Food Ordering and Help & Support Chatbot

Food Ordering Chatbot using RASA.

Steps:

Clone the repo.

1. Install Anaconda3
2. Create Environment and install requirements and activate the environment

$conda create --name <env-name> --file requirements.txt

Main Bot file to run:

$python bot.py

Some points about RASA:

1. Requires installations of multiple components
2. Requires tech knowledge
3. Open-source
4. No interface needed, can write down in json or md format
5. Host the data or bot on our own server

Building the conversational states

First let's define how our chatbot should converse like. Just a flow diagram.

RASA core training

1. training_data: these are the stories that define the normal converation.

   ## Generated Story -7329490837376575624
   * greetings.hello
       - utter_greetings.hello
       - utter_agent.welcome
       - utter_ask.cuisine
   * cuisine.type{"cuisine": "mexican"}
       - slot{"cuisine": "mexican"}
       - utter_display.menu
       - utter_select.item
   * confirm.affirm
       - utter_order.placed

2. domains: this defines the environment in which bot operates. It specifies intents,entities,slots,actions the bot should know about and templatesfor the things bot can say.

3. load_agent: bot is first loaded with some parameters to determine how the stories and other data can be converted into features for training the bot.

4. policies: this is one of the parameter in load_agent. The policy decides which action to take at every step in the conversation. Find more about policies here. Note: tensorflow_embedding pipeline can be used to assign two or more intents to a single input message.

5. load_data: you load the data using the stories training file and defined domain file.

6. training: policies work in ensemble, we can pass more than 1 policy, it will train separately and will be used together in ensemble for prediction.

7. persist: when model is trained, it is then persisted to some storage.

Predictions

1. load model: first task is to load the trained model.
2. start_server: we can user Flask.
3. interpreter: it's job is to classify the intent with entity extraction and these entities can be used as slots in the conversation.
4. create_tracker: this is created to track all the objects of a key say user_id. To make it easy to use slots of a particular user_id. RASA has inbuilt Memory Tracker. We can use MongoTrackerStore when scaling up.
5. create_features: create freatures of the objects present in tracker based on the policy
6. policy_ensemble: if we have more than one policy, each policy will output a score and max is taken to determine the next action
7. update_tracker: once the action is performed, we will update the tracker.
8. bot_text: the bot message to user (the action).

Preparing and training the chatbot

The code is provide here : github/food-ordering-chatbot

Other components that are required

• database for item availability
• storage to keep order until it's placed/confirmed.
• storage to keep user's data (id, name, phone-number, address, previous orders)
• payment gateway/wallet
• notifications : can be a genral notification or for some coupons/codes
• assigned person for delivery ( to update on chatbot) ( person_name, phone-number)

TODO

• Flask web app

I will include basic flask application for chatting with the bot.