NLU Performance¶
NLU Performance Dashboard¶
The currently displayed dashboard can be seen next to the agent in the top left of the screen.
The data presented on this dashboard is related to the Natural Language Understanding (NLU) section.
More information regarding "NLU" and "Intents" you will find here: NLU Overview.
| Graph title | Description |
|---|---|
| Understood messages | This percentage value, displayed as a report arranged in a tile, relates to the evaluated messages and is helpful to improve your agent in natural language understanding. A message is counted as understood if it has triggered an intent, a slot is matched, or it is marked as understood via the Code Node or Overwrite Analytics Node. A message is not counted as understood if it is marked as Don't count in the Overwrite Analytics Node or sent during an active handover without configuring Agent Assist Flow. Clicking on the three dots menu item you can select the "Go to Intent Trainer" option that leads you to the Intent Trainer in Cognigy.AI to improve your Agent. |
| Average Execution Time | The average time it took to execute the Flow in ms. The execution time is displayed as a report arranged in a tile and as a graph on a timeline scale as well A tooltip will display additional info in the line chart depending on the cursor position. Time scaling depends on the global filter "Timeframe" setting. |
| Avg Intent Score | This tiled report shows the average figure between 0 and 1 and indicates the quality of your Agent how "good" Intents were found in conversations, depending on the global filter "Timeframe" setting. More information about Intents you will find here: Machine Learning Intents |
| Max Intent Score | This tiled report shows the figure of how many Intents the Agent matched with 100%. |
| Understood/Misunderstood Messages | This stacked bar chart shows bars on a timeline (per month) which figures out in percentage the understood /misunderstood messages in conversations. Tooltip info displays the exact number, percentage value, and month. The "Understood" values form the lower parts of the stacked bar chart. A message is counted as understood/misunderstood if it has triggered an intent, a slot is matched, or it is marked as understood or misunderstood via the Code Node or Overwrite Analytics Node. A message is not counted as understood/misunderstood if it is marked as Don't count in the Overwrite Analytics Node or sent during an active handover without configuring Agent Assist Flow.Clicking on the three dots menu item you can select the "Go to Intent Trainer" option that leads you to the Intent Trainer in Cognigy.AI to improve your Agent. |
Note
The on top of the dashboard placed blue tiles show currently valid data in a "live" mode. That means that data will be refreshed every 30 seconds to be updated.
Charts display data based on the global filter settings.
| Graph title | Description |
|---|---|
| Top Intents | The highest-rated numbers of found intents are displayed in a bar chart (the x-axis is scaled in thousands). More information about Intents you will find here: Machine Learning Intents |
| Top Slots | The highest number of found slots are displayed in a bar chart (the x-axis is scaled in thousands). More information about Slots you will find here: Slot Mapping |
| Intents by Score Range | The scoring range of found Intents is visualized in this bar chart. Every score range bar shows a min-max value and has an average value marker. Clicking on the three dots menu item you can select the "Go to Intent Trainer" option that leads you to the Intent Trainer in Cognigy.AI to improve your Agent. More information about Intent Trainer you will find here: Intent Trainer |
| Top Intent Scores | Top-rated numbers of Intents are displayed in a bar chart. More information about Intent analysis you will find here: Intent Analyzer |
Note
When hovering over any chart or metrics field, rating details will be displayed in a tooltip. Consider that rating data depend on the settings in your global filter options.
Download Report¶
To download a report as a CSV file for the chart, use 
menu.
To create a report file, refer to the Download Report article.