Chat

The chat endpoint uses all the previous techniques to prepare a message:

  1. Tokenization

  2. Part-of-speech tagging

  3. Lemmatization

The chatbot is then trained on lemmatized classification data, which maps various potential input expressions to a set of possible chat responses. Finally, the endpoint replies with the most probable response from the set of possible responses.

Example request

Send the following chat message:

POST /zoo-chatbot/chat
{
  "message": "any giraffes?"
}

Response

Data is truncated for readability.

Notice that the most relevant reply is returned:

Response
{
  "category": "animal-list",
  "reply": "The zoo has 10 giraffes, 3 elephants, 2 lions and 12 penguins",
  "probabilities": [
    {
      "category": "animal-list",
      "probability": 0.6893424773
    },
    {
      "category": "directions",
      "probability": 0.0776643807
    },
    {
      "category": "opening-times",
      "probability": 0.0776643807
    }
  ]
}