# Documentation - Class name: LLMHTMLTagReader - Category: SALT/Language Toolkit/Readers - Output node: False LLMHTMLTagReader节点旨在从指定文件中读取并解析HTML标签,将其转化为结构化的文档格式。它利用BeautifulSoup来解析HTML内容,重点关注特定的标签和属性以提取相关信息,同时通过可选参数允许自定义处理过程。 # Input types ## Required - path - 指定要读取的HTML文件的文件路径。这是一个关键参数,因为它决定了要处理的HTML内容的来源。 - Comfy dtype: STRING - Python dtype: str ## Optional - tag - 定义在解析过程中要重点关注的特定HTML标签。这允许从HTML文件中有针对性地提取信息。 - Comfy dtype: STRING - Python dtype: str - ignore_no_id - 一个布尔标志,设置后会指示阅读器忽略没有ID属性的HTML元素。这对于过滤掉不必要的元素很有用。 - Comfy dtype: COMBO[BOOLEAN] - Python dtype: bool - extra_info - 允许以字符串形式包含额外的自定义信息,可用于进一步自定义解析行为。 - Comfy dtype: STRING - Python dtype: str # Output types - documents - 结构化文档格式的输出,代表了经过解析和解释的HTML内容。 - Comfy dtype: DOCUMENT - Python dtype: tuple ## Usage tips - Infra type: `CPU` - Common nodes: unknown ## Source code ```python class LLMHTMLTagReader(HTMLTagReader): """ @NOTE: Reads HTML tags into a llama_index Document @Source: https://github.com/run-llama/llama_index/blob/main/llama-index-integrations/readers/llama-index-readers-file/llama_index/readers/file/html/base.py @Documentation: https://docs.llamaindex.ai/en/latest/api_reference/readers/file/#llama_index.readers.file.HTMLTagReader @Imports: from bs4 import BeautifulSoup """ def __init__(self): super().__init__() @classmethod def INPUT_TYPES(cls): return { "required": { "path": ("STRING", {"default": ""}), }, "optional": { "tag": ("STRING", {"default":"section"}), "ignore_no_id": ([False, True],), "extra_info": ("STRING", {"multiline": True, "dynamicPrompts": False, "default": "{}"}), } } RETURN_TYPES = ("DOCUMENT", ) RETURN_NAMES = ("documents",) FUNCTION = "execute" CATEGORY = f"{MENU_NAME}/{SUB_MENU_NAME}/Readers" def execute(self, path:str, tag:str="section", ignore_no_id:bool=False, extra_info:str="{}"): get_full_path(1, path) if not os.path.exists(path): raise FileNotFoundError(f"No file available at: {path}") path = Path(path) self._tag = tag self._ignore_no_id = ignore_no_id extra_info = read_extra_info(extra_info) data = self.load_data(path, extra_info) return (data, ) ```