EnglishPal/app/Article.py

141 lines
5.6 KiB
Python
Raw Normal View History

from WordFreq import WordFreq
from wordfreqCMD import youdao_link, sort_in_descending_order
import pickle_idea, pickle_idea2
import os
import random, glob
import hashlib
from datetime import datetime
from flask import Flask, request, redirect, render_template, url_for, session, abort, flash, get_flashed_messages
from difficulty import get_difficulty_level_for_user, text_difficulty_level, user_difficulty_level
2024-07-04 10:06:44 +08:00
from model.article import get_all_articles, get_article_by_id, get_number_of_articles
import logging
2024-07-04 10:06:44 +08:00
path_prefix = './'
db_path_prefix = './db/' # comment this line in deployment
def total_number_of_essays():
2024-07-04 10:06:44 +08:00
return get_number_of_articles()
def get_article_title(s):
return s.split('\n')[0]
def get_article_body(s):
lst = s.split('\n')
lst.pop(0) # remove the first line
return '\n'.join(lst)
def get_today_article(user_word_list, visited_articles):
if visited_articles is None:
visited_articles = {
"index" : 0, # 为 article_ids 的索引
"article_ids": [] # 之前显示文章的id列表越后越新
}
if visited_articles["index"] > len(visited_articles["article_ids"])-1: # 生成新的文章,因此查找所有的文章
2024-07-04 10:06:44 +08:00
result = get_all_articles()
else: # 生成阅读过的文章,因此查询指定 article_id 的文章
if visited_articles["article_ids"][visited_articles["index"]] == 'null': # 可能因为直接刷新页面导致直接去查询了'null',因此当刷新的页面的时候,需要直接进行“上一篇”操作
visited_articles["index"] -= 1
visited_articles["article_ids"].pop()
2024-07-04 10:06:44 +08:00
article_id = visited_articles["article_ids"][visited_articles["index"]]
result = get_article_by_id(article_id)
random.shuffle(result)
# Choose article according to reader's level
2024-07-04 10:06:44 +08:00
logging.debug('* get_today_article(): start d1 = ... ')
d1 = load_freq_history(user_word_list)
d2 = load_freq_history(path_prefix + 'static/words_and_tests.p')
2024-07-04 10:06:44 +08:00
logging.debug(' ... get_today_article(): get_difficulty_level_for_user() start')
d3 = get_difficulty_level_for_user(d1, d2)
2024-07-04 10:06:44 +08:00
logging.debug(' ... get_today_article(): done')
d = None
result_of_generate_article = "not found"
2024-07-04 10:06:44 +08:00
d_user = load_freq_history(user_word_list)
2024-07-04 10:06:44 +08:00
logging.debug('* get_today_article(): user_difficulty_level() start')
2022-01-27 17:01:03 +08:00
user_level = user_difficulty_level(d_user, d3) # more consideration as user's behaviour is dynamic. Time factor should be considered.
2024-07-04 10:06:44 +08:00
logging.debug('* get_today_article(): done')
text_level = 0
if visited_articles["index"] > len(visited_articles["article_ids"])-1: # 生成新的文章
amount_of_visited_articles = len(visited_articles["article_ids"])
amount_of_existing_articles = result.__len__()
if amount_of_visited_articles == amount_of_existing_articles: # 如果当前阅读过的文章的数量 == 存在的文章的数量,即所有的书本都阅读过了
result_of_generate_article = "had read all articles"
else:
for k in range(3): # 最多尝试3次
for reading in result:
text_level = text_difficulty_level(reading['text'], d3)
factor = random.gauss(0.8, 0.1) # a number drawn from Gaussian distribution with a mean of 0.8 and a stand deviation of 1
if reading['article_id'] not in visited_articles["article_ids"] and within_range(text_level, user_level, (8.0 - user_level) * factor): # 新的文章之前没有出现过且符合一定范围的水平
d = reading
visited_articles["article_ids"].append(d['article_id']) # 列表添加新的文章id下面进行
result_of_generate_article = "found"
break
if result_of_generate_article == "found": # 用于成功找到文章后及时退出外层循环
break
if result_of_generate_article != "found": # 阅读完所有文章或者循环3次没有找到适合的文章则放入空“null”
visited_articles["article_ids"].append('null')
else: # 生成已经阅读过的文章
d = random.choice(result)
text_level = text_difficulty_level(d['text'], d3)
result_of_generate_article = "found"
today_article = None
if d:
today_article = {
2024-07-04 10:06:44 +08:00
"user_level": '%4.1f' % user_level,
"text_level": '%4.1f' % text_level,
"date": d['date'],
"article_title": get_article_title(d['text']),
"article_body": get_article_body(d['text']),
"source": d["source"],
"question": get_question_part(d['question']),
"answer": get_answer_part(d['question'])
}
return visited_articles, today_article, result_of_generate_article
def load_freq_history(path):
d = {}
if os.path.exists(path):
d = pickle_idea.load_record(path)
return d
def within_range(x, y, r):
return x > y and abs(x - y) <= r
def get_question_part(s):
s = s.strip()
result = []
flag = 0
for line in s.split('\n'):
line = line.strip()
if line == 'QUESTION':
result.append(line)
flag = 1
elif line == 'ANSWER':
flag = 0
elif flag == 1:
result.append(line)
return '\n'.join(result)
def get_answer_part(s):
s = s.strip()
result = []
flag = 0
for line in s.split('\n'):
line = line.strip()
if line == 'ANSWER':
flag = 1
elif flag == 1:
result.append(line)
2023-03-30 16:10:22 +08:00
return '\n'.join(result)