access to blockchain-based
product information by use of natural language
dialogue supported by image, speech, and language
processing technologies. As a first prototype, a
system capable of interaction by speech with a
customer that approaches the system was obtained,
distinguishing gender and initial "mood" and capable
of receiving requests for information by voice,
distinguishing the different entities from the
expression used, consulting the information on the
blockchain, and responding to the user by voice, text,
and image.
In its current state of development, several aspects
require evolution, such as: greater flexibility in the
extraction of users’ intentions and feelings; greater
ability to convey intentions and emotions using
speech (e.g., timbre adjustment); evolution of the
blockchain to profit from the constant evolution of
this technology. The system should also be subjected
to user use and evaluation in environments gradually
closer to the real, allowing its user-centric evolution.
ACKNOWLEDGEMENTS
This work was co-financed by project no. 38546
under the Incentive System R&D Projects-Individual
Projects of Portugal 2020.
REFERENCES
Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean,
J., Isard, M. (2016). Tensorflow: A system for large-
scale machine learning. 12th USENIX symposium on
operating systems design and implementation (OSDI
16).
Bocklisch, T., Faulkner, J., Pawlowski, N., & Nichol, A.
(2017). RASA: Open source language understanding
and dialogue management. arXiv preprint
arXiv:1712.0518.
Kawa, A. (2012). SMART logistics chain . Asian
Conference on Intelligent Information and Database
Systems. Springer.
Papangelis, A., Namazifar, M., Khatri, C., Wang, Y.-C.,
Molino, P., & Tur, G. (2020). Plato dialogue system: A
flexible conversational ai research platform. arXiv
preprint arXiv:2001.06463.
Uckelmann, D. (2008). A definition approach to smart
logistics. International Conference on Next Generation
Wired/Wireless Networking. Springer.
Yaga, D., Mell, P., Roby, N., & Scarfone, K. (2019).
Blockchain technology overview. arXiv preprint
arXiv:1906.11078.
ICEIS 2022 - 24th International Conference on Enterprise Information Systems
300
