6. Converse Bot Guide#

We start by importing the converse bot from botiverse.bots and as well import chat_gui for testing purposes

from botiverse.bots import ConverseBot
from botiverse import chat_gui

The converse bot is based on the state-of-the-art model Text-to-text Transfer Transformer (T5) by Raffel et al. It was finetuned on a large corpus of conversations to be suitable as a chatbot here at Botiverse and is capable of being further finetuned using company data to simulate a chatbot the interacts similar to human agents.

Hence, the starting point is a dataset of conversation by human agents similar to the following:

6.1. Dataset Sample#

[
  [
    "how long does it take for an order cancellation to process?",
    "Great question! Generally, if an order hasn't begun processing, the ..."
  ],
  [
    "how can I report an undelivered package with no tracking updates or news from the shipping company?",
    "Greetings! Who is the carrier and what is the current status of your package listed here: [URL] ? ^TL"
  ],
  [
    "i have been waiting for a parcel which has been dispatched and said to arrive on Friday ...?",
    "Uh no! That is not what we expect! We would like to take a closer look into this, in ..."
  ],
]

The json format is very easy to deal with and is just a list of questions and their answers by human agents as found in real conversations. The typical firm with sufficient experience can easily have a lot of this data.

6.1.1. Read the Data#

We start by reading the data

bot = ConverseBot()
bot.read_data("conversations.json")

6.1.2. Train and save the chatbot#

And then training the model and saving it. The train method supports arguments for the number of epochs and the batch size and provides a way to measure the chatbots accuracy on the training data as it trains.

bot.train(epochs=1, batch_size=1)
bot.save_model("conversebot.pt")

Train Acc: 0.93: 100%|██████████| 240/240 [00:01<00:00, 220.21it/s]

6.1.3. Infer#

Once trained we can use the infer method as usual. It takes a string for the user input and returns a string for the chatbot’s response.

bot.infer("What is Wikipedia?")

"Hello! Welcome to our university's website."

6.1.4. Deploy the Chatbot#

Finally, we can deploy the chatbot as needed.

chat_gui('Converse Bot', bot.infer, server=True)

Converse Bot Guide

Contents