论文

初始化文件。

ArxivReader

基类：BaseReader

Arxiv 读取器。

获取搜索查询，返回 Arxiv 上最相关的科学论文 Document 列表。

源代码位于 llama-index-integrations/readers/llama-index-readers-papers/llama_index/readers/papers/arxiv/base.py

class ArxivReader(BaseReader):
    """
    Arxiv Reader.

    Gets a search query, return a list of Documents of the top corresponding scientific papers on Arxiv.
    """

    def __init__(
        self,
    ) -> None:
        """Initialize with parameters."""
        super().__init__()

    def _hacky_hash(self, some_string):
        return hashlib.md5(some_string.encode("utf-8")).hexdigest()

    def load_data(
        self,
        search_query: str,
        papers_dir: Optional[str] = ".papers",
        max_results: Optional[int] = 10,
    ) -> List[Document]:
        """
        Search for a topic on Arxiv, download the PDFs of the top results locally, then read them.

        Args:
            search_query (str): A topic to search for (e.g. "Artificial Intelligence").
            papers_dir (Optional[str]): Locally directory to store the papers
            max_results (Optional[int]): Maximum number of papers to fetch.

        Returns:
            List[Document]: A list of Document objects.

        """
        import arxiv

        arxiv_search = arxiv.Search(
            query=search_query,
            id_list=[],
            max_results=max_results,
            sort_by=arxiv.SortCriterion.Relevance,
        )
        search_results = list(arxiv_search.results())
        logging.debug(f"> Successfully fetched {len(search_results)} paperes")

        if not os.path.exists(papers_dir):
            os.makedirs(papers_dir)

        paper_lookup = {}
        for paper in search_results:
            # Hash filename to avoid bad characters in file path
            hashed_name = self._hacky_hash(f"{paper.title}{paper.entry_id}")
            filename = f"{hashed_name}.pdf"
            paper_lookup[filename] = {
                "Title of this paper": paper.title,
                "Authors": (", ").join([a.name for a in paper.authors]),
                "Date published": paper.published.strftime("%m/%d/%Y"),
                "URL": paper.entry_id,
                # "summary": paper.summary
            }
            paper.download_pdf(dirpath=papers_dir, filename=filename)
            logging.debug(f"> Downloading {filename}...")

        def get_paper_metadata(filename):
            return paper_lookup[os.path.basename(filename)]

        arxiv_documents = SimpleDirectoryReader(
            papers_dir,
            file_metadata=get_paper_metadata,
            exclude_hidden=False,  # default directory is hidden ".papers"
        ).load_data()
        # Include extra documents containing the abstracts
        abstract_documents = []
        for paper in search_results:
            d = (
                f"The following is a summary of the paper: {paper.title}\n\nSummary:"
                f" {paper.summary}"
            )
            abstract_documents.append(Document(text=d))

        # Delete downloaded papers
        try:
            for f in os.listdir(papers_dir):
                os.remove(os.path.join(papers_dir, f))
                logging.debug(f"> Deleted file: {f}")
            os.rmdir(papers_dir)
            logging.debug(f"> Deleted directory: {papers_dir}")
        except OSError:
            print("Unable to delete files or directory")

        return arxiv_documents + abstract_documents

    def load_papers_and_abstracts(
        self,
        search_query: str,
        papers_dir: Optional[str] = ".papers",
        max_results: Optional[int] = 10,
    ) -> Tuple[List[Document], List[Document]]:
        """
        Search for a topic on Arxiv, download the PDFs of the top results locally, then read them.

        Args:
            search_query (str): A topic to search for (e.g. "Artificial Intelligence").
            papers_dir (Optional[str]): Locally directory to store the papers
            max_results (Optional[int]): Maximum number of papers to fetch.

        Returns:
            List[Document]: A list of Document objects representing the papers themselves
            List[Document]: A list of Document objects representing abstracts only

        """
        import arxiv

        arxiv_search = arxiv.Search(
            query=search_query,
            id_list=[],
            max_results=max_results,
            sort_by=arxiv.SortCriterion.Relevance,
        )
        search_results = list(arxiv_search.results())
        logging.debug(f"> Successfully fetched {len(search_results)} paperes")

        if not os.path.exists(papers_dir):
            os.makedirs(papers_dir)

        paper_lookup = {}
        for paper in search_results:
            # Hash filename to avoid bad characters in file path
            hashed_name = self._hacky_hash(f"{paper.title}{paper.entry_id}")
            filename = f"{hashed_name}.pdf"
            paper_lookup[filename] = {
                "Title of this paper": paper.title,
                "Authors": (", ").join([a.name for a in paper.authors]),
                "Date published": paper.published.strftime("%m/%d/%Y"),
                "URL": paper.entry_id,
                # "summary": paper.summary
            }
            paper.download_pdf(dirpath=papers_dir, filename=filename)
            logging.debug(f"> Downloading {filename}...")

        def get_paper_metadata(filename):
            return paper_lookup[os.path.basename(filename)]

        arxiv_documents = SimpleDirectoryReader(
            papers_dir,
            file_metadata=get_paper_metadata,
            exclude_hidden=False,  # default directory is hidden ".papers"
        ).load_data()
        # Include extra documents containing the abstracts
        abstract_documents = []
        for paper in search_results:
            d = (
                f"The following is a summary of the paper: {paper.title}\n\nSummary:"
                f" {paper.summary}"
            )
            abstract_documents.append(Document(text=d))

        # Delete downloaded papers
        try:
            for f in os.listdir(papers_dir):
                os.remove(os.path.join(papers_dir, f))
                logging.debug(f"> Deleted file: {f}")
            os.rmdir(papers_dir)
            logging.debug(f"> Deleted directory: {papers_dir}")
        except OSError:
            print("Unable to delete files or directory")

        return arxiv_documents, abstract_documents

load_data

load_data(search_query: str, papers_dir: Optional[str] = '.papers', max_results: Optional[int] = 10) -> List[Document]

在 Arxiv 上搜索主题，将最相关结果的 PDF 下载到本地，然后读取。

参数

名称	类型	描述	默认
search_query	str	要搜索的主题 (例如 "Artificial Intelligence")。	必需
papers_dir	可选[str]	存储论文的本地目录	'.papers'
max_results	可选[int]	最大抓取论文数量。	10

返回

类型	描述
列表[Document]	List[Document]: Document 对象列表。

源代码位于 llama-index-integrations/readers/llama-index-readers-papers/llama_index/readers/papers/arxiv/base.py

def load_data(
    self,
    search_query: str,
    papers_dir: Optional[str] = ".papers",
    max_results: Optional[int] = 10,
) -> List[Document]:
    """
    Search for a topic on Arxiv, download the PDFs of the top results locally, then read them.

    Args:
        search_query (str): A topic to search for (e.g. "Artificial Intelligence").
        papers_dir (Optional[str]): Locally directory to store the papers
        max_results (Optional[int]): Maximum number of papers to fetch.

    Returns:
        List[Document]: A list of Document objects.

    """
    import arxiv

    arxiv_search = arxiv.Search(
        query=search_query,
        id_list=[],
        max_results=max_results,
        sort_by=arxiv.SortCriterion.Relevance,
    )
    search_results = list(arxiv_search.results())
    logging.debug(f"> Successfully fetched {len(search_results)} paperes")

    if not os.path.exists(papers_dir):
        os.makedirs(papers_dir)

    paper_lookup = {}
    for paper in search_results:
        # Hash filename to avoid bad characters in file path
        hashed_name = self._hacky_hash(f"{paper.title}{paper.entry_id}")
        filename = f"{hashed_name}.pdf"
        paper_lookup[filename] = {
            "Title of this paper": paper.title,
            "Authors": (", ").join([a.name for a in paper.authors]),
            "Date published": paper.published.strftime("%m/%d/%Y"),
            "URL": paper.entry_id,
            # "summary": paper.summary
        }
        paper.download_pdf(dirpath=papers_dir, filename=filename)
        logging.debug(f"> Downloading {filename}...")

    def get_paper_metadata(filename):
        return paper_lookup[os.path.basename(filename)]

    arxiv_documents = SimpleDirectoryReader(
        papers_dir,
        file_metadata=get_paper_metadata,
        exclude_hidden=False,  # default directory is hidden ".papers"
    ).load_data()
    # Include extra documents containing the abstracts
    abstract_documents = []
    for paper in search_results:
        d = (
            f"The following is a summary of the paper: {paper.title}\n\nSummary:"
            f" {paper.summary}"
        )
        abstract_documents.append(Document(text=d))

    # Delete downloaded papers
    try:
        for f in os.listdir(papers_dir):
            os.remove(os.path.join(papers_dir, f))
            logging.debug(f"> Deleted file: {f}")
        os.rmdir(papers_dir)
        logging.debug(f"> Deleted directory: {papers_dir}")
    except OSError:
        print("Unable to delete files or directory")

    return arxiv_documents + abstract_documents

load_papers_and_abstracts

load_papers_and_abstracts(search_query: str, papers_dir: Optional[str] = '.papers', max_results: Optional[int] = 10) -> Tuple[List[Document], List[Document]]

在 Arxiv 上搜索主题，将最相关结果的 PDF 下载到本地，然后读取。

参数

名称	类型	描述	默认
search_query	str	要搜索的主题 (例如 "Artificial Intelligence")。	必需
papers_dir	可选[str]	存储论文的本地目录	'.papers'
max_results	可选[int]	最大抓取论文数量。	10

返回

类型	描述
列表[Document]	List[Document]: 代表论文本身的 Document 对象列表
列表[Document]	List[Document]: 仅代表摘要的 Document 对象列表

源代码位于 llama-index-integrations/readers/llama-index-readers-papers/llama_index/readers/papers/arxiv/base.py

def load_papers_and_abstracts(
    self,
    search_query: str,
    papers_dir: Optional[str] = ".papers",
    max_results: Optional[int] = 10,
) -> Tuple[List[Document], List[Document]]:
    """
    Search for a topic on Arxiv, download the PDFs of the top results locally, then read them.

    Args:
        search_query (str): A topic to search for (e.g. "Artificial Intelligence").
        papers_dir (Optional[str]): Locally directory to store the papers
        max_results (Optional[int]): Maximum number of papers to fetch.

    Returns:
        List[Document]: A list of Document objects representing the papers themselves
        List[Document]: A list of Document objects representing abstracts only

    """
    import arxiv

    arxiv_search = arxiv.Search(
        query=search_query,
        id_list=[],
        max_results=max_results,
        sort_by=arxiv.SortCriterion.Relevance,
    )
    search_results = list(arxiv_search.results())
    logging.debug(f"> Successfully fetched {len(search_results)} paperes")

    if not os.path.exists(papers_dir):
        os.makedirs(papers_dir)

    paper_lookup = {}
    for paper in search_results:
        # Hash filename to avoid bad characters in file path
        hashed_name = self._hacky_hash(f"{paper.title}{paper.entry_id}")
        filename = f"{hashed_name}.pdf"
        paper_lookup[filename] = {
            "Title of this paper": paper.title,
            "Authors": (", ").join([a.name for a in paper.authors]),
            "Date published": paper.published.strftime("%m/%d/%Y"),
            "URL": paper.entry_id,
            # "summary": paper.summary
        }
        paper.download_pdf(dirpath=papers_dir, filename=filename)
        logging.debug(f"> Downloading {filename}...")

    def get_paper_metadata(filename):
        return paper_lookup[os.path.basename(filename)]

    arxiv_documents = SimpleDirectoryReader(
        papers_dir,
        file_metadata=get_paper_metadata,
        exclude_hidden=False,  # default directory is hidden ".papers"
    ).load_data()
    # Include extra documents containing the abstracts
    abstract_documents = []
    for paper in search_results:
        d = (
            f"The following is a summary of the paper: {paper.title}\n\nSummary:"
            f" {paper.summary}"
        )
        abstract_documents.append(Document(text=d))

    # Delete downloaded papers
    try:
        for f in os.listdir(papers_dir):
            os.remove(os.path.join(papers_dir, f))
            logging.debug(f"> Deleted file: {f}")
        os.rmdir(papers_dir)
        logging.debug(f"> Deleted directory: {papers_dir}")
    except OSError:
        print("Unable to delete files or directory")

    return arxiv_documents, abstract_documents

PubmedReader

基类：BaseReader

Pubmed 读取器。

获取搜索查询，返回 Pubmed 上最相关的科学论文 Document 列表。

源代码位于 llama-index-integrations/readers/llama-index-readers-papers/llama_index/readers/papers/pubmed/base.py

class PubmedReader(BaseReader):
    """
    Pubmed Reader.

    Gets a search query, return a list of Documents of the top corresponding scientific papers on Pubmed.
    """

    def load_data_bioc(
        self,
        search_query: str,
        max_results: Optional[int] = 10,
    ) -> List[Document]:
        """
        Search for a topic on Pubmed, fetch the text of the most relevant full-length papers.
        Uses the BoiC API, which has been down a lot.

        Args:
            search_query (str): A topic to search for (e.g. "Alzheimers").
            max_results (Optional[int]): Maximum number of papers to fetch.

        Returns:
            List[Document]: A list of Document objects.

        """
        from datetime import datetime

        import requests
        from defusedxml import ElementTree as safe_xml

        pubmed_search = []
        parameters = {"tool": "tool", "email": "email", "db": "pmc"}
        parameters["term"] = search_query
        parameters["retmax"] = max_results
        resp = requests.get(
            "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
            params=parameters,
        )
        root = safe_xml.fromstring(resp.content)

        for elem in root.iter():
            if elem.tag == "Id":
                _id = elem.text
                try:
                    resp = requests.get(
                        f"https://www.ncbi.nlm.nih.gov/research/bionlp/RESTful/pmcoa.cgi/BioC_json/PMC{_id}/ascii"
                    )
                    info = resp.json()
                    title = "Pubmed Paper"
                    try:
                        title = next(
                            [
                                p["text"]
                                for p in info["documents"][0]["passages"]
                                if p["infons"]["section_type"] == "TITLE"
                            ]
                        )
                    except KeyError:
                        pass
                    pubmed_search.append(
                        {
                            "title": title,
                            "url": (
                                f"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC{_id}/"
                            ),
                            "date": info["date"],
                            "documents": info["documents"],
                        }
                    )
                except Exception:
                    print(f"Unable to parse PMC{_id} or it does not exist")

        # Then get documents from Pubmed text, which includes abstracts
        pubmed_documents = []
        for paper in pubmed_search:
            for d in paper["documents"]:
                text = "\n".join([p["text"] for p in d["passages"]])
                pubmed_documents.append(
                    Document(
                        text=text,
                        extra_info={
                            "Title of this paper": paper["title"],
                            "URL": paper["url"],
                            "Date published": datetime.strptime(
                                paper["date"], "%Y%m%d"
                            ).strftime("%m/%d/%Y"),
                        },
                    )
                )

        return pubmed_documents

    def load_data(
        self,
        search_query: str,
        max_results: Optional[int] = 10,
    ) -> List[Document]:
        """
        Search for a topic on Pubmed, fetch the text of the most relevant full-length papers.

        Args:
            search_query (str): A topic to search for (e.g. "Alzheimers").
            max_results (Optional[int]): Maximum number of papers to fetch.


        Returns:
            List[Document]: A list of Document objects.

        """
        import time

        import requests

        pubmed_search = []
        parameters = {"tool": "tool", "email": "email", "db": "pmc"}
        parameters["term"] = search_query
        parameters["retmax"] = max_results
        resp = requests.get(
            "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
            params=parameters,
        )
        root = safe_xml.fromstring(resp.content)

        for elem in root.iter():
            if elem.tag == "Id":
                _id = elem.text
                url = f"https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?id={_id}&db=pmc"
                print(url)
                try:
                    resp = requests.get(url)
                    info = safe_xml.fromstring(resp.content)

                    raw_text = ""
                    title = ""
                    journal = ""
                    for element in info.iter():
                        if element.tag == "article-title":
                            title = element.text
                        elif element.tag == "journal-title":
                            journal = element.text

                        if element.text:
                            raw_text += element.text.strip() + " "

                    pubmed_search.append(
                        {
                            "title": title,
                            "journal": journal,
                            "url": (
                                f"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC{_id}/"
                            ),
                            "text": raw_text,
                        }
                    )
                    time.sleep(1)  # API rate limits
                except Exception as e:
                    print(f"Unable to parse PMC{_id} or it does not exist:", e)

        # Then get documents from Pubmed text, which includes abstracts
        pubmed_documents = []
        for paper in pubmed_search:
            pubmed_documents.append(
                Document(
                    text=paper["text"],
                    extra_info={
                        "Title of this paper": paper["title"],
                        "Journal it was published in:": paper["journal"],
                        "URL": paper["url"],
                    },
                )
            )

        return pubmed_documents

load_data_bioc

load_data_bioc(search_query: str, max_results: Optional[int] = 10) -> List[Document]

在 Pubmed 上搜索主题，抓取最相关的全文论文文本。使用 BoiC API，该 API 经常不可用。

参数

名称	类型	描述	默认
search_query	str	要搜索的主题 (例如 "Alzheimers")。	必需
max_results	可选[int]	最大抓取论文数量。	10

返回

类型	描述
列表[Document]	List[Document]: Document 对象列表。

源代码位于 llama-index-integrations/readers/llama-index-readers-papers/llama_index/readers/papers/pubmed/base.py

def load_data_bioc(
    self,
    search_query: str,
    max_results: Optional[int] = 10,
) -> List[Document]:
    """
    Search for a topic on Pubmed, fetch the text of the most relevant full-length papers.
    Uses the BoiC API, which has been down a lot.

    Args:
        search_query (str): A topic to search for (e.g. "Alzheimers").
        max_results (Optional[int]): Maximum number of papers to fetch.

    Returns:
        List[Document]: A list of Document objects.

    """
    from datetime import datetime

    import requests
    from defusedxml import ElementTree as safe_xml

    pubmed_search = []
    parameters = {"tool": "tool", "email": "email", "db": "pmc"}
    parameters["term"] = search_query
    parameters["retmax"] = max_results
    resp = requests.get(
        "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
        params=parameters,
    )
    root = safe_xml.fromstring(resp.content)

    for elem in root.iter():
        if elem.tag == "Id":
            _id = elem.text
            try:
                resp = requests.get(
                    f"https://www.ncbi.nlm.nih.gov/research/bionlp/RESTful/pmcoa.cgi/BioC_json/PMC{_id}/ascii"
                )
                info = resp.json()
                title = "Pubmed Paper"
                try:
                    title = next(
                        [
                            p["text"]
                            for p in info["documents"][0]["passages"]
                            if p["infons"]["section_type"] == "TITLE"
                        ]
                    )
                except KeyError:
                    pass
                pubmed_search.append(
                    {
                        "title": title,
                        "url": (
                            f"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC{_id}/"
                        ),
                        "date": info["date"],
                        "documents": info["documents"],
                    }
                )
            except Exception:
                print(f"Unable to parse PMC{_id} or it does not exist")

    # Then get documents from Pubmed text, which includes abstracts
    pubmed_documents = []
    for paper in pubmed_search:
        for d in paper["documents"]:
            text = "\n".join([p["text"] for p in d["passages"]])
            pubmed_documents.append(
                Document(
                    text=text,
                    extra_info={
                        "Title of this paper": paper["title"],
                        "URL": paper["url"],
                        "Date published": datetime.strptime(
                            paper["date"], "%Y%m%d"
                        ).strftime("%m/%d/%Y"),
                    },
                )
            )

    return pubmed_documents

load_data

load_data(search_query: str, max_results: Optional[int] = 10) -> List[Document]

在 Pubmed 上搜索主题，抓取最相关的全文论文文本。

参数

名称	类型	描述	默认
search_query	str	要搜索的主题 (例如 "Alzheimers")。	必需
max_results	可选[int]	最大抓取论文数量。	10

返回

类型	描述
列表[Document]	List[Document]: Document 对象列表。

源代码位于 llama-index-integrations/readers/llama-index-readers-papers/llama_index/readers/papers/pubmed/base.py

def load_data(
    self,
    search_query: str,
    max_results: Optional[int] = 10,
) -> List[Document]:
    """
    Search for a topic on Pubmed, fetch the text of the most relevant full-length papers.

    Args:
        search_query (str): A topic to search for (e.g. "Alzheimers").
        max_results (Optional[int]): Maximum number of papers to fetch.


    Returns:
        List[Document]: A list of Document objects.

    """
    import time

    import requests

    pubmed_search = []
    parameters = {"tool": "tool", "email": "email", "db": "pmc"}
    parameters["term"] = search_query
    parameters["retmax"] = max_results
    resp = requests.get(
        "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
        params=parameters,
    )
    root = safe_xml.fromstring(resp.content)

    for elem in root.iter():
        if elem.tag == "Id":
            _id = elem.text
            url = f"https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?id={_id}&db=pmc"
            print(url)
            try:
                resp = requests.get(url)
                info = safe_xml.fromstring(resp.content)

                raw_text = ""
                title = ""
                journal = ""
                for element in info.iter():
                    if element.tag == "article-title":
                        title = element.text
                    elif element.tag == "journal-title":
                        journal = element.text

                    if element.text:
                        raw_text += element.text.strip() + " "

                pubmed_search.append(
                    {
                        "title": title,
                        "journal": journal,
                        "url": (
                            f"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC{_id}/"
                        ),
                        "text": raw_text,
                    }
                )
                time.sleep(1)  # API rate limits
            except Exception as e:
                print(f"Unable to parse PMC{_id} or it does not exist:", e)

    # Then get documents from Pubmed text, which includes abstracts
    pubmed_documents = []
    for paper in pubmed_search:
        pubmed_documents.append(
            Document(
                text=paper["text"],
                extra_info={
                    "Title of this paper": paper["title"],
                    "Journal it was published in:": paper["journal"],
                    "URL": paper["url"],
                },
            )
        )

    return pubmed_documents