How can I identify negative answers using Watson Conversation

Question

For example: if the user writes to the Watson conversation service:

"I would not want to have a pool in my new house, but I would like to live in Kondo"

How can you know that the user does not want to have a pool, but he likes to live in a condo?

Answer 1

This is a good question, and yes, it is a bit complicated ...

Currently, it is best to provide as many examples of statements that need to be classified as specific intentions as examples of learning for that intention - the more examples you will provide the more reliable NLU (natural language understanding).

Having said that, note that using examples such as:

"I would like to have a pool in my new home, but I would not want to live in a condo"

for intent-pool and

"I would not want to have a pool in my new house, but I would like to live in Kondo"

for intent-condo will force the system to correctly classify these sentences, but the difference between them can be quite small (due to the fact that they are very similar when you look only at the text).

So the question here is whether the system should classify such intentions “out of the box” or instead teach the system simpler examples and use some form of meaning if you see that the top N intentions have low confidence differences.

Answer 2

Sergio, in this case you can check all the conditions that apply with peers node (continue), and your negative (example else) you can use "true".

Try using intentions to define flow and an entity to determine conditions.

More details: https://www.ibm.com/watson/developercloud/doc/conversation/tutorial_basic.shtml

PS: you can get the value of an object using:

Answer 3

This is a typical multi-use conversation service scenario. Each user says something, all 10 best intentions are identified. You can change your JSON dialog editor to see all intentions.

 { "output": { "text": { "values": [ "<? intents ?>" ], "selection_policy": "sequential" } } } 

For example, when the user makes an expression, it will cause two intentions, you will see that the intentions [0] .confidence and intents [1] .confidence will both be quite high, which means that the “Conversation” identified both intentions from the user's text.

But at present there is a serious limitation, there is no guaranteed order for identified intentions, that is, if you said, “I would not like to have a pool in my new house, but I would like to live in Kondo,” there is no guarantee that a positive intention will will_not_want "will be intentional [0]. The alleged and intentional "will_want" will be intentions [1] .intent. Therefore, it will be quite difficult to implement this scenario with greater accuracy in your application.