How to judge the types of questions and extract the relevant information in questions

[gao-stu]

I have several fixed question types, for example

"What is the relationship between Zhang SAN and Li Si?" What is the relationship between A and B? Is it type 1

"What is the characteristic of Zhang SAN", what is the characteristic of A, is it type 2

"What friends does Zhang SAN have", what xx does A have (xx is a concept, or a large range, such as friends, classmates, teachers), is type 3

How to use the large language model to return the type of question for any input question, and extract the key words in the question, such as "What friends does Zhang SAN have?" to tell me that the name is Zhang SAN and the relationship is friends. They may also ask "Who are Wang Wu's classmates?" Do you just ask a large model, or do you want to combine other tools, preferably with code examples.

[dosubot[bot]]

🤖

To categorize different types of questions and extract key information from them using a large language model, you can use Named Entity Recognition (NER) and dependency parsing techniques. Here's a simple example using the Spacy library in Python:

import spacy

nlp = spacy.load("en_core_web_sm")

def categorize_question(question):
    doc = nlp(question)
    # Extract entities
    entities = [ent.text for ent in doc.ents]
    # Extract noun chunks
    noun_chunks = [chunk.text for chunk in doc.noun_chunks]
    # Extract root of the sentence (usually the main verb)
    root = [token.text for token in doc if token.dep_ == "ROOT"]

    # Categorize based on the root verb and the entities/noun chunks
    if "relationship" in root and len(entities) == 2:
        return "Type 1", entities
    elif "characteristic" in root and len(entities) == 1:
        return "Type 2", entities
    elif "friends" in noun_chunks and len(entities) == 1:
        return "Type 3", entities
    else:
        return "Unknown type", entities

question = "What is the relationship between Zhang SAN and Li Si?"
print(categorize_question(question))  # Output: ('Type 1', ['Zhang SAN', 'Li Si'])

This is a very basic example and might not work perfectly for all types of questions. You might need to adjust the categorization rules based on your specific use case.

As for using a large language model like GPT-3, you can use it to generate answers to the categorized questions. However, the model itself does not categorize questions or extract entities. You would need to use NER and dependency parsing techniques (like in the example above) to do that.

Please note that the LangChain framework currently does not provide built-in support for question categorization or entity extraction. You would need to implement these features yourself or use a third-party library like Spacy.

Sources

• libs/langchain/langchain/chains/qa_with_sources/map_reduce_prompt.py
• docs/extras/use_cases/question_answering/question_answering.ipynb

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.