When predicting entities in a document, the NLU Service does not return any information about where the entities were found in the text. Since we use custom models with a lot of numerical entities, it becomes very hard to find the sentences where an entity was found by using only the text mention of that entity. One solution would be to split the text into sentences and call the entity prediction for each sentence, but that would increase both the cost and processing time. We would like the NLU entity prediction to return the start/end indexes of each predicted entity in the text. That way, we would be able to give more context to the users by showing the sentence where an entity was found instead of only the entity mention.
Why is it useful?
|Who would benefit from this IDEA?||As a developer, I want to get the indexes of the predicted entities in the text, so that I can show more context to the user when extracting information from large text documents.|
How should it work?