This commit is contained in:
机器人总动员 2021-05-05 16:21:06 +08:00
parent bec8b109cf
commit 498463b00e
2 changed files with 110 additions and 109 deletions

View File

@ -2,11 +2,11 @@ from telegram.ext import Updater
from config import TOKEN
from func import start_handler, chat_content_handler, check_schedule
import schedule
import task
from task import do_task
import threading
# schedule.every().day.at('16:20').do(task).tag('task')
schedule.every(1).minutes.do(task).tag('task')
schedule.every(1).minutes.do(do_task).tag('task')
threading.Thread(target=check_schedule).start()

215
task.py
View File

@ -11,120 +11,121 @@ from config import TOKEN
bot = telegram.Bot(token=TOKEN)
r = connector.get_connection()
key_list = r.keys()
group_list = []
for i in key_list:
if "chat_content" in i:
group_list.append(i[:i.find("_")])
# print(group_list)
def do_task():
r = connector.get_connection()
key_list = r.keys()
group_list = []
for i in key_list:
if "chat_content" in i:
group_list.append(i[:i.find("_")])
# print(group_list)
mk = imageio.imread("/root/word_cloud_bot/circle.png")
# 构建并配置词云对象w注意要加scale参数提高清晰度
w = wordcloud.WordCloud(width=800,
height=800,
background_color='white',
font_path='/root/word_cloud_bot/font.ttf',
mask=mk,
scale=5)
mk = imageio.imread("/root/word_cloud_bot/circle.png")
# 构建并配置词云对象w注意要加scale参数提高清晰度
w = wordcloud.WordCloud(width=800,
height=800,
background_color='white',
font_path='/root/word_cloud_bot/font.ttf',
mask=mk,
scale=5)
for group in group_list:
try:
print("当前处理的群组:" + str(group))
start_time = float(time.time())
# 生成词云图片
jieba.enable_paddle() # 启动paddle模式。 0.40版之后开始支持,早期版本不支持
words = pseg.cut(r.get("{}_chat_content".format(group)), use_paddle=True) # paddle模式
word_list = []
for word, flag in words:
# print(word + "\t" + flag)
if flag in ["n", "nr", "nz", "PER", "f", "ns", "LOC", "s", "nt", "ORG", "nw"]:
# 判断该词是否有效,不为空格
if re.match(r"^\s+?$", word) is None:
word_list.append(word)
# print(word_list)
for group in group_list:
try:
print("当前处理的群组:" + str(group))
start_time = float(time.time())
# 生成词云图片
jieba.enable_paddle() # 启动paddle模式。 0.40版之后开始支持,早期版本不支持
words = pseg.cut(r.get("{}_chat_content".format(group)), use_paddle=True) # paddle模式
word_list = []
for word, flag in words:
# print(word + "\t" + flag)
if flag in ["n", "nr", "nz", "PER", "f", "ns", "LOC", "s", "nt", "ORG", "nw"]:
# 判断该词是否有效,不为空格
if re.match(r"^\s+?$", word) is None:
word_list.append(word)
# print(word_list)
# 分析高频词
word_amount = {}
# print(word_amount)
for word in word_list:
# 判断该词是否之前已经出现
if word_amount.get(word) is not None:
word_amount[word] = word_amount.get(word) + 1
else:
word_amount[word] = 1
# print(word_amount)
word_amount = sorted(word_amount.items(), key=lambda kv: (kv[1]), reverse=True)
# print("排序后的热词:" + str(word_amount))
hot_word_string = ""
for i in range(min(5, len(word_amount))):
hot_word_string += "\t\t\t\t\t\t\t\t" + "`" + str(word_amount[i][0]) + "`" + ": " + str(
word_amount[i][1]) + "\n"
# print(hot_word_string)
# 获取消息总数
total_message_amount = r.get("{}_total_message_amount".format(group))
# 分析高频词
word_amount = {}
# print(word_amount)
for word in word_list:
# 判断该词是否之前已经出现
if word_amount.get(word) is not None:
word_amount[word] = word_amount.get(word) + 1
else:
word_amount[word] = 1
# print(word_amount)
word_amount = sorted(word_amount.items(), key=lambda kv: (kv[1]), reverse=True)
# print("排序后的热词:" + str(word_amount))
hot_word_string = ""
for i in range(min(5, len(word_amount))):
hot_word_string += "\t\t\t\t\t\t\t\t" + "`" + str(word_amount[i][0]) + "`" + ": " + str(
word_amount[i][1]) + "\n"
# print(hot_word_string)
# 获取消息总数
total_message_amount = r.get("{}_total_message_amount".format(group))
# print("总发言数: " + total_message_amount)
# print("总发言数: " + total_message_amount)
# 获取发言用户数
user_amount = len(r.hkeys("{}_user_message_amount".format(group)))
# 获取所有用户发言数字典
user_message_amount = r.hgetall("-1001403536948_user_message_amount")
user_message_amount = sorted(user_message_amount.items(), key=lambda kv: (kv[1]), reverse=True)
# print("排序后的用户:" + str(user_message_amount))
top_5_user = ""
for i in range(min(5, len(user_message_amount))):
top_5_user += "\t\t\t\t\t\t\t\t" + "🎖`" + str(user_message_amount[i][0]) + "`" + " 贡献: " + str(
user_message_amount[i][1]) + "\n"
# print(top_5_user)
string = " ".join(word_list)
# 将string变量传入w的generate()方法,给词云输入文字
w.generate(string)
# 将词云图片导出到当前文件夹
w.to_file('{}_chat_word_cloud.png'.format(group))
bot.send_message(
chat_id=group,
text="🎤 今日话题榜 🎤\n"
"📅 {}\n"
"⏱ 截至今天{}\n"
"🗣️ 本群{}位朋友共产生{}条发言\n"
"🤹‍ 大家今天讨论最多的是:\n\n"
"{}\n"
"看下有没有你感兴趣的话题? 👏".format(
time.strftime("%Y年%m月%d", time.localtime()),
time.strftime("%H:%M", time.localtime()),
user_amount,
total_message_amount,
hot_word_string),
parse_mode="Markdown"
)
# 获取发言用户数
user_amount = len(r.hkeys("{}_user_message_amount".format(group)))
# 获取所有用户发言数字典
user_message_amount = r.hgetall("-1001403536948_user_message_amount")
user_message_amount = sorted(user_message_amount.items(), key=lambda kv: (kv[1]), reverse=True)
# print("排序后的用户:" + str(user_message_amount))
top_5_user = ""
for i in range(min(5, len(user_message_amount))):
top_5_user += "\t\t\t\t\t\t\t\t" + "🎖`" + str(user_message_amount[i][0]) + "`" + " 贡献: " + str(
user_message_amount[i][1]) + "\n"
# print(top_5_user)
string = " ".join(word_list)
# 将string变量传入w的generate()方法,给词云输入文字
w.generate(string)
# 将词云图片导出到当前文件夹
w.to_file('{}_chat_word_cloud.png'.format(group))
bot.send_message(
chat_id=group,
text="🎤 今日话题榜 🎤\n"
"📅 {}\n"
"⏱ 截至今天{}\n"
"🗣️ 本群{}位朋友共产生{}条发言\n"
"🤹‍ 大家今天讨论最多的是:\n\n"
"{}\n"
"看下有没有你感兴趣的话题? 👏".format(
time.strftime("%Y年%m月%d", time.localtime()),
time.strftime("%H:%M", time.localtime()),
user_amount,
total_message_amount,
hot_word_string),
parse_mode="Markdown"
)
bot.send_message(
chat_id=group,
text="🏵 今日活跃用户排行榜 🏵\n"
"📅 {}\n"
"⏱ 截至今天{}\n\n"
"{}\n"
"感谢这些朋友今天的分享! 👏 \n"
"遇到问题,向他们请教说不定有惊喜😃".format(
time.strftime("%Y年%m月%d", time.localtime()),
time.strftime("%H:%M", time.localtime()),
top_5_user),
parse_mode="Markdown"
)
bot.send_message(
chat_id=group,
text="🏵 今日活跃用户排行榜 🏵\n"
"📅 {}\n"
"⏱ 截至今天{}\n\n"
"{}\n"
"感谢这些朋友今天的分享! 👏 \n"
"遇到问题,向他们请教说不定有惊喜😃".format(
time.strftime("%Y年%m月%d", time.localtime()),
time.strftime("%H:%M", time.localtime()),
top_5_user),
parse_mode="Markdown"
)
bot.send_photo(
chat_id=group,
photo=open("{}_chat_word_cloud.png".format(group), "rb")
)
bot.send_photo(
chat_id=group,
photo=open("{}_chat_word_cloud.png".format(group), "rb")
)
os.remove("{}_chat_word_cloud.png".format(group))
os.remove("{}_chat_word_cloud.png".format(group))
stop_time = float(time.time())
print("当前群组处理耗时:" + str(stop_time - start_time))
except Exception as e:
print(e)
continue
r.flushall()
print("已清空数据库")
stop_time = float(time.time())
print("当前群组处理耗时:" + str(stop_time - start_time))
except Exception as e:
print(e)
continue
r.flushall()
print("已清空数据库")