$200 劝退，无缘 Deep Research，可以试试 AutoSurvey

paper_texts = ''
for t, p in zip(references_titles, references_papers):
    paper_texts += f'---\n\npaper_title: {t}\n\npaper_content:\n\n{p}\n'
paper_texts+='---\n'

'''
Here is the requirement you must follow:
1. The content you write must be more than [WORD NUM] words.
2. When writing sentences that are based on specific papers above, you cite the "paper_title" in a '[]' format to support your content. An example of citation: 'the emergence of large language models (LLMs) [Language models are few-shot learners; PaLM: Scaling language modeling with pathways]'
    Note that the "paper_title" is not allowed to appear without a '[]' format. Once you mention the 'paper_title', it must be included in '[]'. Papers not existing above are not allowed to cite!!!
    Remember that you can only cite the paper provided above and only cite the "paper_title"!!!
3. Only when the main part of the paper support your claims, you cite it.
'''

'''
You are an expert in artificial intelligence who wants to write a overall and comprehensive survey about [TOPIC].\n\
You have created a overall outline below:\n\
---
[OVERALL OUTLINE]
---
<instruction>

Now you need to help to refine one of the subsection to improve th ecoherence of your survey.

You are provied with the content of the subsection "[SUBSECTION NAME]" along with the previous subsections and following subsections.

Previous Subsection:
---
[PREVIOUS]
---

Subsection to Refine:
---
[SUBSECTION]
---

Following Subsection:
---
[FOLLOWING]
---

If the content of Previous Subsection is empty, it means that the subsection to refine is the first subsection.
If the content of Following Subsection is empty, it means that the subsection to refine is the last subsection.

Now edit the middle subsection to enhance coherence, remove redundancies, and ensure that it connects more fluidly with the previous and following subsections.
Please keep the essence and core information of the subsection intact.
</instruction>

Directly return the refined subsection without any other informations:
'''

def replace_citations_with_numbers(self, citations, markdown_text):
    ids = self.db.get_titles_from_citations(citations)
    citation_to_ids = {citation: idx for citation, idx in zip(citations, ids)}
    paper_infos = self.db.get_paper_info_from_ids(ids)
    temp_dic = {p['id']:p['title'] for p in paper_infos}
    titles = [temp_dic[_] for _ in tqdm(ids)]
    ids_to_titles = {idx: title for idx, title in zip(ids, titles)}
    titles_to_ids = {title: idx for idx, title in ids_to_titles.items()}
    title_to_number = {title: num+1 for  num, title in enumerate(titles)}
    title_to_number = {title: num+1 for  num, title in enumerate(title_to_number.keys())}
    number_to_title = {num: title for  title, num in title_to_number.items()}
    number_to_title_sorted =  {key: number_to_title[key] for key in sorted(number_to_title)}

    def replace_match(match):
        citation_text = match.group(1)
        individual_citations = citation_text.split(';')
        numbered_citations = [str(title_to_number[ids_to_titles[citation_to_ids[citation.strip()]]]) for citation in individual_citations]
        return '[' + '; '.join(numbered_citations) + ']'

    # 利用正则表达式进行标题到文本编号的替换

    updated_text = re.sub(r'\[(.*?)\]', replace_match, markdown_text)

    # 生成完整的文献列表

    references_section = "\n\n## References\n\n"
    references = {num: titles_to_ids[title] for num, title in number_to_title_sorted.items()}
    for idx, title in number_to_title_sorted.items():
        t = title.replace('\n','')
        references_section += f"[{idx}] {t}\n\n"

    return updated_text + references_section, references