ai_qback/gpt/get_article2.py

# -*- coding: utf-8 -*-

import json
from collections import defaultdict
from random import randint, shuffle, sample

import httpx
import requests
from fastapi import BackgroundTasks
from openpyxl import load_workbook
from tenacity import retry, stop_after_attempt, wait_fixed

from common.common_data import all_exchange_words
from common.split_text import split_text_to_word, get_article_words_count
from gpt.chatgpt import get_article_gpt_pydantic
from gpt.gpt_check import CheckArticleResult
from tools.loglog import logger, log_err_e
from tools.new_mysql import MySQLUploader


def get_article_difficulty(article) -> int:
    """获取文章的难度值"""
    url = "http://qbank.yunzhixue.cn/api/article/analysis"

    data = {"body": article, "question": ""}
    try:
        response = requests.post(url, json=data)
    except Exception as e:
        log_err_e(e, msg="获取文章难度值；")
        return 0

    if response.status_code == 200:
        difficult_value = response.json()['data']['difficult']
        return difficult_value
    else:
        logger.error(f"错误状态码{response.status_code}")


def find_interval(number) -> int:
    """
    判断一个数字属于哪个难度等级区间。31级是例外情况，需要排查

    :param number: 要检查的数字。
    :return: 返回包含该数字的区间，如果没有找到，则返回 None。
    """
    intervals = [(1, 200), (201, 250), (251, 300), (301, 350), (351, 400), (401, 450), (451, 550), (551, 650), (651, 750), (751, 850),
                 (851, 950),
                 (951, 1100),
                 (1101, 1250), (1251, 1400), (1401, 1550), (1551, 1700), (1701, 1900), (1901, 2100), (2101, 2300), (2301, 2600),
                 (2601, 2900),
                 (2901, 3200),
                 (3201, 3500), (3501, 3900), (3901, 4300), (4301, 4700), (4701, 5100), (5101, 5500), (5501, 5900), (5901, 6500),
                 (6501, 99999)]
    for index, (start, end) in enumerate(intervals, start=1):
        if start <= number <= end:
            return index
    logger.error(f"文章难度判断不对:{number}")
    return 0


def merge_and_split(list1, list2):
    combined = list1 + list2
    import random
    random.shuffle(combined)

    two_thirds = []
    one_third = []

    total_length = len(combined)
    if total_length > 15:
        two_thirds = combined[:15]
        one_third = combined[15:]
    else:
        two_thirds = combined
        one_third = []

    return two_thirds, one_third


class GetArticle:
    def __init__(self):
        self.m = MySQLUploader()

        self.callback_url_dict = defaultdict(str)
        self.real_ip_dict = defaultdict(str)
        self.demo_name = defaultdict(str)

        self.article_result = {}

        self.punctuation = [",", ".", "!", "?", ":", ";", '"', "–", "_", "-", "...", "......"]
        all_exchange_words.update(self.punctuation)

        self.exchange_data: dict[str, list] = {}
        self.read_spring_bamboo_exchange_table()

    def read_spring_bamboo_exchange_table(self):
        """变形是键，原型是值"""
        wb = load_workbook(r"data/春笋单词对照变形.xlsx", read_only=True, data_only=True)
        ws = wb.active
        for row in ws.values:
            prototype = row[0]
            exchange = row[1]
            if prototype not in self.exchange_data:
                self.exchange_data[prototype] = [exchange]
            else:
                self.exchange_data[prototype].append(exchange)
        wb.close()

    def parser_insert_to_mysql(self, resp_result):
        try:
            for single_article in resp_result['articles']:
                article = single_article['body']
                article_json = json.dumps(single_article)
                difficult_value = find_interval(get_article_difficulty(article))
                if not difficult_value:
                    logger.error("文章难度等级为0；")
                sql = "INSERT INTO spring_bamboo_article (article_json,difficult_level) VALUES (%s,%s)"
                self.m.execute_(sql, (article_json, difficult_value))
        except Exception as e:

            logger.error(f"插入数据库时发生错误: {str(e)}")

    def submit_task(self, real_ip: str, core_words: list, take_count: int,
                    demo_name: str, reading_level: int, article_length: int, exercise_id: int,
                    background_tasks: BackgroundTasks):
        """
        core_words: 词义数据组
        take_count: 取文章数量 （int类型，正常是2篇，最大8篇）
        demo_name: 项目名称
        reading_level:阅读等级
        article_length:文章长度
        exercise_id:学案id
        background_tasks: FastAPI的后台任务管理器
        """
        task_id = randint(10000000, 99999999)
        logger.info(f"reading-comprehension 生成文章id。学案id:{exercise_id},task_id:{task_id}")

        try:
            self.real_ip_dict[task_id] = real_ip
            self.demo_name[task_id] = demo_name

            resp_result = self.run_task(core_words, task_id, exercise_id, take_count, reading_level, article_length)

            background_tasks.add_task(self.parser_insert_to_mysql, resp_result)

            logger.success(f"reading-comprehension 文章2任务完成。学案id:{exercise_id},taskid:{task_id}")
            return resp_result
        except Exception as e:
            err_msg = f"GetArticle提交任务失败{type(e).__name__},{e}"
            log_err_e(e, msg="GetArticle提交任务失败;")
            return err_msg
        finally:

            self.real_ip_dict.pop(task_id, None)
            self.demo_name.pop(task_id, None)

    def __parse_gpt_resp(self, gpt_resp: dict, core_words: list):
        return_json = {"articles": []}
        for choice in gpt_resp["choices"]:
            single_article_dict = json.loads(choice["message"]["content"])

            allWordAmount = 0

            articleWordAmount = get_article_words_count(single_article_dict["englishArticle"])
            allWordAmount += articleWordAmount

            for i in single_article_dict["questions"]:
                count_trunk = get_article_words_count(i["trunk"])
                count_candidates = sum([get_article_words_count(ii["text"]) for ii in i["candidates"]])
                allWordAmount += count_trunk
                allWordAmount += count_candidates

            usedMeanIds: list = single_article_dict['usedMeanIds']

            article_words = split_text_to_word(single_article_dict['englishArticle'])

            for i in core_words:
                meaning_id = i.get('meaning_id', 0)
                if not meaning_id:
                    continue
                word = i["spell"]
                if meaning_id not in usedMeanIds and word in self.exchange_data:
                    words_exchanges_list = self.exchange_data[word]
                    for exchange_word in words_exchanges_list:
                        if exchange_word in article_words:
                            usedMeanIds.append(meaning_id)
                            break

            single_article_dict["body"] = single_article_dict.pop("englishArticle")
            single_article_dict["chinese"] = single_article_dict.pop("chineseArticle")

            for q in single_article_dict['questions']:
                data = q['candidates']
                shuffled_candidates = sample(data, len(data))

                labels = ['A', 'B', 'C', 'D']
                for index, candidate in enumerate(shuffled_candidates):
                    candidate['label'] = labels[index]
                q['candidates'] = shuffled_candidates

            return_json['articles'].append({**single_article_dict, "allWordAmount": allWordAmount, "articleWordAmount": articleWordAmount})

        return return_json

    @retry(stop=stop_after_attempt(3), wait=wait_fixed(2), reraise=True)
    def get_article(self, core_words: list, task_id: int, exercise_id: int, reading_level, article_length, n) -> dict:

        if not article_length:
            if 0 < reading_level <= 10:
                article_length = 50 + 10 * reading_level
            elif 10 < reading_level <= 20:
                article_length = 150 + 30 * (reading_level - 10)
            else:
                article_length = 450 + 20 * (reading_level - 20)

        for index, (start, end) in enumerate([(1, 8), (9, 16), (17, 24), (24, 30)], start=1):
            if start <= reading_level <= end:
                difficulty_control_stage = index
                break
        else:
            difficulty_control_stage = 2

        diffculty_control = {
            1: {"grade": "小学", "desc_difficulty": "最简单最容易没有难度", "paragraph_count": "1-2",
                "desc2": "文章整体非常简洁，通俗易懂，适合初学者，刚入门，单词全是最常见的，语句通顺即可。",
                "choice_desc": "选择题难度尽可能简单，参考中国小学生水平"},
            2: {"grade": "初中", "desc_difficulty": "简单、常见、难度低", "paragraph_count": "2-3",
                "desc2": "文章整体难度适中，大约和中国初中生，中国CET-3，雅思4分这样的难度标准。",
                "choice_desc": "选择题难度适中，但是不要所有选择题让其直接在文中找到答案，参考中国初中生水平，中考标准。"},
            3: {"grade": "初中", "desc_difficulty": "简单、常见、难度低", "paragraph_count": "2-3",
                "desc2": "文章整体难度适中，大约和中国初中生，中国CET-3，雅思4分这样的难度标准。",
                "choice_desc": "选择题难度适中，但是不要所有选择题让其直接在文中找到答案，参考中国初中生水平，中考标准。"},
            4: {"grade": "高中", "desc_difficulty": "常见、高中难度的", "paragraph_count": "3-5",
                "desc2": "文章整体难度适中，大约和中国的高中生，中国CET-6，雅思6分这样的难度标准。",
                "choice_desc": "选择题难度偏难，要有迷惑性混淆性，答案不要出现直接在文中，4个选项要学生推理或逻辑判断，参考中国高中生水平，高考标准。"}
        }

        grade = diffculty_control[difficulty_control_stage]["grade"]
        select_diffculty = diffculty_control[difficulty_control_stage]["desc_difficulty"]
        select_paragraph_count = diffculty_control[difficulty_control_stage]["paragraph_count"]
        desc2 = diffculty_control[difficulty_control_stage]["desc2"]
        choice_desc = diffculty_control[difficulty_control_stage]["choice_desc"]

        shuffle(core_words)
        core_words_meaning_str = "; ".join([f"[{i['meaning_id']}  {i['spell']} {i['meaning']}]" for i in core_words])

        no_escape_code = r"\\n\\n"

        sys_prompt = "你是一个专业的英语老师，擅长根据用户提供的词汇生成对应的英语文章和中文翻译和4个配套选择题。"

        q = f"""下面我会为你提供一组数据，[单词组]（里面包含词义id，英语单词，中文词义），请根据这些单词的中文词义，\
生成一篇带中文翻译的考场英语文章，英语文章和中文翻译要有[标题]。特别注意这个单词有多个词义时，生成的英语文章一定要用提供的中文词义，例如我提供单词[change 零钱]，就不要使用[变化]的词义。

要求:
1.必须用提供的这个词义的单词，其他单词使用{select_diffculty}的单词。{desc2}{choice_desc}
2.优先保证文章语句通顺，意思不要太生硬。不要为了使用特定的单词，造成文章语义前后不搭，允许不使用个别词义。
3.文章中使用提供单词，一定要和提供单词的中文词义匹配，尤其是一词多义时，务必使用提供单词的词义。必须要用提供单词的词义。如果用到的词义与提供单词词义不一致，请不要使用这个单词。
4.生成的文章要求{article_length}词左右，可以用{no_escape_code}字符分段，一般{select_paragraph_count}个段落左右。第一段是文章标题。不需要markdown格式。
5.允许不使用[单词组]的个别单词，优先保证文章整体意思通顺连贯和故事完整。

提供[单词组]：{core_words_meaning_str};
"""

        try:
            real_ip = self.real_ip_dict[task_id]
            demo_name = self.demo_name[task_id]

            gpt_resp = get_article_gpt_pydantic(q, temperature=1.2, real_ip=real_ip, demo_name=demo_name, model='gpt-4.1',
                                                check_fucn=CheckArticleResult.get_article_1, max_tokens=15000,
                                                sys_prompt=sys_prompt, n=n, task_id=task_id, exercise_id=exercise_id)

            multi_articles_dict = self.__parse_gpt_resp(gpt_resp=gpt_resp, core_words=core_words)
            return multi_articles_dict

        except httpx.HTTPError as e:
            logger.error(f"HTTP请求错误: {str(e)}")
            raise
        except json.JSONDecodeError as e:
            logger.error(f"JSON解析错误: {str(e)}")
            raise
        except Exception as e:
            log_err_e(e, f"gpt生成文章回复其他错误.")
            raise

    def run_get_article_task(self, core_words, task_id, exercise_id, take_count, reading_level, article_length) -> dict:
        """
        :param core_words: 核心单词数据，优先级1;可能为空
        :param task_id: 任务id
        :param take_count: 文章数量
        :param reading_level:阅读等级
        :param article_length:文章长度
        :return:
        """
        try:
            return_json = self.get_article(core_words, task_id, exercise_id, reading_level, article_length, n=take_count)
            return return_json
        except Exception as e:
            logger.error(f"运行文章任务时发生错误: {str(e)}")
            raise

    def run_task(self, core_words, task_id, exercise_id, take_count, reading_level, article_length):
        try:
            outside_json = self.run_get_article_task(core_words, task_id, exercise_id, take_count, reading_level, article_length)
            return outside_json
        except Exception as e:
            log_err_e(e, msg="外层总任务捕获错误")

    def cleanup(self):
        """清理所有资源"""
        pass