How to use llama_index to summarize information from multiple articles

[madawei2699]

When I use llama_index to summarize information from multiple articles, my code is like this:

def get_answer_from_llama_web(message, urls, logger):
    logger.info('=====> Use llama with chatGPT to answer!')
    combained_urls = get_urls(urls)
    logger.info(combained_urls)
    documents = get_documents_from_urls(combained_urls)
    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo"))
    logger.info(documents)
    index = GPTSimpleVectorIndex(documents, text_qa_template=QUESTION_ANSWER_PROMPT)
    return index.query(message, llm_predictor=llm_predictor)

This code will fetch multiple articles from the URLs and use the GPTSimpleVectorIndex to summarize information. What I want is to summarize the information by article. But when given the prompt is:

For the following articles to write an article recommendation email out, my requirements.
1. write a separate summary paragraph for each article, sorting out the most valuable parts

https://mp.weixin.qq.com/s/JFbdL0oUAlZU-os6CerSog

https://mp.weixin.qq.com/s/SZsW-SlkWdQq-aAyrCikDw

https://mp.weixin.qq.com/s/Wx5q-rEwG4sROvnewGxWrw

But the answer is totally wrong which is looks nothing to do with the content of the article. My question is what index is suitable to do this work?

[iraadit]

I would also be interested.
And also by how to compare some elements between the different articles.

[Disiok]

Hey @madawei2699 @iraadit, if you want to summarize articles, using a GPTListIndex would be more suitable.
See https://gpt-index.readthedocs.io/en/latest/guides/use_cases.html for more details.

Going to close this since it's not an issue. Please join the discord community (https://discord.gg/dGcwcsnxhU) for better support!