Summarizer

The Summarizer tool in Therix condenses lengthy texts into concise summaries. It analyzes documents, extracts key information, and generates coherent summaries, enabling users to grasp core content efficiently.

The Summarizer uses two main methods: "Extractive" and "Abstractive" summarization. They work differently but both help make summaries, depending on what you need.

Summarizer - Abtractive

Abstractive Summarization in Therix is a smart way to make short summaries from long texts. Unlike Extractive Summarization that just rearranges the text, Abstractive Summarization really understands the text. It figures out the main ideas and context, then writes original summaries that capture those ideas in a new way.

This method lets you be creative in summarizing and isn't limited by how the original text is structured. Abstractive Summarization is great for simplifying complex information, making it easier to understand for making decisions and sharing knowledge in different areas.

Sample Output:

{'answer': '\nBeets are a root vegetable rich in fiber, folate, manganese, and antioxidants, and can improve athletic performance and reduce blood pressure. More research is needed to fully understand their health benefits.', 'session_id': UUID('b9548328-6982-40e9-9cea-06208090f25f')}

Summarizer - Extractive

In the Extractive Summarizer of Therix, the aim is to make a brief summary of a text in a JSON format. It picks out important sentences or phrases from the text and puts them in a structured way in JSON.

Users can also give a custom model to customize how the information is organized in the JSON. This gives more control over how the summary looks. If you don't specify a custom model, the system itself decides how to organize the summary in JSON, making sure it's clear and concise.

Whether using the default JSON setup or a custom model, the result is a neat and informative summary in JSON. This helps in understanding big chunks of text quickly and making decisions based on them in different fields.

Custom model is made by extending from the SummarizerOutputModel of therix.

from therix.core.summarizer_output_model import SummarizerOutputModel

class TopicModel(SummarizerOutputModel):
    mainTopic: str
    subTopic1: str
    subTopic2: str

Sample Output:

{'answer': {"mainTopic": "Superfoods", "subTopic1": "Health Benefits", "subTopic2": "10 Superfoods"} , 'session_id': UUID('3f2d3cef-8091-4e0d-accd-71b6387938c7')}

How you can integrate summarizer in your code:

• Import Necessary Modules: Begin by importing the required modules for configuring the Summarizer.

from therix.core.summarizer_config import SummarizerConfig
from therix.utils.summarizer import SummarizerTypeMaster
from therix.core.summarizer_output_model import SummarizerOutputModel  # Import SummarizerOutputModel if needed 

• Instantiate SummarizerConfig:

Create an instance of the SummarizerConfig class by specifying the type of summarizer and, optionally, the Pydantic model for structuring the JSON output.

• For Extractive Summarizer with custom Pydantic model

summarizer_config = SummarizerConfig(SummarizerTypeMaster.EXTRACTIVE, TopicModel)

• For Extractive Summarizer with custom system prompt

Use the same placeholders for {context} and {response_schema_json}

#For example:
sys_prompt = 
        """
        You are a doctor chatbot
        Summarize the provided context below as a doctor:
        {context}

        ---

        Craft your response with conciseness and accuracy, including only the information provided in the context. 
        Use null values for any missing information.

        Please structure your response in the following JSON format:
        {response_schema_json}
        """

• For Abstractive Summarizer

summarizer_config = SummarizerConfig(SummarizerTypeMaster.ABSTRACTIVE)

• Add Summarizer Configuration to Agent: Use the .add() method to add the SummarizerConfig to your Therix agent.

agent.add(summarizer_config)

Example

agent = Agent(name="Summarizer Agent")
    (agent
    .add(SummarizerConfig(SummarizerTypeMaster.EXTRACTIVE,TopicModel))
    #custom-prompt
    .add(SystemPromptConfig(config={"system_prompt" : sys_prompt}))
    .add(// Any other configuration you want to add)
    .save())

     
    answer = agent.invoke(text)

By following these steps and adjusting the parameters as needed, you can seamlessly integrate the Summarizer into your Therix agent. Whether you require Extractive or Abstractive summarization, with or without a custom model, Therix provides the flexibility and functionality to meet your summarization needs effectively.

How to pass dynamic system prompt

You can also pass a dynamic system prompt while invoking already created agent, by just adding a parameter to agent.invoke method

Example

#just define s dynamic system prompt if you don't want to use already added system prompt or don.t want to add a system prompt to your agent


dynamic_system_prompt = """
        You are a an eXpert in summarizing documents
        Summarize the provided context below as an expert:
        {context}

        ---

        Craft your response with conciseness and accuracy, including only the information provided in the context. 
        Use null values for any missing information.

        Please structure your response in the following JSON format:
        {response_schema_json}
        """


#And instead of adding this system prompt to the agent configuration we can directly pass it to our invoke method 


agent = Agent(name="Summarizer Agent")
    (agent
    .add(SummarizerConfig(SummarizerTypeMaster.EXTRACTIVE,TopicModel))
    .add(// Any other configuration you want to add)
    .save())

     
    ans = agent.invoke(question = text, dynamic_system_prompt=dynamic_system_prompt )
    print(ans)