推荐 | 文本分析库cntext2.x使用手册

cd desktop

pip3 install cntext-2.1.6-py3-none-any.whl

import cntext as ct

print('当前cntext版本: ', ct.__version__)
help(ct)

当前cntext版本: 2.1.6

Help on package cntext:

NAME
    cntext

PACKAGE CONTENTS
    io
    mind
    model
    stats
    llm
...

import cntext as ct

ct.get_dict_list()

['zh_common_NTUSD.yaml',
 'zh_common_DUTIR.yaml',
 'enzh_common_StopWords.yaml',
 'en_valence_Concreteness.yaml',
 'en_common_LoughranMcDonald.yaml',
 'zh_common_FinanceSenti.yaml',
 'zh_common_FLS.yaml',
 'zh_common_TsinghuaPraiseDegrade.yaml',
 'zh_common_FEPU.yaml',
 'en_common_ANEW.yaml',
 'en_common_NRC.yaml',
 'zh_valence_ChineseEmoBank.yaml',
 'zh_valence_SixSemanticDimensionDatabase.yaml',
 'zh_common_FinacialFormalUnformal.yaml',
 'zh_common_LoughranMcDonald.yaml',
 'enzh_common_AdvConj.yaml',
 'en_common_SentiWS.yaml',
 'zh_common_Digitalization.yaml',
 'en_common_LSD2015.yaml',
 'zh_common_HowNet.yaml',
 'zh_common_EPU.yaml']

import cntext as ct
print(ct.read_yaml_dict('zh_common_Digitalization.yaml'))

{'Name': '中文数字化词典',
'Desc': '基于这篇论文，构建了中文数字化词典，含人工智能技术、大数据技术、云计算技术、区块链技术、数字技术应用等关键词列表。 ', 'Refer': '吴非,胡慧芷,林慧妍,任晓怡. 企业数字化转型与资本市场表现——来自股票流动性的经验证据[J]. 管理世界,2021,37(07):130-144+10.',
'Category': ['Artificial_Intelligence', 'Big_Data', 'Cloud_Computing', 'Block_Chains', 'Usage_of_Digitalization'],

'Dictionary':
    {'Artificial_Intelligence': ['人工智能', '商业智能', '图像理解', '投资决策辅助系统', '智能数据分析', '智能机器人', '机器学习', '深度学习', '语义搜索', '生物识别技术', '人脸识别', '语音识别', '身份验证', '自动驾驶', '自然语言处理'],
    'Big_Data': ['大数据', '数据挖掘', '文本挖掘', '数据可视化', '异构数据', '征信', '增强现实', '混合现实', '虚拟现实'],
    'Cloud_Computing': ['云计算', '流计算', '图计算', '内存计算', '多方安全计算', '类脑计算', '绿色计算', '认知计算', '融合架构', '亿级并发', 'EB级存储', '物联网', '信息物理系统'],
    'Block_Chains': ['区块链', '数字货币', '分布式计算', '差分隐私技术', '智能金融合约'],
    'Usage_of_Digitalization': ['移动互联网', '工业互联网', '移动互联', '互联网医疗', '电子商务', '移动支付', '第三方支付', 'NFC支付', '智能能源', 'B2B', 'B2C', 'C2B', 'C2C', 'O2O', '网联', '智能穿戴', '智慧农业', '智能交通', '智能医疗', '智能客服', '智能家居', '智能投顾', '智能文旅', '智能环保', '智能电网', '智能营销', '数字营销', '无人零售', '互联网金融', '数字金融', 'Fintech', '金融科技', '量化金融', '开放银行']}}

ct.detect_encoding(file)

import cntext as ct

#读取data文件夹下的【三体.txt】
#识别编码方式
ct.detect_encoding(file='data/三体.txt')

utf-8

#查看【文件夹data】内所有的 txt文件。
ct.get_files(fformat='data/*.txt')

['data/三体.txt',
 'data/santi.txt',
 'data/w2v_corpus.txt',
 'data/sopmi_corpus.txt',
 'data/brown_corpus.txt',
 'data/sopmi_seed_words.txt']

ct.read_pdf(file)

import cntext as ct

text = ct.read_pdf('格力电器2023.pdf')
print(text)

珠海格力电器股份有限公司 2023年年度报告全文
珠海格力电器股份有限公司
2023年年度报告


二〇二四年四月
珠海格力电器股份有限公司 2023年年度报告全文
 第 2 页 共 249 页 第一节 重要提示、目录和释义
公司董事会、监事会及董事、监事、高级管理人员保证年度报告内容
的真实、准确、完整，不存在虚假记载、误导性陈述或重大遗漏，并承担
个别和连带的法律
......

ct.read_docx(file)

import cntext as ct

text = ct.read_docx('test.docx')
text

这是来自test.docx里内容

ct.read_file(file, encoding='utf-8')

import cntext as ct

#默认encoding='utf-8'
#sdf = ct.read_file(file='data/三体.txt')

sdf = ct.read_file(file='data/三体.txt', encoding='utf-8')
sdf

ct.read_files(fformat, encoding='utf-8'）

import cntext as ct

#默认encoding='utf-8'
#ddf = ct.read_files(fformat='data/*.txt')

ddf = ct.read_files(fformat='data/*.txt', encoding='utf-8')
ddf

ct.extract_mda(text, kws_pattern='')

kws_pattern = '董事会报告|董事会报告与管理讨论|企业运营与管理评述|经营总结与分析|管理层评估与未来展望|董事局报告|管理层讨论与分析|经营情况讨论与分析|经营业绩分析|业务回顾与展望|公司经营分析|管理层评论与分析|执行摘要与业务回顾|业务运营分析'

import cntext as ct

text = ct.read_pdf('格力电器2023.pdf')
mda_text = ct.extract_mda(text)
print(mda_text)

'管理层讨论与分析  \n一、报告期内公司所处行业情况  \n（一）行业发展现状  \n1.消费领域 ——家电行业稳定增长，空调市场恢复明显  \n2023年，中国经济保持了整体恢复向好的态势，激发消费是稳增长的重中之重。国家鼓励和推动消费品以旧换\n新，促进消费经济大循环，加速更新需求释放，推动高能效产品设备销售和出口增长，进一步激发绿色消费潜力。  \n1）家电行业稳定增长  \n2023年，国内经济恢复明显，家电行业稳定增长。根据全国家用电器工业信息中心发布的《 2023年中国家电\n行业年度报告》，家电行业外销明显增长，出口规模为 6,174亿元，同比增长 9.9%；国内市场实现稳步增长，销售\n规模为7'
.......
.......

import glob
import cntext as ct

print('extract_mda识别能力')
for year in range(2001, 2024):
    num = 0
    for file in glob.glob(f'年报txt/{year}/*.txt'):
        mda_text = ct.extract_mda(open(file).read())
        if mda_text!='':
            num = num + 1

    volume = len(glob.glob(f'年报txt/{year}/*.txt'))
    ratio = num/volume

    print(f'{year}: {ratio:.2f}')

2001: 0.24
2002: 0.37
2003: 0.43
2004: 0.70
2005: 0.77
2006: 0.78
2007: 0.79
2008: 0.77
2009: 0.79
2010: 0.82
2011: 0.84
2012: 0.96
2013: 0.95
2014: 0.98
2015: 0.98
2016: 0.99
2017: 0.98
2018: 0.98
2019: 0.99
2020: 0.97
2021: 0.98
2022: 0.99
2023: 0.99

ct.traditional2simple(text, mode='t2s')

import cntext as ct

text = '簡體漢字'
ct.traditional2simple(text)

'简体汉字'

text = '简体汉字'
ct.traditional2simple(text, mode='s2t')

'簡體漢字'

import cntext as ct

raw_text = '今日起可中遇到技术问题，可以拨打电话０３７１－６６３２１９９１、６６３２１９７３咨询。'

text = ct.fix_text(raw_text)
text

今日起可中遇到技术问题，可以拨打电话0371-66321991、66321973咨询。

- you're -> you are
- yall  -> you all
- gotta  -> got to
...

import cntext as ct

raw_text = "yall're happy now"

text = ct.fix_contractions(raw_text)
text

"you all are happy now"

ct.word_count(text, lang='chinese', return_df=False)

import cntext as ct

text = '致力于致力于以零文章处理费或订阅费发布优质研究软件。'

#ct.word_count(text, lang='chinese')
ct.word_count(text)

Counter({'致力于': 2,
         '文章': 1,
         '处理费': 1,
         '订阅费': 1,
         '发布': 1,
         '优质': 1,
         '研究': 1,
         '软件': 1})

ct.word_count(text, return_df=True)

ct.readability(text, lang='chinese', syllables=3, return_series=False)

Gunning Fog Index = 0.4 * (Total_Words/Total_Sentences + 100 * Complex_Words/Total_Words)
SMOG Index = 1.0430 * sqrt(Complex_Words/Total_Sentences) * 30 + 3.1291
Coleman-Liau Index = 0.0588 * (100*Total_Letters/Total_Words) -0.296*(100*Total_Sentences/Total_Words) - 15.8
Automated Readability Index(ARI) = 4.71 * (Total_Characters/Total_Words) + 0.5*(Total_Words/Total_Sentences) - 21.43
Readability Index(RIX) = Complex_Words * (6 + Total_characters) / Total_Sentences

import cntext as ct

text = '致力于以零文章处理费或订阅费发布优质研究软件。'

ct.readability(text, lang='chinese', syllables=3)

{'fog_index': 120.4,
 'flesch_kincaid_grade_level': 20.2,
 'smog_index': 57.32,
 'coleman_liau_index': 83.96,
 'ari': 87.4,
 'rix': 87.0}

sentiment(text, diction, lang='chinese', return_series=False)

import cntext as ct

diction = {'pos': ['高兴', '快乐', '分享'],
           'neg': ['难过', '悲伤'],
           'adv': ['很', '特别']}

text = '我今天得奖了，很高兴，我要将快乐分享大家。'
ct.sentiment(text=text,
             diction=diction,
             lang='chinese')

{'pos_num': 3,
 'neg_num': 0,
 'adv_num': 1,
 'stopword_num': 8,
 'word_num': 14,
 'sentence_num': 1}

ct.sentiment_by_valence(text, diction, lang='chinese', return_series=False)

import cntext as ct
import pandas as pd

concreteness_dict = ct.read_yaml_dict('en_valence_Concreteness.yaml')['Dictionary']
concreteness_dict

{'roadsweeper': {'concreteness': 4.85},
 'traindriver': {'concreteness': 4.54},
 'tush': {'concreteness': 4.45},
 'hairdress': {'concreteness': 3.93},
 'pharmaceutics': {'concreteness': 3.77},
 'hoover': {'concreteness': 3.76},
 'shopkeeping': {'concreteness': 3.18},
 'pushiness': {'concreteness': 2.48},
 ......
 }

reply = "I'll go look for that"

score=ct.sentiment_by_valence(text=reply,
                              diction=concreteness_dict,
                              lang='english')

score

{'concreteness': 9.28,
'word_num': 6}

employee_replys = ["I'll go look for that",
                   "I'll go search for that",
                   "I'll go search for that top",
                   "I'll go search for that t-shirt",
                   "I'll go look for that t-shirt in grey",
                   "I'll go search for that t-shirt in grey"]

for idx, reply in enumerate(employee_replys):
    score=ct.sentiment_by_valence(text=reply,
                                  diction=concreteness_dict,
                                  lang='english')

    template = "Concreteness Score: {score:.2f} | Example-{idx}: {exmaple}"

    print(template.format(score=score['concreteness'],
                          idx=idx,
                          exmaple=reply))

Concreteness Score: 9.28 | Example-0: I'll go look for that
Concreteness Score: 9.32 | Example-1: I'll go search for that
Concreteness Score: 13.25 | Example-2: I'll go search for that top
Concreteness Score: 14.25 | Example-3: I'll go search for that t-shirt
Concreteness Score: 21.32 | Example-4: I'll go look for that t-shirt in grey
Concreteness Score: 21.36 | Example-5: I'll go search for that t-shirt in grey

ct.word_in_context(text, keywords, window=3, lang='chinese')

import cntext as ct

#测试代码，假设zh_text是年报文本，从找找出丝网词相关词的上下文
zh_text = """
【插入一条自家广告】大邓自己家的家，
安平县多隆丝网制品，生产销售不锈钢轧花网、
电焊网、石笼网、刀片刺绳、冲孔网等丝网制品。
联系人 邓颖静 0318-7686899

人生苦短，我学Python
在社科中，可以用Python做文本分析
Python是一门功能强大的编程语言，广泛应用在经管社科领域。
可以做网络爬虫、文本分析、LDA话题模型、相似度分析等。

今年经济不景气，形势异常严峻。
由于疫情不景气，静默管理， 产品积压， 公司经营困难。
保就业促就业，任务十分艰巨。
"""

#【python】上下文
ct.word_in_context(text = zh_text,
                   keywords = ['python'],
                   window=10,
                   lang='chinese')

epu(df, freq='Y', e_pattern='', p_pattern='', u_pattern='')

import cntext as ct

#省略，读取数据得到 news_df

epu_df = ct.epu(df=news_df, freq='m')
epu_df

ct.fepu(text, ep_pattern, u_pattern)

import cntext as ct

#省略，读取数据得到 mda_df

fepu_df = df['经营讨论与分析内容'].apply(ct.fepu)
res_df = pd.concat([df[['会计年度', '股票代码']], fepu_df],   axis=1)
res_df

ct.semantic_brand_score(text, brands, lang='chinese')

import cntext as ct

brands = ['汪淼', '史强', '罗辑', '叶文洁', '伊文斯']

#准备santi_test_text
#小说等分20份， 读取第一份得到santi_test_text

sbs_df = ct.semantic_brand_score(text=santi_test_text,
                               brands=brands,
                               lang='chinese')
sbs_df

ct.cosine_sim(text1, text2, lang='chinese')   cos余弦相似
ct.jaccard_sim(text1, text2, lang='chinese')  jaccard相似
ct.minedit_sim(text1, text2, lang='chinese')  最小编辑距离相似度；
ct.simple_sim(text1, text2, lang='chinese')   更改变动算法

import cntext as ct

text1 = '编程真好玩编程真好玩'
text2 = '游戏真好玩编程真好玩'

print('cosine', ct.cosine_sim(text1, text2, lang='chinese'))
print('jaccard', ct.jaccard_sim(text1, text2, lang='chinese'))
print('minedit', ct.minedit_sim(text1, text2, lang='chinese'))
print('simple', ct.simple_sim(text1, text2, lang='chinese'))

cosine  0.82
jaccard 0.67
minedit 1.00
simple 0.84

import cntext as ct


text1 = 'Programming is fun!'
text2 = 'Programming is interesting!'

print('cosine', ct.cosine_sim(text1, text2, lang='english'))
print('jaccard', ct.jaccard_sim(text1, text2, lang='english'))
print('minedit', ct.minedit_sim(text1, text2, lang='english'))
print('simple', ct.simple_sim(text1, text2, lang='english'))

cosine  0.67
jaccard 0.50
minedit 1.00
simple 0.78

import cntext as ct

personA = '这场音乐会太嗨了'
personB = '这场音乐会说出来令你不敢相信，主办方策划有方，群众激情满满，我印象深刻，体验感拉满'


print('A-hhi', ct.word_hhi(personA))
print('B-hhi', ct.word_hhi(personB))

print('A词汇多样性', 1 - ct.word_hhi(personA))
print('B词汇多样性', 1 - ct.word_hhi(personB))

A-hhi 0.20000000000000004
B-hhi 0.07024793388429751

A词汇多样性 0.7999999999999999
B词汇多样性 0.9297520661157025

import cntext as ct

plt = ct.matplotlib_chinese()
plt.figure(figsize=(7, 4))
plt.plot([1, 2, 3, 4], [1, 4, 9, 16])
plt.title('中文图表', fontsize=10)
plt.show()

ct.lexical_dispersion_plot1(text, targets_dict, lang='chinese', figsize=(12, 6), title='特定词汇在不同文本来源的相对离散图', prop=True)

import cntext as ct

roles_dict = {
    "汪淼": ['汪淼'],
    "叶文洁": ['叶文洁'],
    "罗辑": ['罗辑']
}

santi_text = open('三体.txt', encoding='utf-8').read()

ax = ct.lexical_dispersion_plot1(text = santi_text,  #文本数据
                            targets_dict = roles_dict, #角色
                            figsize = (10, 4),  #尺寸大小
                            lang = 'chinese',  #中文数据
                            title = '《三体》小说角色出现位置', #标题
                            prop = True)    #相对位置(横坐标轴取值范围0-100)
ax

ct.lexical_dispersion_plot1(text = santi_text,  #文本数据
                            targets_dict = roles_dict, #角色
                            figsize = (10, 4),  #尺寸大小
                            lang = 'chinese',  #中文数据
                            title = '《三体》小说角色出现位置', #标题
                            prop = False)    #绝对位置(横坐标轴取值范围与小说文本长度有关)

import cntext as ct

# diy了一个小词典
senti_dict = {
    'pos': ['开心', '幸福', '快乐', '安宁', '希望'],
    'neg': ['紧张', '恐惧', '害怕', '绝望']
}

santi_text = open('三体.txt', encoding='utf-8').read()

ax = ct.lexical_dispersion_plot1(text = santi_text,
                            targets_dict = senti_dict,
                            figsize = (10, 2),
                            lang = 'chinese',
                            title = '《三体》情绪词出现位置',
                            prop = True)
ax

ct.lexical_dispersion_plot2(texts_dict, targets, lang='chinese', figsize=(12, 6), title='特定词汇在不同文本来源的相对离散图')

targets = ['太空', '宇宙']

texts_dict = {'三体': open('三体.txt', encoding='utf-8').read(),
              '基地': open('基地.txt', encoding='utf-8').read()}

ax = ct.lexical_dispersion_plot2(texts_dict = texts_dict,
                            targets = targets,
                            figsize = (10, 2),
                            title = '"太空/宇宙"词语出现位置',
                            lang = 'chinese')
ax

ct.Word2Vec(corpus_file, lang='chinese', dict_file=None, stopwords_file=None, vector_size=100, window_size=6, min_count=5, max_iter=5, chunksize=10000, only_binary=True, **kwargs)

import cntext as ct

w2v = ct.Word2Vec(corpus_file = 'data/三体.txt',
                  lang = 'chinese',
                  window_size = 6,
                  vector_size = 50)


w2v

Mac(Linux) System, Enable Parallel Processing
Cache output/三体_cache.txt Not Found or Empty, Preprocessing Corpus
Reading Preprocessed Corpus from output/三体_cache.txt
Start Training Word2Vec
Word2Vec Training Cost 10 s.
Output Saved To: output/Word2Vec/三体-Word2Vec.50.6.bin

ct.GloVe(corpus_file, lang='chinese', dict_file=None, stopwords_file=None, vector_size=100, window_size=15, min_count=5, max_memory=4.0, max_iter=15, x_max=10, only_binary=True, chunksize=10000)

import cntext as ct

glove = ct.GloVe(corpus_file='data/三体.txt',
                 lang='chinese',
                 vector_size=50,
                 window_size=15)

glove

Mac(Linux) System, Enable Parallel Processing
Cache output/三体_cache.txt Not Found or Empty, Preprocessing Corpus
Start Training GloVe
BUILDING VOCABULARY
Using vocabulary of size 6975.

COUNTING COOCCURRENCES
Merging cooccurrence files: processed 2106999 lines.

Using random seed 1743474106
SHUFFLING COOCCURRENCES
Merging temp files: processed 2106999 lines.

TRAINING MODEL
Read 2106999 lines.
Using random seed 1743474106
04/01/25 - 10:21.46AM, iter: 001, cost: 0.055981
04/01/25 - 10:21.46AM, iter: 002, cost: 0.050632
......
04/01/25 - 10:21.48AM, iter: 014, cost: 0.030047
04/01/25 - 10:21.48AM, iter: 015, cost: 0.029100

GloVe Training Cost 9 s.
Output Saved To: output/三体-GloVe.50.15.bin
<gensim.models.keyedvectors.KeyedVectors at 0x331517440>

ct.evaluate_similarity(wv, file=None)

足球	足球	4.98
老虎	老虎	4.8888888889
恒星	恒星	4.7222222222
入场券	门票	4.5962962963
空间	化学	0.9222222222
股票	电话	0.92
国王	车	0.9074074074
中午	字符串	0.6
收音机	工作	0.6
教授	黄瓜	0.5
自行车	鸟	0.5
蛋白质	文物	0.15

import cntext as ct

# 可在 https://cntext.readthedocs.io/zh-cn/latest/embeddings.html 下载该模型文件
dm_w2v = ct.load_w2v('output/douban-movie-1000w-Word2Vec.200.15.bin')

# 使用内置评估文件
ct.evaluate_similarity(wv=dm_w2v)
# 使用自定义评估文件
# ct.evaluate_similarity(wv=dm_w2v, file='diy_similarity.txt')

近义测试: similarity.txt
/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/cntext/model/evaluate_data/similarity.txt
Processing Similarity Test: 100%|██████████| 537/537 [00:00<00:00, 85604.55it/s]

评估结果：
+----------+------------+----------------------------+
| 发现词语 | 未发现词语 | Spearman's Rank Coeficient |
+----------+------------+----------------------------+
|   459    |     78     |            0.43            |
+----------+------------+----------------------------+

ct.evaluate_analogy(wv, file=None)

: CapitalOfCountries
雅典 希腊 巴格达 伊拉克
哈瓦那 古巴 马德里 西班牙
河内 越南 伦敦 英国
: CityInProvince
石家庄 河北 南昌 江西
沈阳 辽宁 南昌 江西
南京 江苏 郑州 河南
: FamilyRelationship
男孩 女孩 兄弟 姐妹
男孩 女孩 国王 王后
父亲 母亲 国王 王后
丈夫 妻子 叔叔 阿姨
: SocialScience-Concepts
社会 社会结构 家庭 家庭结构
文化 文化传承 语言 语言传承
群体 群体行为 组织 组织行为

import cntext as ct

# 可在 https://cntext.readthedocs.io/zh-cn/latest/embeddings.html 下载该模型文件
dm_w2v = ct.load_w2v('output/douban-movie-1000w-Word2Vec.200.15.bin')

# 使用内置评估文件
ct.evaluate_analogy(wv=dm_w2v)
# 使用自定义评估文件
# ct.evaluate_analogy(wv=dm_w2v, file='diy_analogy.txt')

类比测试: analogy.txt
/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/cntext/model/evaluate_data/analogy.txt
Processing Analogy Test: 100%|█████████████| 1198/1198 [00:11<00:00, 103.52it/s]

评估结果：
+--------------------+----------+------------+------------+----------+
|      Category      | 发现词语 | 未发现词语 | 准确率 (%) | 平均排名 |
+--------------------+----------+------------+------------+----------+
| CapitalOfCountries |   615    |     62     |   39.02    |   2.98   |
|   CityInProvince   |   175    |     0      |   28.57    |   4.74   |
| FamilyRelationship |   272    |     0      |   92.65    |   1.48   |
|   SocialScience    |    8     |     62     |   25.00    |   6.00   |
+--------------------+----------+------------+------------+----------+

ct.SoPmi(corpus_file, seed_file)       #人工标注的初始种子词

import cntext as ct

ct.SoPmi(corpus_file='data/sopmi_corpus.txt',
         seed_file='data/sopmi_seed.txt')       # 人工标注的初始种子词

Step 1/4:...Preprocess   Corpus ...
Step 2/4:...Collect co-occurrency information ...
Step 3/4:...Calculate   mutual information ...
Step 4/4:...Save    candidate words ...
Finish! used 19.74 s

ct.load_w2v(w2v_path)

import cntext as ct

santi_w2v = ct.load_w2v(w2v_path='output/三体-Word2Vec.50.6.bin')
# santi_w2v = ct.load_wv(wv_path='output/三体-Word2Vec.50.6.txt')

santi_glove = ct.load_w2v(w2v_path='output/三体-GloVe.50.15.bin')
# santi_glove = ct.load_wv(wv_path='output/三体-GloVe.50.15.bin')

santi_w2v

Loading output/三体-Word2Vec.50.6.bin...
Loading output/三体-GloVe.50.15.bin...
<gensim.models.keyedvectors.KeyedVectors at 0x33aa9cf80>

ct.glove2word2vec(glove_file, word2vec_file)

import cntext as ct
ct.glove2word2vec(glove_file='data/GloVe.6B.50d.txt',
                  word2vec_file='output/word2vec_format_GloVe.6B.50d.txt')

ct.expand_dictionary(wv,  seeddict, topn=100)

seeddict = {
    '人物': ['叶文洁', '史强', '罗辑'],
    '物体': ['飞船', '车辆']
}


ct.expand_dictionary(wv=santi_w2v.wv,
                     seeddict=seeddict,
                     topn=10)

import cntext as ct

# 获取词向量文件 https://cntext.readthedocs.io/zh-cn/latest/embeddings.html
w2v = ct.load_w2v('专利摘要-Word2Vec.200.15.bin')
semantic_centroid(wv=w2v, words=['创新', '颠覆'])

array([ 0.15567462, -0.05117003, -0.18534171,  0.20808656, -0.01133028,
        0.10738188, -0.02571066,  0.06051835,  0.00107351,  0.08017981,
        0.08914138,  0.01845527,  0.06232869, -0.03851539, -0.17092938,
        0.02196799, -0.04136903,  0.11350462, -0.09539546,  0.04907424,
        0.01268489,  0.05294977,  0.08449743, -0.02762416,  0.02332745,
        0.08865491, -0.06260188, -0.0378293 ,  0.04771722,  0.05745243,
        0.04417403, -0.04126203, -0.02403288, -0.03834526,  0.08115771,
        0.01508994,  0.07678635,  0.01395652,  0.1360324 ,  0.03027042,
       -0.02819572,  0.02339242,  0.11504567,  0.02910597,  0.06149592,
        0.01126606, -0.10132807,  0.07762785, -0.01214836,  0.03780747,
        0.12758181, -0.03115267, -0.19343086, -0.21930983,  0.05253006,
       -0.01452067, -0.07067247, -0.04237257, -0.08911953,  0.08573315,
        0.02742999,  0.05392318,  0.02916237,  0.04465031, -0.0788566 ,
       -0.07088121,  0.03111146,  0.00387428, -0.04032568,  0.14935694,
       -0.03880607,  0.07259471,  0.01711774, -0.05551507,  0.01039889,
        0.00666137,  0.03313185,  0.03169986,  0.08127907,  0.0239668 ,
       -0.00991806, -0.04201584,  0.01199235, -0.08669737, -0.02087858,
       -0.03440931,  0.02360864,  0.06623896, -0.01020982,  0.01200165,
        0.01059455,  0.13041293,  0.01103112,  0.03814259, -0.01519256,
        0.02946554,  0.00593279,  0.08796389,  0.0198915 , -0.0569265 ,
       -0.14622693,  0.07680258, -0.02288322, -0.04959924,  0.03325186,
        0.11031196,  0.06893978,  0.04289736, -0.0307357 , -0.09662723,
        0.02554002,  0.05394766,  0.047071  , -0.09522557, -0.08160087,
       -0.01467315, -0.01304489,  0.07513782,  0.04484766, -0.0516454 ,
        0.00648148,  0.01093231, -0.00303798, -0.06217093,  0.02755075,
       -0.10749754, -0.05205868, -0.02562402,  0.09068517,  0.05208463,
       -0.11790312,  0.02881086, -0.02414756,  0.00192055,  0.03881926,
       -0.05390498,  0.06648378,  0.02055933, -0.07083403, -0.07248309,
       -0.12991821,  0.0603951 ,  0.14131376, -0.01507344, -0.06480791,
       -0.08994781, -0.03397571,  0.0108852 , -0.02777362,  0.01159309,
        0.00121858, -0.0690551 , -0.07747664,  0.03437752, -0.14576062,
        0.06320656, -0.10743124, -0.01910913,  0.15803815, -0.03027673,
       -0.02909171, -0.03350233, -0.0694584 , -0.09807504, -0.09133697,
       -0.01123043,  0.04894681, -0.01971908, -0.08290677, -0.00336836,
        0.09619438, -0.03496556,  0.09733834, -0.0421683 ,  0.01408717,
        0.03355598,  0.00748263,  0.011903  , -0.12909584,  0.01545653,
        0.07656407,  0.09496018,  0.0608537 ,  0.00597665, -0.01628997,
        0.06285962, -0.16796936, -0.0486528 ,  0.01525079, -0.03067709,
       -0.02952635, -0.02731965, -0.06351878,  0.03577968,  0.0457835 ,
        0.08370785, -0.03491699, -0.12606403, -0.08686454, -0.04782247])

import cntext as ct

# 获取词向量文件
# https://github.com/hiDaDeng/Chinese-Pretrained-Word-Embeddings
dm_w2v = ct.load_w2v('douban-movie-1000w-Word2Vec.200.15.bin')
gender_axis_vector = ct.generate_concept_axis(wv=dm_w2v,
                                              poswords=['男', '男人', '父亲'],
                                              negwords=['女', '女人', '母亲'])
gender_axis_vector

array([-0.0118976 ,  0.03178174, -0.04656127,  0.00613294, -0.03692355,
       -0.06293361, -0.04739443,  0.01368712,  0.02603469, -0.02268519,
       -0.09925436,  0.05780286,  0.11218373,  0.07519485,  0.06885784,
        0.05505687, -0.04097392,  0.1737831 ,  0.05118835, -0.06879821,
        0.04762978,  0.02224233, -0.04891564, -0.08712718, -0.01432874,
       -0.07395219,  0.01229804,  0.06655715, -0.01864985, -0.04864848,
        0.00260787,  0.06843776,  0.00472286,  0.03623124,  0.11959086,
       -0.04683099, -0.11005358,  0.0271024 , -0.05976011,  0.12669185,
        0.03592191, -0.01125782, -0.02587771, -0.02719228,  0.0507662 ,
       -0.09198377,  0.09546432, -0.01937146,  0.06106697, -0.0405688 ,
       -0.1311393 ,  0.06090249,  0.03515694,  0.01364273, -0.02491697,
        0.03379048, -0.06635275,  0.01432849,  0.01212378, -0.0625283 ,
       -0.03481676, -0.0422427 , -0.17145215, -0.06323837,  0.02563147,
       -0.02371969,  0.01217621, -0.00346871,  0.07024875,  0.08295133,
        0.00731711, -0.01932047,  0.02165518, -0.09927654, -0.08531073,
        0.01949702,  0.00536061,  0.10426087, -0.02010326,  0.02297032,
       -0.10657956,  0.1035546 ,  0.00569263, -0.0849498 ,  0.1098236 ,
        0.05310893, -0.0802139 , -0.01034231, -0.12204715,  0.01407488,
       -0.01781198, -0.0134118 ,  0.09836894,  0.16098371,  0.00609895,
        0.05433145, -0.08940306,  0.00136946, -0.08455469, -0.08432727,
        0.04675778, -0.03415223, -0.18552355, -0.05219543, -0.01127822,
        0.02059881, -0.08120015, -0.15610164,  0.01439221,  0.01727759,
       -0.14516874,  0.01783531, -0.13099317,  0.03820422,  0.03033866,
       -0.01779634,  0.07759558,  0.15866944,  0.00191632, -0.00905253,
        0.0312649 , -0.05698524,  0.07270953, -0.00734233,  0.06289094,
        0.01014149, -0.0052088 ,  0.02478063, -0.0112649 , -0.0930789 ,
        0.14639418, -0.08183327, -0.08392337, -0.01458992, -0.0163887 ,
        0.06790476, -0.03252221,  0.08593727,  0.10469338, -0.01363467,
        0.00749907, -0.01320484,  0.08405331,  0.0489707 , -0.11343482,
       -0.10319041, -0.02415894,  0.13382405, -0.01983603, -0.00990637,
       -0.03335103,  0.11718886, -0.05802442, -0.18935862, -0.07409969,
       -0.08306517, -0.04423901,  0.11331058,  0.00588326,  0.06339834,
        0.04405889,  0.1263905 , -0.007273  , -0.02706875,  0.02325469,
       -0.13092995,  0.02056245, -0.0442118 , -0.01964739, -0.06501938,
        0.02196051, -0.1823353 ,  0.04273191,  0.01935809, -0.01464438,
       -0.02626805,  0.09194217,  0.02489716,  0.05376589, -0.00484252,
        0.02822759,  0.06744799, -0.14196248,  0.03016541, -0.05347864,
       -0.16907257,  0.05094757,  0.0721257 , -0.00421157,  0.03022675,
       -0.00047884,  0.07792547, -0.00209365,  0.0669208 ,  0.02009218,
        0.11358768, -0.05002993,  0.01760067,  0.03407429, -0.0893421 ],
      dtype=float32)

ct.sematic_distance(wv, words1, words2)

male_concept = ['male', 'man', 'he', 'him']
female_concept = ['female', 'woman', 'she', 'her']
engineer_concept  = ['engineer',  'programming',  'software']

dist(male, engineer) = distance(male_concept,  engineer_concept)
dist(female, engineer) = distance(female_concept,  engineer_concept)

import cntext as ct

# glove_w2v.6B.100d.txt链接: https://pan.baidu.com/s/1MMfQ7M0YCzL9Klp4zrlHBw 提取码: 72l0
g_wv = ct.load_w2v('data/glove_w2v.6B.100d.txt')

engineer = ['program', 'software', 'computer']
man_words =  ["man", "he", "him"]
woman_words = ["woman", "she", "her"]

dist_male_engineer = ct.sematic_distance(male_concept,  engineer_concept)
dist_female_engineer = ct.sematic_distance(female_concept,  engineer_concept)

dist_male_engineer - dist_female_engineer

-0.5

ct.sematic_projection(wv, words, poswords, negwords, return_full=False, cosine=False)

import cntext as ct
animals = ['mouse', 'cat', 'horse',  'pig', 'whale']
small_words= ["small", "little", "tiny"]
large_words = ["large", "big", "huge"]

# wiki_wv = ct.load_w2v('wiki的word2vec模型文件路径')
# wiki_wv

# In size conception, mouse is smallest, horse is biggest.
# 在大小概念上，老鼠最小，马是最大的。
ct.sematic_projection(wv=wiki_wv,
                      words=animals,
                      poswords=large_words,
                      negwords=small_words,
                      )

[('mouse', -1.68),
 ('cat', -0.92),
 ('pig', -0.46),
 ('whale', -0.24),
 ('horse', 0.4)]

project_word(wv, a, b, cosine=False)

b='苗条'
for a in ['性感','美丽', '可爱', '丑陋']:
    proj = ct.project_word(dm_w2v, a, b)
    print(f'[{a}]在[{b}]投影值: {proj}')


b='修长'
for a in ['性感','美丽', '可爱', '丑陋']:
    proj = ct.project_word(dm_w2v, a, b)
    print(f'[{a}]在[{b}]投影值: {proj}')

[性感]在[苗条]投影值: 14.172947883605957
[美丽]在[苗条]投影值: 7.0944623947143555
[可爱]在[苗条]投影值: 6.935092926025391
[丑陋]在[苗条]投影值: 1.235807180404663

[性感]在[修长]投影值: 14.599699974060059
[美丽]在[修长]投影值: 9.360642433166504
[可爱]在[修长]投影值: 4.740543842315674
[丑陋]在[修长]投影值: 4.010622501373291

for a in ['性感','美丽', '可爱', '丑陋']:
    proj = ct.project_word(wv=dm_w2v, a=a, b=['修长', '苗条'])
    print(f'[{a}]在[修长，苗条]投影值: {proj}')

[性感]在[修长，苗条]投影值: 15.807487487792969
[美丽]在[修长，苗条]投影值: 9.040315628051758
[可爱]在[修长，苗条]投影值: 6.414511203765869
[丑陋]在[修长，苗条]投影值: 2.882350444793701

ct.divergent_association_task(wv, words)

low_words = ["arm", "eyes", "feet", "hand", "head", "leg", "body"]
average_words = ["bag", "bee", "burger", "feast", "office", "shoes", "tree"]
high_words = ["hippo", "jumper", "machinery", "prickle", "tickets", "tomato", "violin"]

# 导入模型，得到wv。
# wv = ct.load_w2v('wiki的word2vec模型文件路径')


print(ct.divergent_association_task(wv, low_words)) # 50
print(ct.divergent_association_task(wv, average_words)) # 78
print(ct.divergent_association_task(wv, high_words)) # 95

50
78
95

ct.discursive_diversity_score(wv, words)

ct.procrustes_align(base_wv, other_wv, words=None)

import cntext as ct

PROMPT = '从口味taste、速度speed、服务service三个维度， 对外卖评论内容进行文本分析， 分别返回不同维度的分值(分值范围-1.0 ~ 1.0)'
BASE_URL = 'https://dashscope.aliyuncs.com/compatible-mode/v1'
API_KEY = '你的API-KEY'
MODEL_NAME = 'qwen-max'

#味道、速度、服务
OUTPUT_FORMAT = {'taste': float, 'speed': float, 'service': float}

COMMENT_CONTENT = '太难吃了'

# 使用
# result = ct.analysis_by_llm(text=COMMENT_CONTENT,
# 或
result = ct.text_analysis_by_llm(text=COMMENT_CONTENT,
                                 prompt=PROMPT,
                                 base_url=BASE_URL,
                                 api_key=API_KEY,
                                 model_name=MODEL_NAME,
                                 temperature=0,
                                 output_format=OUTPUT_FORMAT,
                                 max_retries=3,
                                 return_df=False)

result

{'taste': -1.0, 'speed': 0.0, 'service': 0.0}

import pandas as pd
import cntext as ct


# 构造实验数据
data = ['速度非常快，口味非常好， 服务非常棒！',
        '送餐时间还是比较久',
        '送单很快，菜也不错赞',
        '太难吃了']
df = pd.DataFrame(data, columns=['comment'])


# 分析函数
def llm_analysis(text):
    result = ct.analysis_by_llm(text=text,
                                prompt= '从口味taste、速度speed、服务service三个维度， 对外卖评论内容进行文本分析， 分别返回不同维度的分值(分值范围-1.0 ~ 1.0)',
                                base_url='https://dashscope.aliyuncs.com/compatible-mode/v1',
                                api_key='你的API-KEY',
                                model_name='qwen-max',
                                output_format={'taste': float, 'speed': float, 'service': float}
                               )
    return pd.Series(result)


# 批量运算
df2 = df['comment'].apply(llm_analysis)
res_df = pd.concat([df, df2], axis=1)
# 保存分析结果
res_df.to_csv('result.csv', index=False)
res_df

@misc{deng2022cntext,
  author       = {Deng, X. and Nan, P.},
  title        = {cntext: a Python tool for text mining},
  year         = {2022},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.7063523},
  url          = {https://doi.org/10.5281/zenodo.7063523},
  howpublished = {[Computer software]},
  note         = {Source Code URL: \url{https://github.com/hiDaDeng/cntext}}
}

%0 Generic
%A Deng, X.
%A Nan, P.
%T cntext: a Python tool for text mining
%Y [Computer software]
%D 2022
%I Zenodo
%R 10.5281/zenodo.7063523
%U https://doi.org/10.5281/zenodo.7063523
%Z Source Code URL: https://github.com/hiDaDeng/cntext
%@