Page 100 - 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

of use we needed. Therefore, various attributes have been you want to do other operations?” answering “no”, they
created within each entity, with their synonyms, in order activate the reset of the contexts and the “exit_facts” in‑
to be able to extract the values of the input parameters tent, initializing the bot for new requests.
without these being necessarily identical to those we had
foreseen in the phase of creating the intents. In this way, 7. EXPERIMENTAL RESULTS
every time an intent is activated, the platform will return a
JSON ile with the information on: Intents; Action; Event; The experiments have been conducted to assess how ef‑
Response; Contexts; Parameters; Score. fective the chatbot was in understanding the user re‑
quests and perform action accordingly. In the following
At each request, these parameters are updated based on
section we describe the performed tests with reference
the intent that is activated at that time so that, at the next
to the access to the platform and perform device setting
call, the new intent to be activated is also chosen based on
and request data of interest. The performance of query
the previous parameters. In this way, it is possible to com‑
matching results has been also analysed.
pletely contextualize the conversation. In fact, the con‑
texts section contains all the active contexts in that call
ordered according to their lifetime. With this mechanism 7.1 Access to the platform
it was therefore possible to implement a management of Fig. 7 shows the interaction with the chatbot with the in‑
the state of the bot allowing for the exchange of variables tent of accessing the Lysis platform. The igure shows the
between subsequent requests. low of questions and answers between the bot and the
user. Remembering that it is necessary to authenticate,
6. IMPLEMENTATION to be able to use both the chatbot services and to have
access to the resources made available by the platform,
the irst question asked was on how it was possible to au‑
thenticate. The bot’s response was a message with the
instructions on how to log in and, once logged in, it sug‑
gested to the user that it was necessary to enter some ad‑
ditional information to complete the iguration. The
user then asked how to enter the owner key and the SVO
root, to which the bot answered by providing a descrip‑
tion of where to ind them and information on how to en‑
ter this information. This was possible thanks to the fact
that when you ask for information either on the key or on
the root SVO, the respective context is activated allowing
you to keep track of what was previously requested. Af‑
ter completing the con iguration and logging in again, you
can see that the welcome message is simply given, a sign
that the con iguration was successful.
Fig. 6 – Flow diagram of the bot
7.2 Setting the devices and data retrieval
As we can see in the diagram in Fig. 6, it is possible to en‑
ter one of the intents based on what is asked. Let’s refer to We also tested the ability to request data from the plat‑
our use case previously described related to the surveil‑ form and apply the desired settings to the available de‑
lance. As we said, the user has to ask the bot for send‑ vices, all with the most natural language possible. Fig. 8
ing the video stream from the video camera device; if it is shows how it was simple to request the list of devices and
turned off, it will irst be asked to turn on and then send their current status. If you want to switch a device on or
the data. When the user needs to use the bot, they have to off, simply specify which of these actions should be ap‑
log in and then send the message “show me what happens plied and the setting will be performed. The request for
at home”. This activates the “svo_facts” intent through the data was also handled in a similar way; in the sentence it
“svo_facts” event and setting “svo _followup context” with is just needed to specify which data is needed and from
a certain lifetime set as the current context. The bot then which environment to obtain the requested data.
responds by giving the list of objects that are indexed in
their home. By selecting the video surveillance, the user 7.3 Elaboration time and latency evaluation
remains within the “svo_followup context” and then ac‑
tivates the loop indicated with “yes” in the diagram; by The time spent processing the information sent to the
reactivating the svo_facts intent, the context is updated chatbot was very low. This happens thanks to the use of
again and the bot responds by displaying the state of the an ML engine that is fully running in external services and
object. On the basis of this, then it allows for the choice for the ef icient implementation of the sample questions
whether to activate it or not. Once concluded, if the user on the platform. This allows for a low latency between
decides to perform different actions to the question “do a request and the response and this makes the user

88 © International Telecommunication Union, 2021