Sentiment reviewer
The sentiment reviewer allows users to extract NER (named entity recognition) and intents from any given sentence. The responses (or output) of the agent is generated by decorating the input text with metadata to give visibility to the user of the results.
The agent can be fine tuned to extract specific entities and intents by creating Training data
for the Sentiment Reviewer and in combination with the Trainer
module to keep track of the evolution of the agent capabilities.
Both Training data
and Trainer
can be accessed only by the following user types:
Root
SuperAdmin
AIEngineer
Menu location
The Sentiment reviewer is located from the main menu:
Sentiment reviewer
Review feature
Users have the ability to review content using the What You See Is What You Get (WYSIWYG) interface and when a user clicks the Review
button, the agent will provide analyse the content provided extrapolating the NER(entities) and intents (e.g. sentiments) found in the text.
The user cannot edit the NER or intents extracted by the agent however, during the training process it is possible to leverage the Review
functionality to identify the entities and intents extracted by the agent by default to only add/edit or delete the entities and intents as required.
Due to the nature of the NER and intent extraction, this feature can be affected by cold starts and may require more than 30 seconds to complete depending on load and complexity of the content.
Train feature
Users enabled to train the agent can choose to use the Train mode
at the top left of the screen. This will keep all the data generated by the agent allowwing the users to make corrections or add new data to the agent to train on.
The user can choose to use the Train mode
without saving the document, with the ability to save the document at a later stage after the annotation is complete
The documents saved in Train mode
will be saved in the Training data
module.
Similarly to the documents present in the Knowledge hub
, users can real time collaborate on the same document.