Tasks: Article (Q&A) Recognition

This dashboard shows you some suggestions for improving recognition. Most of the standard dashboard filters apply. However, there is no context filter on this dashboard. The answers of Conversational AI Cloud can be context-dependent, but recognition works on all questions in the same way.

Note that you only have access to this dashboard if you have the right to edit entities in Conversational AI Cloud. The widgets on this dashboard are filtered from the Recognition Analysis dashboard or intended to improve the suggestions of that dashboard. You can use the Recognition Analysis and Q&A Questions Dashboards for further investigation.

Task: Add These Words to Entities or Stop Words OR Use Them in Compounds

These two tables show words from your users' questions that are not included in an entity or assigned to the stop words. Based on how meaningful these words are, there are a couple of things you can do.

• High-value words: If they are important for correct recognition of the question, they should be in entities. You can add them to existing entities or create new ones

• Low/no value words: If these words do not add to the meaning of your users' questions, you should add them to the stop words. This will help to improve your overall recognition

• Miscellaneous words: In some cases, you may choose to leave some of these words out of the stop words to make a fine distinction in recognition between two very closely related questions. These are probably not very meaningful on their own, but they are in combination with another word. In that case, you could create a compound entity in which you combine the word presented in the table with the word you need to distinguish between in your recognition. For example: If your logged-in environment is called "My Account", you may want to create the entities MY_ACCOUNT and ACCOUNT as separate entities, instead of leaving 'my' as a word that is not in an entity or creating MY as an entity on its own. Add the word to the stop words for all other cases

Note: Once you've added words to your stop words, you should also check the overlaps on the Task screen. You may have created an article overlap by marking some words as irrelevant.

Task: Add These Questions to Your KB or Work on These Entities - There Is Missing Recognition Here

This table shows the Top 5 missing entity combinations from the Recognition Analysis dashboard, together with a stereotypical user question. You should consider adding these questions to your knowledge base. The entity already exists, but there is no or not sufficient recognition in your articles. Even if the recognition rate for these questions is already quite high, these are the most recurring questions that aren't being recognized as well as they could be.

If (almost) this exact question is already in your knowledge base, you should investigate the entities associated with the question. There are two possible entity issues here:

• Full overlap: two entities have the same word(s) in them

• Partial overlap: one entity has a compound term and another entity matches on a single word from the compound

There is a table further down on this dashboard that gives information about full overlaps between entities. You should resolve any full overlaps. Partial overlaps are not usually harmful to the recognition of longer questions, but they can have a disproportionate effect on recognition for short questions. In that case, consider rephrasing the questions in your knowledge base.

Task: Review Entity - Recognition May Be Too Broad

Entities that are shown here cover so many different terms that the contents may no longer be proper synonyms. This can be intended, e.g. a list of locations for your stores or a range of products may easily be longer than 20 items. For other entities, you should consider whether the entity is still specific enough, or if it has expanded to cover more than it should. If you encounter such an entity, you could divide it into two or more specific entities.