0. 说明

这是我自己写的一个时间标准化工具，功能是把一些常见的时间词描述转换为相对标准的YYYY-MM-DD格式。原理很简单，就是写规则匹配。

但是我刚写完这个代码之后，就找到了更好的项目，所以就把这个方法给淘汰了。

相比之下，这个功能不是很完备，也有很多亟待优化的地方，但是有这样一个方便的工具，总好过没有，所以把这个工具放出来给有需要的同学们使用。

1. 示例

等一下我们会写一个TimeNormalizer，现在假设已经写好了，我们来实例化它，然后就可以使用了：

tn = TimeNormalizer()norm_time = tn.normalize('Sep. 14, 2009')
# 2009-09-14norm_time = tn.normalize('in July, 1999')
# 1999-07-01__1999-07-31norm_time = tn.normalize('in Friday')
# 2022-08-05norm_time = tn.normalize('last Friday')
# 2022-07-29norm_time = tn.normalize('early Friday')
# 2022-08-05:06__2022-08-05:12norm_time = tn.normalize('in the autumn of 1997')
# 1997-09-01__1997-11-30norm_time = tn.normalize('in the past thirty-five days')
# 2022-06-29__2022-08-03norm_time = tn.normalize('yesterday night')
# 2022-08-02:20__2022-08-02:24

可以看到，其中很多时间描述词都是相对时间，所以我们需要一个基准时间，于是在设计这个类的时候，有一个参数pub_time（因为我原本是设计的以新闻的发布时间作为基准，所以起名叫publish了），这个参数需要是一个自定义的时间格式类型，下面介绍。

2. 时间格式

开始之前先import一下：

import re
from datetime import datetime, timedelta

在这里我定义了两种时间类型，时间点和时间片段，前者是用来描述一个时刻，例如某一天，后者则例如一个月，一年等。

时间点的定义如下：

class StandardTimePoint:"""标准时间点---------------ver: 2021-12-13by: changhongyu"""def __init__(self, std_y=None, std_m=None, std_d=None, std_h=None, pub_time=None, require_pub_time=True):""":param std_y: str: 标准年:param std_m: str: 标准月:param std_d: str: 标准日:param std_h: str: 标准时:param pub_time: StandardTimePoint: 基准时间:param require_pub_time: bool: 是否需要传入基准时间"""self.specific = ''  # 用于记录哪些字段是一开始就有确切数值的assert std_d, Exception("If you don't specify the date, this time object is a time period.")if std_y:assert len(std_y) == 4, ValueError("Attribute `std_y` must have format like `2021`.")self.specific += 'y'if std_m:assert len(std_m) == 2, ValueError("Attribute `std_m` must have format like `01`.")self.specific += 'm'if std_d:assert len(std_d) == 2, ValueError("Attribute `std_d` must have format like `31`.")self.specific += 'd'if std_h:assert len(std_h) == 2, ValueError("Attribute `std_h` must have format like `24`.")self.specific += 'h'self.std_y = std_yself.std_m = std_mself.std_d = std_dself.std_h = std_hif require_pub_time:if pub_time:self.pub_time = pub_timeelse:pub_time = datetime.now()_month = str(pub_time.month)_day = str(pub_time.day)if len(_day) == 1:_day = '0' + _dayif len(_month) == 1:_month = '0' + _monthself.pub_time = StandardTimePoint(std_y=str(pub_time.year),std_m=_month,std_d=_day,require_pub_time=False)self._fill()def __repr__(self):msg = ''if self.std_y:msg += self.std_yif self.std_m:if len(msg):msg += '-'msg += self.std_mif self.std_d:if len(msg) and not self.std_m:raise Exception("Unspecific month.")if len(msg):msg += '-'msg += self.std_dif not len(msg):raise Exception("Empty date.")if self.std_h:msg += ':'msg += self.std_hreturn msgdef _fill(self):if not self.std_y:self.std_y = self.pub_time.std_yif not self.std_m:self.std_m = self.pub_time.std_m

时间片段的定义如下：

class StandardTimePeriod:"""标准时间片段---------------ver: 2021-12-13by: changhongyu"""def __init__(self, std_y_s=None, std_m_s=None, std_d_s=None, std_h_s=None,std_y_e=None, std_m_e=None, std_d_e=None, std_h_e=None, pub_time=None, require_pub_time=True):""":param std_y_s: str: 标准年, 开始:param std_m_s: str: 标准月, 开始:param std_d_s: str: 标准日, 开始:param std_h_s: str: 标准时, 开始:param std_y_e: str: 标准年, 结束:param std_m_e: str: 标准月, 结束:param std_d_e: str: 标准日, 结束:param std_h_e: str: 标准时, 结束:param pub_time: StandardTimePoint: 基准时间:param require_pub_time: bool: 是否需要传入基准时间"""self.specific = ''  # 用于记录哪些字段是一开始就有确切数值的assert std_y_s or std_m_s or std_d_s, Exception("Empty date.")if std_y_s:assert std_y_e, Exception("If there is a start time, the end time is expected.")assert len(std_y_s) == len(std_y_e) == 4, \ValueError("Attribute `std_y` must have format like `2021`.")self.specific += 'y'if std_m_s:assert len(std_m_s) == len(std_m_e) == 2, \ValueError("Attribute `std_m` must have format like `01`.")self.specific += 'm'if std_d_s:assert len(std_d_s) == len(std_d_e) == 2, \ValueError("Attribute `std_d` must have format like `31`.")self.specific += 'd'if std_h_s:assert len(std_h_s) == len(std_h_e) == 2, \ValueError("Attribute `std_h` must have format like `24`.")self.specific += 'h'if require_pub_time:if pub_time:self.pub_time = pub_timeelse:pub_time = datetime.now()base_month = str(pub_time.month)base_day = str(pub_time.day)if len(base_month) == 1:base_month = '0' + base_monthif len(base_day) == 1:base_day = '0' + base_dayself.pub_time = StandardTimePoint(std_y=str(pub_time.year),std_m=base_month,std_d=base_day,require_pub_time=False)self.month_days = {i + 1: num for i, num in enumerate([31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31])}self.std_y_s = std_y_sself.std_m_s = std_m_sself.std_d_s = std_d_sself.std_h_s = std_h_sself.std_y_e = std_y_eself.std_m_e = std_m_eself.std_d_e = std_d_eself.std_h_e = std_h_eself._fill()self.start_time = StandardTimePoint(self.std_y_s, self.std_m_s, self.std_d_s, self.std_h_s)self.end_time = StandardTimePoint(self.std_y_e, self.std_m_e, self.std_d_e, self.std_h_e)def _fill(self):if not self.std_y_s:# 时间缺省，用基准时间补self.std_y_s = self.pub_time.std_yself.std_y_e = self.pub_time.std_yif not self.std_m_s:# 月份缺省，补全年self.std_m_s = '01'self.std_m_e = '12'if not self.std_d_s:# 日期缺省，补整月self.std_d_e = str(self.month_days[int(self.std_m_e)])self.std_d_s = '01'# 闰月调整if int(self.std_y_e) % 4 == 0 and int(self.std_y_e) % 100 != 0 and self.std_m_e == '02':self.std_d_e = '29'def __repr__(self):msg = ''msg += self.std_y_smsg += '-'msg += self.std_m_smsg += '-'msg += self.std_d_sif self.std_h_s:msg += ':'msg += self.std_h_smsg += '__'msg += self.std_y_emsg += '-'msg += self.std_m_emsg += '-'msg += self.std_d_eif self.std_h_e:msg += ':'msg += self.std_h_ereturn msg

那么当我们希望传入一个基准时间时，则需要实例化一个时间点：

std = StandardTimePoint(std_y='2022', std_m='01', std_d='01')

然后把它作为参数传给标准化工具进行实例化，就可以以指定的日期作为基准时间了，例如：

tn = TimeNormalizer(std)
tn.normalize('yesterday')
# 2021-12-31

3. 主类

下面是主类，直接上代码：

class TimeNormalizer:"""【时间标准化主类】分为三个方面对时间描述进行标准化：1. 基本格式化2. 节日, 特殊日期计算(TODO)3. 描述性推断具体过程：1) 先对完全标准结构进行匹配,例如`20020101`, `2002-01-01`, `2002-01-01:05`2) 然后匹配描述性推断,例如`last 3 years`3） 接下来处理半标准结构的十年例如`in the 90s`4) 半标准结构的年月日(包含缺省)例如`Nov. 1st, 1925`5) 时间分割切片例如`in the autumn of 1923`6) 最后处理简单结构例如`in Summer`, `on Wednesday`, `yesterday`---------------ver: 2021-12-16by: changhongyu"""def __init__(self, pub_time: StandardTimePoint = None):""":param pub_time: StandardTimePoint: 新闻时间作为基准时间"""if not pub_time:pub_time = datetime.now()base_month = str(pub_time.month)base_day = str(pub_time.day)if len(base_month) == 1:base_month = '0' + base_monthif len(base_day) == 1:base_day = '0' + base_dayself.pub_time = StandardTimePoint(std_y=str(pub_time.year),std_m=base_month,std_d=base_day,require_pub_time=False)else:self.pub_time = pub_time# 0-100的英文转数字num_str = ['zero', 'one', 'two', 'three', 'four', 'five', 'six', 'seven', 'eight', 'nine', 'ten','eleven', 'twelve', 'thirteen', 'fourteen', 'fifteen', 'sixteen', 'seventeen', 'eighteen','nineteen']num_str_ty = ['twenty', 'thirty', 'forty', 'fifty', 'sixty', 'seventy', 'eighty', 'ninety']num_str_to_int = {n: i for i, n in enumerate(num_str)}for i, ty in enumerate(num_str_ty):num_str_to_int[ty] = (i + 2) * 10for j in range(1, 10):ns = ty + '-' + num_str[j]num_str_to_int[ns] = (i + 2) * 10 + j# 做个反转，防止个位数先于十位数被匹配到key_list = [k for k in num_str_to_int.keys()][::-1]self.num_str_to_int = {k: num_str_to_int[k] for k in key_list}# 1. 基本格式化, 需注意所有表述均为title形式# 1) 月份相关self.month_long_string = ["January", "February", "March", "April", "May", "June","July", "August", "September", "October", "November", "December"]self.month_short_string = ["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"]self.num_days_of_every_month = [str(n) for n in [31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31]]# 2) 星期相关self.weekday_long_string = ['Monday', 'Tuesday', 'Wednesday', 'Thursday', 'Friday', 'Saturday', 'Sunday']self.weekday_short_string = ['Mon', 'TUE', 'WED', 'Thu', 'Fri', 'Sat', 'Sun']# 3) 分割描述self.y_sp = {'Early': ['01', '04'], 'Mid': ['05', '08'], 'Late': ['09', '12'],'Spring': ['03', '05'], 'Summer': ['06', '08'], 'Autumn': ['09', '11'], 'Winter': ['12', '02'],'01+Quarter': ['01', '03'], '02+Quarter': ['04', '06'],'03+Quarter': ['07', '09'], '04+Quarter': ['10', '12'], 'Last+Quarter': ['10', '12']}self.m_sp = {'Early': ['01', '10'], 'Mid': ['11', '20'], 'Late': ['21', '31']}self.d_sp = {'Morning': ['06', '10'], 'Noon': ['11', '14'], 'Afternoon': ['15', '18'], 'Evening': ['19', '22'],'Dawn': ['05', '07'], 'Day': ['08', '16'], 'Dusk': ['17', '19'], 'Night': ['20', '24'],'Mid+Night': ['01', '03'], 'Mid+Day': ['11', '13'],'Early': ['06', '12'], 'Late': ['18', '24']}# 2. 节日计算# 3. 描述性推断# 1) 介词推断self.preps = [s.title for s in ['on', 'in', 'at', 'before', 'after', 'during', 'since', 'toward']]# 2) 描述词self.present_words = ['This', 'Current', 'Present']self.previous_words = ['Past', 'Last', 'Previous']self.future_words = ['Next', 'Following', 'Coming', 'Up-Coming', 'Upcoming']self.delta_dict = {"Day": {'y': 0, 'm': 0, 'd': 1},"Week": {'y': 0, 'm': 0, 'd': 7},"Fortnight": {'y': 0, 'm': 0, 'd': 14},"Month": {'y': 0, 'm': 1, 'd': 0},"Season": {'y': 0, 'm': 3, 'd': 0},"Quarter": {'y': 0, 'm': 3, 'd': 0},"Year": {'y': 1, 'm': 0, 'd': 0},"Decade": {'y': 10, 'm': 0, 'd': 0},"Century": {'y': 100, 'm': 0, 'd': 0},"Centuries": {'y': 100, 'm': 0, 'd': 0},}def _init_pub(self):"""每次抽取完之后重置基准时间"""pub_time = datetime.now()base_month = str(pub_time.month)base_day = str(pub_time.day)if len(base_month) == 1:base_month = '0' + base_monthif len(base_day) == 1:base_day = '0' + base_dayself.pub_time = StandardTimePoint(std_y=str(pub_time.year),std_m=base_month,std_d=base_day,require_pub_time=False)def normalize(self, str_time, pub_time=None):"""将一个时间字符串标准化同类规则按照从复杂到简单的顺序对内容进行结构化当前边复杂的规则匹配到，则直接return结果否则会继续判断是否符合简单的规则:param str_time: str: 抽取出来的时间:param pub_time: StandardTimePoint:return:TODO: 顺序调整"""if pub_time:self.pub_time = pub_time# <0 完全标准结构, `20020101`, `2002-01-01`, `2002-01-01:05`normed_time = self._normalize_rule_0(str_time)if normed_time:self._init_pub()return normed_time# <4 `last 3 years`normed_time = self._normalize_rule_4(str_time)if normed_time:self._init_pub()return normed_time# <1 抽取十年单位, `in the 90s`normed_time = self._normalize_rule_1(str_time)if normed_time:self._init_pub()return normed_time# <2 抽取相对标准结构, `Nov. 1st, 1925`normed_time = self._normalize_rule_2(str_time)if normed_time:self._init_pub()return normed_time# <3 抽取分割时间描述, `in the autumn of 1923`normed_time = self._normalize_rule_3(str_time)if normed_time:self._init_pub()return normed_time# <5 单个的周几或者季节描述, 'in Summer'normed_time = self._normalize_rule_5(str_time)if normed_time:self._init_pub()return normed_timeself._init_pub()return None@staticmethoddef _normalize_rule_0(str_time):"""完全标准的结构进行转换例如：`20020101`, `2002-01-01`, `2002-01-01:05`:param str_time: str: 抽取出来的时间:return:"""str_time = str_time.replace('-', '').replace(':', '')if not str_time.isdigit():# 只允许存在`:`和`-`两种非数字字符returnif not len(str_time) == 8 or len(str_time) == 10:return_year = str_time[0: 4]_month = str_time[4: 6]_day = str_time[6: 8]_hour = str_time[8: 10] if len(str_time) == 10 else Nonereturn StandardTimePoint(std_y=_year, std_m=_month, std_d=_day, std_h=_hour)def _normalize_rule_1(self, str_time):"""抽取十年单位例如：in the 90s， in the early 1920s:param str_time: str: 抽取出来的时间:return:"""rule_1 = u"the [0-9]0s"pattern_1 = re.compile(rule_1)match = pattern_1.findall(str_time)if len(match):# 匹配到'the 90s'match_str = match[0]year_ty = match_str[-3]if year_ty > self.pub_time.std_y[-2]:# 大于新闻时间所在年份，则认定为上个世纪，否则默认本世纪year_thou = '19'else:year_thou = '20'if 'early' in str_time:std_y_s = year_thou + year_ty + '0'std_y_e = year_thou + year_ty + '4'elif 'late' in str_time:std_y_s = year_thou + year_ty + '5'std_y_e = year_thou + year_ty + '9'else:std_y_s = year_thou + year_ty + '0'std_y_e = year_thou + year_ty + '9'return StandardTimePeriod(std_y_s=std_y_s, std_y_e=std_y_e)else:rule_2 = u"the [0-9]{3}0s"pattern_2 = re.compile(rule_2)match = pattern_2.findall(str_time)if len(match):# 匹配到'the 1900s'match_str = match[0]std_y_s = match_str[4: 8]std_y_e = match_str[4: 7] + '9'return StandardTimePeriod(std_y_s=std_y_s, std_y_e=std_y_e)else:rule_3 = u"in early [1-2][0-9]{2}0s"pattern_3 = re.compile(rule_3)match = pattern_3.findall(str_time)if len(match):match_str = match[0]std_y_s = match_str[-5: -1]std_y_e = match_str[-5: -2] + '5'return StandardTimePeriod(std_y_s=std_y_s, std_y_e=std_y_e)rule_4 = u"in late [1-2][0-9]{2}0s"pattern_4 = re.compile(rule_4)match = pattern_4.findall(str_time)if len(match):match_str = match[0]std_y_s = match_str[-5: -2] + '6'std_y_e = match_str[-5: -2] + '9'return StandardTimePeriod(std_y_s=std_y_s, std_y_e=std_y_e)rule_5 = u"in [1-2][0-9]{2}0s"pattern_5 = re.compile(rule_5)match = pattern_5.findall(str_time)if len(match):match_str = match[0]std_y_s = match_str[-5: -1]std_y_e = match_str[-5: -2] + '9'return StandardTimePeriod(std_y_s=std_y_s, std_y_e=std_y_e)returndef _normalize_rule_2(self, str_time):"""抽取相对标准日期结构例如，`Nov. 1st, 1925`:param str_time: str: 抽取出来的时间:return:"""str_time = self.pre_format(str_time)_, _year, _month, _day, _hour, _split, __ = self._get_time_parts(str_time)if _split:# 如果有分割信息，则交给rule 3处理returnif not _month:returnelif _day:# 如果明确了日期，返回时间点if len(_day) == 1:_day = '0' + _dayif len(_month) == 1:_month = '0' + _monthreturn StandardTimePoint(std_y=_year, std_m=_month, std_d=_day, std_h=_hour)else:# 如果没有明确日期，返回时间片段return StandardTimePeriod(std_y_s=_year, std_y_e=_year, std_m_s=_month, std_m_e=_month)def _normalize_rule_3(self, str_time):"""抽取模糊表述例如：`in the autumn of 1923`, `in early April`, `at morning of the 7th.`:param str_time::return:"""str_time = self.pre_format(str_time)_prep, _year, _month, _day, _hour, _split, _weekday = self._get_time_parts(str_time)if _hour:# 如果最小单位是小时，直接返回时间点return StandardTimePoint(_year, _month, _day, _hour, self.pub_time)if _weekday:# 如果最小单位是星期几，则先找对应的星期，再对当日进行划分if _split and _split in self.d_sp:_hour_s, _hour_e = self.d_sp[_split]_weekday_num = Nonefor i, (wd_l, wd_s) in enumerate(zip(self.weekday_long_string, self.weekday_short_string)):if _weekday.lower() in [wd_l.lower(), wd_s.lower()]:_weekday_num = i + 1 if i != 6 else 0   # 0对应周天，1对应周一if not _weekday_num:raise ValueError('{} not found.'.format(_weekday))week_day_point = self._get_week_day(_weekday_num)week_day = StandardTimePeriod(std_y_s=week_day_point.std_y, std_m_s=week_day_point.std_m,std_d_s=week_day_point.std_d, std_h_s=_hour_s,std_y_e=week_day_point.std_y, std_m_e=week_day_point.std_m,std_d_e=week_day_point.std_d, std_h_e=_hour_e)return week_dayelif _day:# 如果最小单位是天，根据是否有分割描述，返回相应的时间片段或时间点if _split and _split in self.d_sp:# 如果有分割，则按照时间片段处理_hour_s, _hour_e = self.d_sp[_split]if not _month:_month = self.pub_time.std_mreturn StandardTimePeriod(std_y_s=_year, std_m_s=_month, std_d_s=_day, std_h_s=_hour_s,std_y_e=_year, std_m_e=_month, std_d_e=_day, std_h_e=_hour_e,pub_time=self.pub_time)else:# 如果没有分割描述，精确到天的时间按照时间点处理return StandardTimePoint(_year, _month, _day, None, self.pub_time)if _month:# 如果最小单位是月，则返回时间片段if _split and _split in self.m_sp:_day_s, _day_e = self.m_sp[_split]return StandardTimePeriod(std_y_s=_year, std_m_s=_month, std_d_s=_day_s,std_y_e=_year, std_m_e=_month, std_d_e=_day_e,pub_time=self.pub_time)else:return StandardTimePeriod(std_y_s=_year, std_m_s=_month, std_d_s='01',std_y_e=_year, std_m_e=_month, std_d_e=self.num_days_of_every_month[_month],pub_time=self.pub_time)if _year:# 如果最小单位是年，则返回时间片段if _split and _split in self.y_sp:_month_s, _month_e = self.y_sp[_split]return StandardTimePeriod(std_y_s=_year, std_m_s=_month_s,std_y_e=_year, std_m_e=_month_e,pub_time=self.pub_time)else:return StandardTimePeriod(std_y_s=_year, std_y_e=_year, pub_time=self.pub_time)if not _year and _split in self.y_sp:# 如果只说了是这年的什么时间，而没有明确年，则默认基准时间年_month_s, _month_e = self.y_sp[_split]return StandardTimePeriod(std_y_s=self.pub_time.std_y, std_m_s=_month_s,std_y_e=self.pub_time.std_y, std_m_e=_month_e,pub_time=self.pub_time)returndef _normalize_rule_4(self, str_time):"""描述性推断：描述词 + 时间例如：`this Wednesday`, `last month`, `past 2 weeks`:param str_time::return:TODO: two days ago"""str_time = self.__replace_str_num(str_time)for word in self.present_words + self.previous_words + self.future_words:# 第一遍循环, 匹配"last 2 days"结构if word in str_time.title():# 先尝试找两位数的rule = u"{} [0-9][0-9]".format(word)pattern = re.compile(rule)matched = pattern.findall(str_time.title())if len(matched):matched = matched[0]else:# 如果没有找到两位数的，再找一位数的rule_2 = u"{} [0-9]".format(word)pattern = re.compile(rule_2)matched = pattern.findall(str_time.title())if len(matched):matched = matched[0]else:continuetense_word, num_word = matched.split()num_word = int(num_word)assert tense_word == wordif tense_word in self.present_words:continueelif tense_word in self.previous_words:tense = -1else:tense = 1for time_span in self.delta_dict:# 判断时间词是年还是月还是日if time_span in str_time.title():if self.delta_dict[time_span]['y']:# 如果以年为单位计_delta_years = num_word * self.delta_dict[time_span]['y']if tense == -1:# 如果是过去到现在, 例如`in past two years`return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)-_delta_years),std_m_s=self.pub_time.std_m, std_d_s=self.pub_time.std_d,std_y_e=self.pub_time.std_y, std_m_e=self.pub_time.std_m,std_d_e=self.pub_time.std_d)else:# 如果是现在到将来，例如`in next two years`return StandardTimePeriod(std_y_s=self.pub_time.std_y, std_m_s=self.pub_time.std_m,std_d_s=self.pub_time.std_d,std_y_e=str(int(self.pub_time.std_y)+_delta_years),std_m_e=self.pub_time.std_m, std_d_e=self.pub_time.std_d)elif self.delta_dict[time_span]['m']:# 如果以月为单位计_delta_months = num_word * self.delta_dict[time_span]['m']_delta_years = 0while _delta_months >= 12:# 超过一年的换算成年_delta_years += 1_delta_months -= 12if tense == -1:return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)-_delta_years),std_m_s=str(int(self.pub_time.std_m)-_delta_months),std_d_s=self.pub_time.std_d, std_y_e=self.pub_time.std_y,std_m_e=self.pub_time.std_m, std_d_e=self.pub_time.std_d)elif tense == 1:return StandardTimePeriod(std_y_s=self.pub_time.std_y, std_m_s=self.pub_time.std_m,std_d_s=self.pub_time.std_d,std_y_e=str(int(self.pub_time.std_y)+_delta_years),std_m_e=str(int(self.pub_time.std_m)+_delta_months),std_d_e=self.pub_time.std_d)else:# 如果是以日为单位计# 这里其实可以改成_get_time_delta_days, 但是我懒得改了_delta_days = num_word * self.delta_dict[time_span]['d'] * tense_delta_days = timedelta(days=_delta_days)datetime_ = datetime(year=int(self.pub_time.std_y), month=int(self.pub_time.std_m),day=int(self.pub_time.std_d))target_datetime_ = datetime_ + _delta_days# 补零_m = str(datetime_.month) if datetime_.month > 9 else '0' + str(datetime_.month)_tar_m = str(target_datetime_.month) if target_datetime_.month > 9 \else '0' + str(target_datetime_.month)_d = str(datetime_.day) if datetime_.day > 9 else '0' + str(datetime_.day)_tar_d = str(target_datetime_.day) if target_datetime_.day > 9 \else '0' + str(target_datetime_.day)if tense == -1:return StandardTimePeriod(std_y_s=str(target_datetime_.year),std_m_s=_tar_m,std_d_s=_tar_d,std_y_e=str(datetime_.year),std_m_e=_m,std_d_e=_d)else:return StandardTimePeriod(std_y_s=str(datetime_.year),std_m_s=_m,std_d_s=_d,std_y_e=str(target_datetime_.year),std_m_e=_tar_m,std_d_e=_tar_d)for word in self.present_words + self.previous_words + self.future_words:# 第二遍循环, 匹配"last year"结构if word in str_time.title():if word in self.present_words:tense = 0elif word in self.previous_words:tense = -1else:tense = 1for time_span in self.delta_dict:if time_span in str_time.title():if self.delta_dict[time_span]['y']:# 如果以年为单位计if tense == -1:# 如果是过去到现在, 例如`last year`if time_span.lower() == 'year':return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)-1),std_y_e=str(int(self.pub_time.std_y)-1))elif time_span.lower() == 'decade':return self._get_decade_span(str(int(self.pub_time.std_y)-1))elif time_span.lower() == 'century':return self._get_century_span(str(int(self.pub_time.std_y)-1))else:passelif tense == 0:# 如果是当前m, 例如`this year`if time_span.lower() == 'year':return StandardTimePeriod(std_y_s=self.pub_time.std_y,std_y_e=self.pub_time.std_y)elif time_span.lower() == 'decade':return self._get_decade_span(self.pub_time.std_y)elif time_span.lower() == 'century':return self._get_century_span(self.pub_time.std_y)else:passelse:# 如果是现在到将来，例如`next decade`if time_span.lower() == 'year':return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)+1),std_y_e=str(int(self.pub_time.std_y+1)))elif time_span.lower() == 'decade':return self._get_decade_span(str(int(self.pub_time.std_y)+1))elif time_span.lower() == 'century':return self._get_century_span(str(int(self.pub_time.std_y)+1))else:passelif self.delta_dict[time_span]['m']:# 如果以月为单位计if tense == -1:if time_span.lower() == 'month':# 考虑推算到去年的情况if int(self.pub_time.std_m)-1 <= 0:_year = str(int(self.pub_time.std_y)-1)_month = str(12 + int(self.pub_time.std_m)-1)if len(_month) == 1:_month = '0' + _monthelse:_year = self.pub_time.std_y_month = str(int(self.pub_time.std_m)-1)if len(_month) == 1:_month = '0' + _monthreturn StandardTimePeriod(std_y_s=_year, std_m_s=_month,std_y_e=_year, std_m_e=_month)elif time_span.lower() == 'quarter':return self._get_quarter_span(year=self.pub_time.std_y,month=self.pub_time.std_m)else:passelif tense == 0:if time_span.lower() == 'month':return StandardTimePeriod(std_m_s=self.pub_time.std_m,std_m_e=self.pub_time.std_m)elif time_span.lower() == 'quarter':return self._get_quarter_span(year=self.pub_time.std_y,month=self.pub_time.std_m)else:passif tense == 1:if time_span.lower() == 'month':# 考虑推算到明年的情况if int(self.pub_time.std_m) + 1 >= 0:_year = str(int(self.pub_time.std_y) + 1)_month = str(int(self.pub_time.std_m) + 1 - 12)else:_year = self.pub_time.std_y_month = str(int(self.pub_time.std_m) + 1)return StandardTimePeriod(std_y_s=_year, std_m_s=_month,std_y_e=_year, std_m_e=_month)elif time_span.lower() == 'quarter':return self._get_quarter_span(year=self.pub_time.std_y,month=self.pub_time.std_m)else:# 如果是以日为单位计if tense == -1:if time_span.lower() == 'week':last_week_time_point = self._get_time_delta_days(7, 'backward')return self._get_week_span(year=last_week_time_point.std_y,month=last_week_time_point.std_m,day=last_week_time_point.std_d)elif time_span.lower() == 'fortnight':# 如果是过去的两周则直接向前推算14天return self._get_time_delta_days(14, direction='backward')elif time_span.lower() == 'day':# last day 这种说法似乎不太合理return self._get_time_delta_days(1, direction='backward')else:passif tense == 0:if time_span.lower() == 'week':return self._get_week_span(year=self.pub_time.std_y,month=self.pub_time.std_m,day=self.pub_time.std_d)elif time_span.lower() == 'fortnight':# 没有这种说法passelif time_span.lower() == 'day':return StandardTimePoint(self.pub_time.std_y,self.pub_time.std_m,self.pub_time.std_d)else:passif tense == 1:if time_span.lower() == 'week':next_week_time_point = self._get_time_delta_days(7, 'forward')return self._get_week_span(year=next_week_time_point.std_y,month=next_week_time_point.std_m,day=next_week_time_point.std_d)elif time_span.lower() == 'fortnight':return self._get_time_delta_days(14, direction='forward')elif time_span.lower() == 'day':return self._get_time_delta_days(1, direction='forward')else:passfor week_day in self.weekday_long_string:# `this Sunday`if week_day in str_time.title():if tense == -1:# `last Monday`last_week_time_point = self._get_time_delta_days(7, 'backward')return self._get_week_day(self._weekday_to_num(week_day), last_week_time_point)elif tense == 0:# `this Sunday`return self._get_week_day(self._weekday_to_num(week_day))else:next_week_time_point = self._get_time_delta_days(7, 'forward')return self._get_week_day(self._weekday_to_num(week_day), next_week_time_point)for season in ['Spring', 'Summer', 'Autumn', 'Winter']:# `this Summer`if season in str_time.title():if season != 'Winter':if tense == -1:return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)-1),std_m_s=self.y_sp[season][0],std_y_e=str(int(self.pub_time.std_y)-1),std_m_e=self.y_sp[season][1])elif tense == 0:return StandardTimePeriod(std_m_s=self.y_sp[season][0],std_m_e=self.y_sp[season][1])else:return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)+1),std_m_s=self.y_sp[season][0],std_y_e=str(int(self.pub_time.std_y)+1),std_m_e=self.y_sp[season][1])else:if tense == -1:return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)-1),std_m_s=self.y_sp[season][0],std_y_e=self.pub_time.std_y,std_m_e=self.y_sp[season][1])elif tense == 0:# 如果当前时间是冬天，则this winter指的是去年底到今年初# 如果当前时间不是冬天，则this winter指的是今年底到明年初if 3 <= int(self.pub_time.std_m) < 12:return StandardTimePeriod(std_y_s=self.pub_time.std_y,std_m_s=self.y_sp[season][0],std_y_e=str(int(self.pub_time.std_y)+1),std_m_e=self.y_sp[season][1])else:return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y)-1),std_m_s=self.y_sp[season][0],std_y_e=self.pub_time.std_y,std_m_e=self.y_sp[season][1])else:return StandardTimePeriod(std_y_s=self.pub_time.std_y,std_m_s=self.y_sp[season][0],std_y_e=str(int(self.pub_time.std_y)+1),std_m_e=self.y_sp[season][1])returndef _normalize_rule_5(self, str_time):"""星期、季节推断rule4中包含了一部分星期，这里的星期是单个的星期，不存在描述词，所以在4之后`in Summer`, `on Tuesday`:param str_time::return:"""for week_day in self.weekday_short_string + self.weekday_long_string:if week_day in str_time.title():if week_day in self.weekday_short_string:w_word = self._weekday_short_to_long(week_day)else:w_word = week_dayreturn self._get_week_day(self._weekday_to_num(w_word))for season in ['Spring', 'Summer', 'Autumn', 'Winter']:if season in str_time.title():if season != 'Winter':return StandardTimePeriod(std_m_s=self.y_sp[season][0],std_m_e=self.y_sp[season][1])else:if 3 <= int(self.pub_time.std_m) < 12:return StandardTimePeriod(std_y_s=self.pub_time.std_y,std_m_s=self.y_sp[season][0],std_y_e=str(int(self.pub_time.std_y) + 1),std_m_e=self.y_sp[season][1])else:return StandardTimePeriod(std_y_s=str(int(self.pub_time.std_y) - 1),std_m_s=self.y_sp[season][0],std_y_e=self.pub_time.std_y,std_m_e=self.y_sp[season][1])str_time = self.pre_format(str_time)for specific_day in ['Today', 'Yesterday', 'Tomorrow']:if specific_day in str_time.title():if specific_day == 'Today':_tar_day = self.pub_timeelif specific_day == 'Yesterday':_tar_day = self._get_time_delta_days(1, direction='backward')else:_tar_day = self._get_time_delta_days(1, direction='forward')for dsp in self.d_sp:# 一天之内的分割词，例如morning# 如果匹配到，按照时间片段算，否则按照时间点算作某一天if dsp in str_time.title():return StandardTimePeriod(std_y_s=_tar_day.std_y, std_m_s=_tar_day.std_m,std_d_s=_tar_day.std_d, std_h_s=self.d_sp[dsp][0],std_y_e=_tar_day.std_y, std_m_e=_tar_day.std_m,std_d_e=_tar_day.std_d, std_h_e=self.d_sp[dsp][1])return StandardTimePoint(std_y=_tar_day.std_y,std_m=_tar_day.std_m,std_d=_tar_day.std_d)if 'Tonight' in str_time.title():return StandardTimePeriod(std_y_s=self.pub_time.std_y, std_m_s=self.pub_time.std_m,std_d_s=self.pub_time.std_d, std_h_s='18',std_y_e=self.pub_time.std_y, std_m_e=self.pub_time.std_m,std_d_e=self.pub_time.std_d, std_h_e='24')return@staticmethoddef pre_format(str_time):"""预处理格式化"""while '  ' in str_time:str_time = str_time.replace('  ', ' ')str_time = str_time.title()str_time = str_time.replace('1St', '01').replace('2Nd', '02').replace('3Rd', '03').replace('4Th', '04')\.replace('5Th', '05').replace('6Th', '06').replace('7Th', '07').replace('8Th', '08').replace('9Th', '09')str_time = str_time.replace(' Quarter', '+Quarter').replace('Mid ', 'Mid+')  # 用加号防止split将结构切开str_time = str_time.replace('St ', ' ').replace('Nd ', ' ').replace('Rd ', ' ').replace('Th ', ' ')str_time = str_time.replace('First', '01').replace('Second', '02').replace('Third', '03')\.replace('Fourth', '04').replace('Fifth', '05').replace('Sixth', '06').replace('Seventh', '07')\.replace('Eighth', '08').replace('Ninth', '09').replace('Tenth', '10')return str_timedef _get_time_parts(self, str_time):"""对时间切片得到各个成分"""_prep, _year, _month, _day, _hour, _split, _weekday = None, None, None, None, None, None, Nonetime_parts = str_time.split()for part in time_parts:if part in self.preps:_prep = partelif part.title().replace('.', '').replace(',', '') in self.month_long_string + self.month_short_string:_month = part.title().replace('.', '').replace(',', '')if _month in self.month_short_string:_month = self._month_short_to_long(_month)_month = self._month_to_num(_month)elif part.replace(',', '') in ['0' + str(i) if i < 10 else str(i) for i in range(1, 32)] + \[str(i) for i in range(1, 32)]:_day = part.replace(',', '')if len(_day) == 1:_day = '0' + _dayelif '1000' < part < '2100':_year = partelif part in self.y_sp or part in self.m_sp or part in self.d_sp:_split = partelif part.title().replace('.', '').replace(',', '') in self.weekday_long_string + self.weekday_short_string:_weekday = part.title().replace('.', '').replace(',', '')if _day:assert not _weekday, Exception('If there have been a specific day, a weekday is unexpected.')return _prep, _year, _month, _day, _hour, _split, _weekdaydef _month_short_to_long(self, short_month):short_month = short_month.title()month_short_to_long_dict = {k: v for k, v in zip(self.month_short_string, self.month_long_string)}return month_short_to_long_dict[short_month]def _month_to_num(self, month):month_to_num_dict = {m: str(i + 1) if i >= 9 else '0' + str(i + 1) for i, m inenumerate(self.month_long_string)}return month_to_num_dict[month.title()]def _weekday_short_to_long(self, short_weekday):short_weekday = short_weekday.title()weekday_short_to_long_dict = {k: v for k, v in zip(self.weekday_short_string, self.weekday_long_string)}return weekday_short_to_long_dict[short_weekday]def _weekday_to_num(self, weekday):for i, w in enumerate(self.weekday_long_string):if weekday == w:if weekday == "Sunday":return 0else:return i + 1raise ValueError("{} is not a standard weekday.".format(weekday))@staticmethoddef _get_decade_span(year):"""给定一年，获取这一年所在的十年的spanExample: (2021) --> [(2020, 01, 01), (2029, 12, 31)]:param year: str:return: StandardTimePeriod"""_year_s = year[:3] + '0'_year_e = year[:3] + '9'return StandardTimePeriod(std_y_s=_year_s, std_y_e=_year_e)@staticmethoddef _get_century_span(year):"""给定一年，获取这一年所在的世纪的span:param year: str:return: StandardTimePeriod"""_year_s = year[:2] + '00'_year_e = year[:2] + '99'return StandardTimePeriod(std_y_s=_year_s, std_y_e=_year_e)@staticmethoddef _get_quarter_span(year, month):"""给定一个月，获取这个月所在季度的span:param year: str:param month: str:return: StandardTimePeriod"""if int(month) in [1, 2, 3]:_month_s = '01'_month_e = '03'elif int(month) in [4, 5, 6]:_month_s = '04'_month_e = '06'elif int(month) in [7, 8, 9]:_month_s = '07'_month_e = '09'elif int(month) in [10, 11, 12]:_month_s = '10'_month_e = '12'else:raise Exception()return StandardTimePeriod(std_y_s=year, std_m_s=_month_s, std_y_e=year, std_m_e=_month_e)@staticmethoddef _get_week_span(year, month, day):"""给定一个日期，获取这一天所在的一周的span:param year: str:param month: str:param day: str:return: StandardTimePeriod"""this_day = datetime(int(year), int(month), int(day))this_weekday = this_day.weekday()if this_weekday == 6:# 如果是周天，要调整到第一天this_weekday = 0else:this_weekday += 1_time_delta_fore = timedelta(days=-this_weekday)_time_delta_aftr = timedelta(days=6-this_weekday)week_start = this_day + _time_delta_foreweek_end = this_day + _time_delta_aftrm_s = str(week_start.month) if week_start.month > 9 else '0' + str(week_start.month)m_e = str(week_end.month) if week_end.month > 9 else '0' + str(week_end.month)d_s = str(week_start.day) if week_start.day > 9 else '0' + str(week_start.day)d_e = str(week_end.day) if week_end.day > 9 else '0' + str(week_end.day)return StandardTimePeriod(std_y_s=str(week_start.year), std_m_s=m_s, std_d_s=d_s,std_y_e=str(week_start.year), std_m_e=m_e, std_d_e=d_e)def _get_time_delta_days(self, days, direction='backward', base_time=None):"""在基准日期的基础上推算几天前后几天后的日期:param days: int: 推算几天:param direction: str: 前向还是后项, backward or forward:param base_time: StandardTimePeriod:return: StandardTimePoint"""assert direction in ['backward', 'forward'], Exception("Direction must be `forward` or `backward`.")if not base_time:base_time = self.pub_timeif direction == 'backward':_delta = timedelta(days=-days)else:_delta = timedelta(days=days)_dt = datetime(int(base_time.std_y), int(base_time.std_m), int(base_time.std_d))_tar_dt = _dt + _delta_tar_y = str(_tar_dt.year)_tar_m = str(_tar_dt.month) if _tar_dt.month > 9 else '0' + str(_tar_dt.month)_tar_d = str(_tar_dt.day) if _tar_dt.day > 9 else '0' + str(_tar_dt.day)return StandardTimePoint(std_y=_tar_y, std_m=_tar_m, std_d=_tar_d)def _get_week_day(self, weekday, base_week_time=None):"""获取基准时间所在周的周几:param weekday: int: 0: Sunday; 6: Saturday:return: StandardTimePoint"""while weekday >= 7:weekday -= 7if not base_week_time:base_week_time = self.pub_timeweek = self._get_week_span(base_week_time.std_y, base_week_time.std_m, base_week_time.std_d)tar_day = datetime(int(week.std_y_s), int(week.std_m_s), int(week.std_d_s)) + timedelta(days=weekday)_year = str(tar_day.year)_month = str(tar_day.month) if tar_day.month > 9 else '0' + str(tar_day.month)_day = str(tar_day.day) if tar_day.day > 9 else '0' + str(tar_day.day)return StandardTimePoint(std_y=_year, std_m=_month, std_d=_day)@staticmethoddef __num_to_int(num):"""0-99的数字转换，英文描述或字符串，转整型:param num: str:return:"""nums = ['zero', 'one', 'two', 'three', 'four', 'five', 'six', 'seven', 'eight', 'nine', 'ten','eleven', 'twelve', 'thirteen', 'fourteen', 'fifteen', 'sixteen', 'seventeen', 'eighteen', 'nineteen']num_to_int = {n: i for i, n in enumerate(nums)}nums_ty = ['twenty', 'thirty', 'forty', 'fifty', 'sixty', 'seventy', 'eighty', 'ninety']num_to_int_ty = {n: (i + 2) * 10 for i, n in enumerate(nums_ty)}try:return int(num)except:try:if '-' in num:num_1, num_2 = num.split('-')  # 如果对不上，则跳转finallyreturn num_to_int_ty[num_1.lower().replace(' ', '')] + num_to_int[num_2.lower().replace(' ', '')]else:return num_to_int[num.lower().replace(' ', '')]except:return 0def __replace_str_num(self, text):"""如果字符串中英文单词描述的数字，则将其替换成数字例如: 'seven days' --> '7 days'"""for k in self.num_str_to_int:text = text.replace(k, str(self.num_str_to_int[k]))return text

总之，我这个时间标准化工具写的不怎么完备，思想也非常朴素，但是它的的确确可以解决一部分时间标准化的问题（如果你有更好用的标准化工具那当我没说）。

如果这个工具对你有所帮助，记得点赞支持哦。

NLP工具——自制英文时间标准化工具相关推荐

【工具】java 时间比较工具
[工具]java 时间比较工具需要依赖hutool /*** 比较时间* @param startDate 开始* @param endDate 结束* @param nowDate 当前* @re ...
15、【商品管理模块开发】——后台获取商品详情功能开发及PropertiesUtil配置工具，DateTimeUtil时间处理工具开发...
1.后台获取商品详情接口: 在上一篇文章所新建的ProudctManageController类中新建下面方法: *Controller: //获取商品详情接口@RequestMapping(&quo ...
java 日期时间工具_Java日期时间类工具
Java日期时间类工具 Java日期时间类 Date类 Java提供的Date类和Calendar类用于处理日期和时间的类,包括创建日期,时间对象,获取系统当前日期,时间等操作.但Date类无法实现 ...
微软官方工具_英文写作在线工具推荐：微软官方免费打分评改网站微软小英
好像给大家推这个工具的时间晚了点,因为四六级和考研已经结束了......不过没关系,如果你需要的话,可以从现在开始使用它备战下一次考试(没有恶意)!今天推荐的「微软小英写作网站」,是正儿八经由微软官方 ...
NLP英文伪原创工具
英文伪原创工具有什么用?英文伪原创是针对英文网站内容更新提供内容更新,通过NLP自然语言处理,实现语言的自动伪原创,英文伪原创的方式很多,可以通过中文文档翻译成英文实现伪原创,也可以通过英文文档翻译成 ...
NLP工具——自制zero-shot事件抽取器
NLP工具--自制zero-shot事件抽取器 0. 简介 1. 抽取全部潜在的事件 2. 抽取特定类型的事件 3. 结语 0. 简介在事件抽取任务中,数据的获取是一件非常关键工作,由于数据标注的成 ...
Java格式化日期用斜杠_[java工具类01]__构建格式化输出日期和时间的工具类
在之前的学习中,我写过一篇关于字符串格式化的,就主要设计到了时间以及日期的各种格式化显示的设置,其主要时通过String类的fomat()方法实现的. 我们可以通过使用不同的转换符来实现格式化显示不同 ...
在线使用的时间管理工具
http://baike.baidu.com/view/102959.htm http://baike.baidu.com/view/470808.htm http://pm.shangxueba.c ...
Unix 时间戳（Unix Timestamp）与 Windows 时间转换工具
Unix 时间戳 (Unix Timestamp) 与 Windows 时间转换工具 2012-08-27 21:30:24 sys 2 评论 9,958 浏览我发现,我在本站发布的文章 ...

NLP工具——自制英文时间标准化工具