User messages
Last updated
Last updated
The User messages section in Dashboards will provide you a more detailed analysis of the virtual agent performance. This section shows information about the conversation, its confidence score, how and what was the NLP response to each user input.
To navigate it, go to the side menu and choose the bar chart icon (image below).
See in the following table the description of the information available in this section:
Messages
Presents the users’ input in the original language. Here, the system will show every message sent by the user to the virtual agent, except for those in image, gif, video, document formats.
Translation
The user's input translated into the virtual agent's primary language. This is handled by the multilingual support feature.
Language
Shows the language that was detected in the user's input. It reflects the language recognized in the user's message, not the language used by the virtual agent.
Session Code
Shows the session code in which the message is part
Code assigned to the user who initiated the conversation
Confidence
Confidence score is represented by a number between 0 and 100. The higher the score, the greater the confidence of the virtual agent answer (see NLP/Knowledge AI Response below). The confidence level of the NLP and the Knowledge AI are set independently in the Parameters section.
NLP/LLM Response
Specification of what was delivered to the user. If the answer gets a higher score, the system will display exactly what was delivered (whether an intent, entity, document, or question) and its name.
NLP: If the message is below the confidence level set, it will appear in the table as "none". In this case, click on "(view details)" to find out which intent was closest to that input.
LLM (Knowledge AI): For documents in TXT and PDF, the system displays the paragraph from which the answer was extracted. If disabled, messages that would be attributed to a document and delivered, will appear as "none".
Answer
Displays the answer that was delivered to the user.
Answer Name
NLP/AL answer name registered in the knowledge base and delivered to the user.
Date/Hour
Corresponds to the exact time the user message was sent.
Important: data will be presented according to the user's timezone.
Channels
Where the user sent the message to the virtual agent. The column displays the channel type and its channel name.
Messages interpreted by NLP have a different confidence score when compared to the messages interpreted by Knowledge AI. Thus, the percentages must be compared separately.
By default, when you access the User messages section, it will display data for “this month”, ie, the current month. That means, if you access the dashboard on June 5, the filter “this month” will resume data from the June 1-4.
If you want to change this setting, click on the Period filters on the top bar (see image below). You can browse the calendar and select any date or date range within the last 12 months.
On the calendar you'll also find the Date/Hour filter, which corresponds to the exact time the user message was sent. The hour follows the AM-PM format.
Important: Date/Hour messages will be presented according to the user's timezone.
Where the user sent the message to the virtual agent. The column displays the channel type and name.
Corresponds to the tags related to this user message, according to what was registered in the cell of intent, entity, answer, not expected answer, and Automated Learning document and questions.
Click on the line for each message to see more details of a particular message. The system will display a box reproducing the conversation of which the message is part, on a specific channel.
Check the “Show info” box to view information such as the name of each component in the Dialog Manager (see example below).
You can refine your data visualization through the Search bar. Messages can be filtered by session code, component type (intent, entity, document, question), period, channels, or tags.
The searched term must be written precisely, and half the word won't make it. We recommend you copy the whole term obtained in a broader search, and filter for a known, specific component.
Click the Export option in the top right corner of the page.
After you click the option, you'll see the modal below.
The Number of messages field displays the total number of messages. It is not possible to edit it.
The Export type field options are:
Random: random messages, within the total clipping (default)
Sequential: shows the most recent messages to the oldest, within the clipping
Higher confidence: messages with the highest to lowest confidence level within clipping
Lower confidence: messages with lowest to highest confidence level, within clipping
A summary report with the current filters applied will be prepared in a XLSX format.
The file will be exported, obtaining data filtered by the currently applied filters on the screen's header, with a maximum 3 months extraction limit.
The exported file will be available for download in the Reports section once generated for the period of 30 days. Reports with a large ammount of data and filters may take a few minutes to show up.
Important: The file will export messages from full conversations. When reaching the limit of 10,000 messages, any incomplete conversation will be discarded.
You will also find an option to export a report from the conversations at the top of the detailed conversation window, by simply clicking a button. In that instance, you won't apply the current filters, but instead, download all of the messages of that given conversation.
As with the other option, the report will be available in the Reports section.
