Follow these instructions to install ChatNode and its prerequisites. The code is developed on Ubuntu version 20.04. The required packages are: speech_recognition, pyttsx3, and openai.
also you need to have "openai_api_key"
As shown in Figure 1:
ChatNode is a system designed to facilitate interactions between a client and a chatbot. It operates by processing inputs and generating outputs based on predefined roles and message histories. The ChatNode object requires five inputs:
- status: Represents the possible status of the node output and is usually set as
status = """{ "status": "true" or "false" or "nan"}"""; - role: Defines the role that the chatbot node will assume during interactions;
- message: A dictionary used in the ChatGPT API prompt, containing the history of messages exchanged between the client and the chatbot. This dictionary has a specific format:
{"role": "system", "content": "Enter the role text"}: Sets the behavior or context for the assistant.{"role": "user", "content": "The client request"}: Represents the user's input or query.{"role": "assistant", "content": "The ChatGPT response to the user request"}: Contains the assistant's response to the user's input.
- output_json_format: Specifies the format of the data extracted by the node, for example,
output_json_format = """{ "name": " ", "age": " " }""".
To initialize a ChatNode object, use the following code: N1 = ChatNode(goal, status, role, messages, output_json_format). To evaluate the ChatNode, use the method: results, summary, json_out, messages = N1.run(messages). To end the conversation due to lack of information or by the client's request, use the method: N1.endconv().
The node outputs the following variables:
- results: The current status of the ChatNode;
- summary: A summary of the findings from the ChatNode;
- json_out: The extracted data formatted according to
output_json_format; - messages: The resulting conversation messages between the client and the ChatNode. This variable is useful for feeding into subsequent ChatNodes in a conversation.
The ChatNode object includes two main methods: 1. run(messages): Executes the ChatNode with the provided messages. 2. endconv(): Ends the conversation when necessary.
The demo showcases a hot drink shop example, where the client interacts with the chatbot to place an order.
As shown in Figure 2, the conversation is divided into five separate nodes:
- The first node is used to collect the client's name and age.
- The second ChatNode checks if the client prefers delivery or will pick it up.
- The third node provides the client with the menu and requests their order.
- The fourth ChatNode summarizes the client's request, including the total price, and requests confirmation.
- The fifth ChatNode is optional and is used if the client requests delivery, asking for the client's address.
For more details about this demo, refer to the YouTube video: Watch the video on YouTube