Page 99 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 5 – Internet of Everything

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 5

To create the bot, the Google Cloud platform was cho‑ We can see in Fig. 5 the work low of requests in a possible
sen, in which all the back end and the various functions user interface. All these interventions are immediate, the
are hosted, alongside the DialogFlow service provided by most expensive response in terms of timing is the one in
Google that offers a retrieval‑model‑based technique for which the data is requested, in this case the video stream.
matching responses with the aid of machine algorithms But in principle, the time between a send‑reply is given by
learning, where the latter can be enabled at the user’s dis‑ the user interaction time with the device plus the delay
cretion. This function, if enabled, allows us to have some introduced by the bot to reply, which is a maximum of a
lexibility in the interpretation of the user’s requests as few seconds.
the answers are given based on the best score obtained
from a classi ication prior to the choice of the answer. In
this way, therefore, it is possible to manage any spelling
or form problems that might cause errors in recognizing
the correct intent for the request made.
The proposed solution allows for a dynamic composition
of the services that can be provided, given the ability of
the bot to query any SVO owned by the user present on
the platform. When the user queries the chatbot, they
will be offered various choices and based on the SVOs that
are selected, a service is composed with only the choices
made by the user. For example, in the car, in addition to
the SVOs relating to the car, you may also need SVOs re‑
lating to other environments, for the purpose of contin‑
uous monitoring, the service offered by the bot therefore
includes the data from these SVOs. In addition, you would
also have the possibility to save them and re‑propose
them at a later time as favorite services. Fig. 4 shows the
sequence diagram which illustrates the simple steps that
take place when a request is sent to the bot.
Fig. 5 – Representative low of messages exchanged between the user
and the bot
We have implemented the chatbot system that allows for
interacting with the platform and for setting and sending
requests to the devices. The queries to the bot are made
in natural language and are taken over and processed by
the DialogFlow platform, with has been integrated in our
system. The messaging system selected has been the Tele‑
gram messaging client which is used by the user.
The development of the bot was divided into two parts:
design and development of all the components necessary
for the NLU functionalities; development of the gateway
and the functions necessary to handle the events and as‑
Fig. 4 – Use case diagram sociated data in the chatbot platform. An agent has been
created within the DialogFlow platform. Agents are NLU
When the user accesses the chatbot interface, they are modules that deal with transforming user requests, ex‑
presented with the various options. Once the desired op‑ pressed in natural language, into usable data, i.e., data that
tion is selected, the bot will send a message to the bot can be associated to actions to be activated. To ensure that
gateway, who will take care of handling the request, la‑ the requests are interpreted correctly, all the possible in‑
beling it and sorting it to an Event Handler (EH). tents and entities have been loaded into the agent. Intents
The Message Handler will recognize that a message has are JSON iles and have been designed and built in order to
arrived and delivers it to a cloud function that takes care map the user’s requests with the actions to be performed
of the part of creating the response message. Then, it has in the best possible way. In order to have the best match
to collect the data in addition to the textual answers by between request and intent, new entities (all synonyms
querying the platform that contains the SVOs necessary for a given word are associated to an intent) have been
for the composition of the answer. At this moment, we developed, in addition to those made available by the plat‑
may have two different scenarios that we analyze below: form, which were able to best characterize the IoT context
• The data request fails
• The data request is achieved

© International Telecommunication Union, 2021 87