Recognition Analysis Dashboard

This dashboard provides users with insights into how their projects intent recognition is performing and will help you to improve the recognition of end users questions in your database.

The recognition rate shown in this dashboard is a calculation of the total amount of interactions that received an answer divided by the total amount of interactions. This new iteration of the recognition rate is easier to understand and better aligns with Conversational AI Cloud's AI-driven recognition.

Recognition Analysis Dashboard (Entity Based)

This dashboard shows where you might be missing recognition, based on the entities you've defined and words from user questions you haven't made a decision about yet.

This lets you quickly analyze where you need to add content, which entities need to be added to your database, or which entities are unnecessary. You select the example question and add it in the Q&A editor, with the answer of your choice.

The dashboard will show you the top 100 entity combinations that were detected in user questions, but not used as such for recognition. You can choose your time period to check larger entities in data over time, and of course, you decide what frequency is worth spending time on.

The standard dashboard filters apply. In addition, you can filter the dashboard on an entity to focus on missing recognition using entity.

Recognition Over Time Widget

This chart allows you to quickly spot trends in the overall recognition of your knowledge base. A dip in the questions answered and/or recognition rate means your customers are asking very different questions to the day(s) before. You may need to update your knowledge base to take these new questions into account. The rest of this dashboard is set up to help you do that.

Intent Based Recognition:

Entity Based Recognition:

No Recognition Widget

Top 50 user questions that did not receive an answer and where no intent was detected.

User Questions Answered Without Intents

This widget shows the top 50 user questions that received an answer but where no intent was detected. Consider adding these questions to your intent model.

Missing Entity Combinations Widget

This widget functions as a filter for the entire dashboard. Right-click on a missing entity combination and choose 'Select' from the menu that appears. The dashboard will adjust to only show the information about that one missing entity combination.

If the dashboard is filtered to a specific missing word, it will show the missing entity combinations for Q&A questions using that word.

The missing entity combinations widget is the core of the dashboard. This widget gives the top 100 combinations of entities for which there was not an exact match in the Q&A of your knowledge base at the moment the question was asked. The columns after give you counts of how often this combination of entities occurred and on how many days in your selected time period.

Do keep in mind that if you add, edit or delete entities, the exact same question may give rise to a different entity combination.

You can use the drill options to see some specific information about the entity combination, such as the interaction values leading to it and the matched words from entities.

Drilling into the missing entity combination also gives you the option to go to the Gap Analysis Log. This will open a pop-up screen with the questions for this entity combination and the answers they received.

Words Missing From Entity Recognition Widget

This widget functions as a filter for the entire dashboard and only shows missing words occurring more than 4 times over the filtered period. Right-click on a missing word and choose "Select" from the menu that appears. The dashboard will adjust to only show the information about that one missing word.

This widget shows the words that were used in user questions but that weren't present in any entities, nor added to the stop words, together with a count of how often they occurred.

If the dashboard is filtered to a specific entities combination, this widget will only show the words from the questions using the selected entity combination.

Recognition Quality Distribution Widget

Recognition quality shows you how well a question was recognized by our entity driven rule-based recognition engine. It’ll only show for recognition that wasn’t caught by your trained intent model and was matched purely through entities, and keywords. The recognition quality is a calculation of the number of condition sets (entities, keywords) present in the end-users question compared to the question on the article that it matched to. Recognition quality can be used to determine where you might want to expand an entity, add additional keyword triggers to a Q&A to improve the performance of your rule-based recognition engine.

If you select a Missing Entity or Missing Word, this widget will reflect the questions with this entity combination or word.

Download Logs

At the bottom of the dashboard you can download logs:

• Get a list of activated entities
• Download the full Gap Analysis log
• Download the log of questions answered by Generative AI (This log will include the sessionuuid so the whole session (and any additional information) can be looked up in the CustomerJourneys dashboard too).

Video

Below you can find videos on how you can work with the Recognition Dashboard in Conversational AI Cloud.