Can Haystack answer questions about a fictional novel?

[TheRabidWolverine]

Can Haystack be used to answer questions concerning a fiction novel? Typical use case of Haystack is asking questions, where the answer is present as a text snippet somewhere in the content (either in a webpage, or in a document). However, a fiction is a complex scenario, where suppose the character A is shown to love B at the beginning, and then, due to some misunderstanding, grows to hate B, and towards the end, realizing his mistake, is reconciled with B again. All this happens over the entire course of a novel, spanning 600 pages, with a host of other characters, with chapters and pages totally unrelated to A and B. Can Haystack, or any related NLP technology answer this questions accurately - "How does A's feelings towards B progress through the novel, and why?"?

[vblagoje]

My best guess, without actually trying, is that you should be able to achieve your task with LLMs good in reasoning. Even though the novel is 600 pages or so, if you setup your own Retriever Augmented Generation pipeline where the document store holds the paragraphs of the novel, you should be able to achieve this goal with OpenAI, Cohere command and Anthorpic Claude models. I also think Llama 2 would be able to generate a correct elaborate answer as well. But these are just guesses. It would help if you could order retrieved passages (e.g. by page number) but that's not too hard with a bit of hacking.