Suppose we have an utterance like I don't want a car but I want a bike. We would have an entity list having 'Car and bike' to find out the model the user wants or not. However, how shall we determine entity for DON'T WANT and WANT. How do we know the user said he 'does not' want car, he 'wants' bike. Can we train Luis on negation and negatives?
2 Answers
Not really except if you expect me two entities. The best way I would do it it is by splitting the sentence in two through words like and or but, then identify the intent for each and then detect if there is any negative entities
You can use any of the following options:
- Add more utterances with syntax related to negative responses and train them before testing. The more the better.
- You can make use of Text Analytics API. The Text Analytics API's Sentiment Analysis feature evaluates text and returns sentiment scores and labels for each sentence. After sending a sentiment analysis request, the API returns sentiment labels (such as "negative", "neutral" and "positive") and confidence scores at the sentence and document-level.
- You can also try creating a new intent that would handle all the negative responses and train them with as many negative utterances as you can.
Hope this helps!!
Related