20210115

20210115-mtg.md

Original : https://lunarwoffie.com/ja/lda-topic-model/

data_0048.txt

"*",
"4468",
"電話",
"(31)ウィルス対策ソフト包括ライセンス(TrendMicro)","WindowsのUpdateをしてしまったが問題あるか",
"2020/11/25",
"2020/11/25",
"",
"11/17お知らせを見る前に、WindowsのUpdateをしてしまったが問題あるか",
"10",
"2020/11/25 13:05",
"10",
"問い合わせ・相談",
"2020/11/25 13:05",
"職員",
"",
"使用しているPCは事務用一括端末（デスクトプ）とのこと
特にウィルス対策ソフトを入れなおした記憶はないとのことなので、ApexOneご利用ではないのでアナウンス対象外であることをお伝えしました",
"NW4",
"2020/11/25 12:58",
"",
""

lunar1.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
display(news_ss.head()) 

lunar12.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

#lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
display(processed_news_ss.head()) 

lunar123.py

# フォントの入手元 https://moji.or.jp/ipafont/ipafontdownload/
# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 


# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

show_wordcloud(processed_news_ss) 

lunar1234.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
print(tfidf_data.toarray()) 
print(tfidf_data.shape)

lunar12345.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar5.py header
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

#lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
print(tfidf_data.toarray()) 
print(tfidf_data.shape)

# lunar5.py
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

def gridsearch_best_model(tfidf_data, plot_enabled=True):
    # Define Search Param
    n_topics = [4,5,6,7,8,9]
    search_params = {'n_components': n_topics}
    
    # Init the Model
    lda = LDA(max_iter=25,               # Max learning iterations
              learning_method='batch',   
              random_state=0,            # Random state
              n_jobs = -1,               # Use all available CPUs)
              )
    
    # Init Grid Search Class
    model = GridSearchCV(lda, param_grid=search_params)
    
    # Do the Grid Search
    model.fit(tfidf_data)
    
    # Best Model
    best_lda_model = model.best_estimator_
    
    # Model Parameters
    print("Best Model's Params: ", model.best_params_)
    
    # Log Likelihood Score
    print("Best Log Likelihood Score: ", model.best_score_)
    
    # Perplexity
    print("Model Perplexity: ", best_lda_model.perplexity(tfidf_data))
    
    # Get Log Likelyhoods from Grid Search Output
    log_likelyhoods_score = [round(score) for score in model.cv_results_["mean_test_score"]]
    
    if plot_enabled:
        # Show graph
        plt.figure(figsize=(12, 8))
        plt.plot(n_topics, log_likelyhoods_score)
        plt.title("Choosing Optimal LDA Model")
        plt.xlabel("Numer of Topics")
        plt.ylabel("Log Likelyhood Scores")
        plt.show()

    return best_lda_model

lunar123456-run.txt

(bachelor) user@MacBook-Pro bachelor % python3 lunar123456.py
File already exists.
Best Model's Params:  {'n_components': 4}
Best Log Likelihood Score:  -138824.45102727457
Model Perplexity:  20994.88944972521

Topic # 0 :
八重歯,犬山,紀伊國屋書店,ﾊﾞｲｷﾝｸﾞ,亜麻,草津,ﾚﾅｳﾝ,木嶋,おまじない,献灯,希子,水原,ｲﾜﾅ,前原,浅香,新駅,一宮,千歳,烏山,紙子,阿波,提灯,うずまく,鳴門,種子,高松,ﾐｯﾁ,章一,ﾔｽ,一平,陣羽織,杏子,庄造,近松,茂兵衛,おさん,金沢,ぬき,影武者,帝国ﾎﾃﾙ,安城,富士川,宇多津,今治,西大寺,黒部,水溶,法体の滝,山の幸,坂出,請負う,鳴門海峡,ぞんぶん,吉野川,南烏山,前川,ﾐｯﾁｰ,ｷﾁ,東武,越谷,商店,犬丸,乗り入れる,泉岳寺,ゆめﾀｳﾝ,光代,芦屋,不出来,推論,じき,亮吉,尾道,持ち掛ける,不義,順二,困苦,商家,裏長屋,門左衛門,透徹,ｺﾞｰﾘｷｰ,悲恋,下宿,艶笑,とらや,附近,大経師,亭主,あや子,奥方,密通,赤貧,寅次郎,溝口,手代,貸本,驟雨,八重,落とし物,ｺﾞﾎﾞｳ,四日市,北口,三郷,鈴鹿,三吉,中山寺,城東,久居,豊橋,河辺,二の宮,江南,網走,八幡,ここち,石野,懐古,東奔西走,久喜,ｱﾋﾟﾀ,天童,銀天,嬉野,道頓堀,河原町,郡上,伊予,武生,段原,北光,商業高校前,北大通,近畿日本鉄道,水堂,梅島,伊勢崎線,扇田,伊万里,野々井,氷見,北長池,さば,津幡,塩冶有原,五橋,砥部,上中居,三ｹ,苦竹,小山東,花堂,花台,三好,発寒,色内,幸田,小菅,能美,能代,常滑,菰野,神の倉,神郡,ﾄﾞﾄｰﾙｺｰﾋｰ,福井東,台原,紙屋,岡南,糠塚,みたけ,古市,箱崎宮前,川筋,等々力,上之町,田家,宇佐,御経塚,二川,安中,和戸,志度,渋川,湯本,せんげ,滑川,宝来,室蘭,観音寺,久留,潮江,丸岡,上里,西梅田,南大阪線,中之,南町,牟佐,袋町,袋井,藤野,不動寺,南貴崎,藍住,名駅南,下之,四番丁,津原,笠松,木屋,新居浜,堀切,日宇,大博,魚津,高平,朝霧,大坪,土橋,東武動物公園,東予,大府,月輪,木津川,別宮,元町,麻布十番,高屋,東川口,東向島,飯山,札幌南,国府,駅東,東金沢,八乙女,月形,旭ヶ丘,士別,高浜,新清水,四国,植木,晴氏,武尾,青葉,わかめ,桑名,下衆,ﾊﾞｲﾊﾟｽ,熱海,香椎,奇人,駅前,新大阪,八王子,真菜,黒松,海産物,ﾋﾞﾀﾐﾝe,草丈,長町,小田原,如意,ちゃめ,山東,たんぽ,東成瀬,焚き火,釣竿,轟音,野趣,高雄,鳥海,かきわける,須賀,本荘,平定,由宇,かぶりつく,田町,大原,長浜,西新橋,館林,名勝,姫路,跡地,的場,急先鋒,春日井,平塚,亀山,佐世保,ﾛｯﾃﾘｱ,御徒,花巻,銚子,関口,門間,傍聴,西日暮里,蒲田,不美人,輪島,紀伊國屋,羽目板,がり,鹿島,草加,一巡,入れ歯,付けまつげ,蟹江,水臭い,ｾﾙﾌｻｰﾋﾞｽ,小樽,辛酸,菊川,一秀,政調,政経,植草,丸亀,法務省,盛岡,愛宕,苫小牧,横須賀,明石,市営,いけ好かない,うるむ,小松,天王寺,立川,一朗,讃岐,富山,真子,七尾,範疇,長屋,磐田,尼崎,おひたし,山頂,塩焼き,足利,仲違い,連れ帰る,東西線,那覇,木更津,浩平,ﾎﾟﾝ酢,諏訪,空港通,さじ,愛子,大田,函館,吉岡,若妻,ﾊﾟﾙｺ,しぶき,浜松,高崎,小児,二子,天然記念物,築地,仲居,難い,ﾗﾃﾝ語,ｼｮｳｶﾞ,家業,黒澤,滝沢,鳥取,京橋,春園,湯立,前橋,日夜,釣り上げる,ﾎｰﾑﾀｳﾝ,うどん,宮内,赤羽,弘樹,高山,山盛り,吉祥寺,南北,上梓,谷崎,潤一郎,佳苗,倦怠期,小春,ﾛｰｿﾝ,犬猿,再燃,科技,なめ,洋食,加勢,甲信越,歯肉,南国,稀少,語録,現存,伊勢,川田,大前,にぎわう,遺志,公募,ﾎﾟｰｸ,南北線,懇意,北陸,花園,年の瀬,ﾏﾄﾞﾝﾅ,靖国神社,土佐,ｱﾋﾙ,戸田,記帳,将棋,天神,青龍,目の敵,岡山,打ち込める,粉末,語源,落差,五反田,由利,慶子,ﾘﾆｱ,本通,new,守山,寝屋川,流儀,東海,ｸｼﾞ,奇異,縁談,着工,東口,駅名,宝塚,歩み,落ち目,鳥羽,南口,丁目,消しｺﾞﾑ,立食,丁度,徳島,必携,歯医者,軽蔑,ﾊﾞｯｸｽ,いかん,jr,釧路,津田沼,戯曲,文通,鮎川,略す,とまどう,よしのり,つり,のべる,ﾉｼﾞﾏ,出雲,脂肪酸,宇都宮,歯並び,ひばりヶ丘,茨木,上越,茅ヶ崎,市役所前,五日市,坂戸,朝霞,作りあげる,竹下,つくば,山道

Topic # 1 :
ﾃﾞｨﾚｸﾄﾘ,転載,引用,大城,相田,ｼﾁﾘｱ,反町,捕鯨,幸広,滝口,紫陽花,省吾,ひったくり,七福神,八神,東淀川,寝息,少林寺拳法,栗田,国屋,都電,根津,ｸｼﾞﾗ,谷中,ひったくる,博巳,孝雄,勇二,太一郎,千鳥足,荒川線,真山,釈放,あつみ温泉,谷根,竜治,求刑,下村,管轄,舞い戻る,目白,柴又,巡査,再演,座禅,起訴,寛永寺,叡山,高僧,金太郎,下町,呉服,鬼子母神,面影橋,雑司ヶ谷,日暮里,芭蕉,弘法大師,墨客,湯治,詩歌,参道,千駄木,恩賜,平家,万引き,義経,塩冶,佳織,野蛮,演目,召使い,敵討ち,御会式,三ﾉ輪,あじさい,音楽専科社,京成,毘沙門天,帝釈天,良観,大黒天,そぞろ歩き,花魁,藤永,しなる,一葉,高3,裏手,白虎隊,尊像,側壁,純然,寺町,武蔵野台,曼珠沙華,清輝,ｱﾙｶﾘ性,ｽﾊﾟｹﾞｯﾃｨ,海抜,左折,線路,学習院大学,鳥羽,沿岸,衛門,差し入れる,拘置,高師直,花街,神通力,女郎,頼朝,いびる,源義経,朝廷,鳴神,逃げ延びる,ばあさん,身請け,由良之助,菊之助,白浪,刃傷,残党,家臣,大星,判官,釣瓶,ﾊﾟﾅﾏ,生い茂る,ﾎﾟﾗﾛｲﾄﾞ,ﾐﾝｸｸｼﾞﾗ,領海,弁財天,ﾗｯｸ,手編み,毒薬,ﾂﾀ,いさむ,通いつめる,日本体育大学,追いやる,立ち去る,川中,始発駅,室町,与謝野,明大,無念,ﾀｶﾗ,脇道,布袋尊,福の神,寿老人,霊気,医王寺,宝生,正井,静寂,御朱印,経時,重々しい,従兄,いふ,名刀,大名,初夜,倣う,東海道,所帯,建立,制止,論告,極刑,上野公園,めんずる,文教堂,下車,隣家,神域,人中,巡り,振り切る,奥寺,突き,紙袋,ﾄﾋﾟｯｸ,ぺったんこ,花袋,公孫樹,腹ごしらえ,いちょう,武芳,露天,世田谷線,江戸城,沁みる,こっぴどい,手本,しし,皆殺し,葛飾,清める,自由,ひとたび,風情,寝言,番頭,栄える,龍神,鬼畜,亡霊,つづける,神出鬼没,町民,文人,貸し切り,畳む,波瀾万丈,お寺,安土,桃山,無期,直角,陳述,柴島,商人,盗賊,すみやか,ｲｽﾞﾑ,浜田,大声,小僧,一帯,地形,ﾃﾞｨｰﾗｰ,恐妻,寺社,しぼむ,北千住,封じ込める,至近,激励,千本,散策,捕れる,償う,武家,山手,こぎつける,けす,稲荷,校内,個体,晶子,境内,天然記念物,弁天,ｾﾋﾟｱ,絶世,善人,過ち,敷地,ﾊﾞﾚﾝﾀｲﾝﾃﾞｰ,喧騒,清らか,美少年,池内,路面,歌舞伎,家屋,ｳｫｰﾎﾙ,おえる,船越,うけあう,酸性,祀る,篠原,務め,本殿,松尾,三国志,位牌,上野,吉原,見殺し,餓死,早稲田,男子校,沙汰,切腹,ﾌｧｲﾙ,殺人,土壌,文彦,四谷,技術評論社,安産,通り抜ける,原型,鎌倉,味気,わる,武道,植える,ほとり,寺院,郷愁,田山,神宮,面影,お化け,沈没,黒田,雑居,逸話,辟易,悪人,疎む,だます,書き方,寝かしつける,赴任,旅館,庶民,廻る,推察,由緒,色紙,気軽い,公判,外回り,こる,ﾉﾍﾞﾙ,蟹江,ﾐｭｰｼﾞｶﾙ,忠臣蔵,ﾎﾋﾞｰ,路地,暴漢,異論,飢餓,踏み入れる,ｼｬｲ,屋台,内容,のろける,青山,呼び起こす,駄菓子,行き届く,隼人,ﾎﾟﾁ,満開,区域,怪談,樹齢,贈り物,うたう,近代,高等,江戸,街路,讃歌,下る,替る,敗北,直射,老いる,ﾓｰﾀｰｽﾎﾟｰﾂ,提供,樋口,作り話,普遍,企む,藤木,先々,情緒,閑静,団子,接触,替わり,博物館,愛人,娯楽,中尾,有す,田舎町,ﾒﾛﾃﾞｨ,思いがけない,ひと息,立ち並ぶ,ｽｰﾊﾟｰﾏﾝ,抜ける,学年,参拝,窮地,男の子,一戦,纏う,中古,木々,すう,破り,ﾘﾌｫｰﾑ,階級,乙女,自伝,災い,甘味,不穏,重厚,不二子,ﾁｬﾝﾈﾙ,活気,論争,演劇,交差,祈願,浅草,ｽﾘｯﾌﾟ,ﾊﾞﾚﾝﾀｲﾝ,無断,一期一会,さまよう,東急,台風,宿る,弱まる,可能,佇む,単身,傍ら,開運,並木,天下,ﾙｰﾄ,おじ,香緒里,通学,賑わう,洋画,警察,思い出,吹き飛ばす,佐藤,まわす,ﾚｷﾞｭﾗｰ,隆史,犯す,老若男女,ｼﾁｭｴｰｼｮﾝ,咲く,焼き鳥,気象,何事,ｲﾝｽﾀﾝﾄ,快眠,吉本興業,屋敷,指す,見下ろす,山形,買い取る,許諾,池田,色濃い,ﾄﾞﾗｯｸﾞ,明治,性行為,裕福,整備,焼きそば,来春,女装,掛ける,先ほど,容疑,一昨年,湿度,水面,共和,童貞,祝う,本業,立たす,真っ最中,転用,改変,妥当,もらう,体育館,白黒,懲役,身勝手,復刻,繰り出す,後押し,朝日新聞

Topic # 2 :
ふる,ぬいぐるみ,永野,つかさ,ｺﾏ,おはこ,載る,ﾊﾞﾝｸ,ｽｶｲﾗｲﾝ,征服,ｱｰｸ,うちわ,清志郎,ねる,おしらせ,釣れる,少林寺,なめこ,おし,ぞくす,暑中,振袖,僧侶,こびる,渓流,日和,ｶﾀﾞﾌｨ,薬丸,ﾎﾞｰｳﾞｫﾜｰﾙ,秘策,国生,忌野,結納,てんかん,亀田製菓,庭木,見舞い,江原,まんま,かくれる,上祐,嚢胞,耳かき,熊田,大浜,白浜,村主,出家,ｵﾘｼﾞﾅﾙｸﾞｯｽﾞ,月頃,ｻﾝｺｰ,曜子,心神,下田,冊子,卵巣,肋骨,弓ヶ浜,撃墜,日吉ｷｬﾝﾊﾟｽ,同志社大学,大阪城ﾎｰﾙ,裕英,ﾘﾋﾞｱ,本店,踏みつける,ﾅﾆﾜ,入田,吉佐美,たのしむ,慈悲,断絶,穢れる,太極,史浩,さいき,にじり寄る,せんせい,祇園,ちかい,磯浜,大佐,きゅう,どじ,にゅｰす,秀美,機先,いする,まける,南伊豆,しかける,巨木,ﾌﾞﾅ,ともだち,生活協同組合,日商,急所,大阪大,水曜,国交,腫瘍,じゅん,千香,おもう,生ずる,流せる,ｻﾘ,断交,一水会,仲人,背伸び,ｵﾙｸﾞ,礼装,入信,安らか,血みどろ,柿沢,仏教徒,改宗,新刊,哨戒,三菱重工業,大雄,祝す,あまた,長生,判例,雅弘,完売,目録,出来上がる,事故死,両家,こん,悪逆,悟道,南無阿弥陀仏,洋史,巣くう,いやしい,ﾋﾞﾗ,大阪大学,ｵｳﾑ真理教,顔合わせ,軟派,少林寺拳法,へちま,生垣,たすける,水難,出で立ち,ﾚﾝﾄｹﾞﾝ,大生,指宿,饗宴,大団円,九郎,ｳｸﾞｲｽ,仲井戸麗市,ﾎﾞﾃﾞｨｶﾞｰﾄﾞ,女帝,早稲田大,釣り針,よそよそしい,ひよこ,さえずり,雑多,河畔,せせらぎ,十和田,奥入瀬川,ｶﾞｰﾘｯｸ,廻り,ほど近い,重罰,食い違う,凶器,赤信号,せいこう,墓石,邦人,ﾄﾘﾎﾟﾘ,評議,供与,家主,つけ込む,用法,召使い,産みの親,家内,さわり,検診,ﾄｹﾞ,雪どけ,触診,天王,建材,左官,山門,図上,修める,苛烈,嵩山,拳法,丸太,庭師,鉄工,京劇,仕る,棟梁,達磨,国共,菩提,禅宗,木工,留袖,配布,花見,つば,ﾑｸﾞﾗ,勉学,これみよがし,礼服,勝男,こんぶ,子生,口上,末広がり,縁起物,麻糸,縁組み,ながの,びわ,熨斗,しらが,祝い酒,長もち,ならわす,喜多,春休み,駒場,他校,島国,受け付け,景勝,しきたり,剪定,獅童,ｵｼﾄﾞﾘ,振り袖,ﾗｳ,和尚,方丈,ﾕｴ,腹心,只者,弟弟子,洋輔,工費,子弟,若僧,ｺｰﾘｰ,撃ち殺す,生い茂る,ｾｰﾌﾃｨ,身幅,ﾐｯｷｰﾏｳｽ,藤崎,大倉,花やしき,いつみ,立川談志,ﾏﾄﾓ,寛平,白無垢,魚類,たこやき,へび,いわし,ざりがに,たいよう,おとす,まんぼう,おとうさん,むし,かじき,秘法,右派,ﾊﾞﾘｶﾝ,湯のみ,石工,不老,座布団,みつかる,ｵｳﾑ,伝搬,さめる,釣り場,てんとう,国民党,腹痛,ひろう,ごちそうさま,だま,河南,死傷,湖畔,抜き,えん,ﾕﾈｽｺ,まち,ちる,ｸﾞｯｽﾞ,激昂,顕子,ｱﾍﾝ,たか子,木漏れ日,ふろ,号館,厳か,さかな,密売,中腰,出立,婚礼,平安,いす,いるか,和装,とうき,空爆,進行,大仏,会館,電柱,啓之,北関東,ｱﾜﾋﾞ,簿記,学術,憲法,直談判,ｻｲﾝ,不仲,介添え,祝祭日,すっごい,荘厳,海岸,振りまく,弱虫,鍛錬,偽り,おいら,合わさる,血まみれ,特大,地色,祭り,せん,ﾊﾞﾝ,取り持つ,封建,授ける,月曜日,秘蔵,篠山,黒地,ふたつ,壁紙,邂逅,納品,病名,作り物,ｺｰﾄﾞﾚｽ,ﾓｽｸﾞﾘｰﾝ,配備,ﾎﾟｽﾄ,土足,釣り竿,流暢,混同,腐れ縁,同志社大,来週,突き出す,みる,拝む,愚弄,硬派,独り占め,ｷﾞｸｼｬｸ,ｶﾞｰﾃﾞﾝ,風習,略式,ぶん,返還,渾然一体,花婿,文字数,連休,百合子,くも,たか,すえ,思い知る,でかける,清らか,取り扱う,昼過ぎ,宝島,fifa,双子,来客,内戦,ずるい,足首,枯れる,共闘,作者,掲示,お盆,町内,ﾛｯｸﾝﾛｰﾙ,床や,鰹節,標榜,ﾌﾞﾗﾝｺ,ﾎｰﾑｾﾝﾀｰ,床の間,かに,見もの,耽美,一足,来月,訊ねる,噴水,ｽﾀﾝﾄﾏﾝ,秋葉原,正広,らん,予告編,貿易,静岡,持病,大江戸,爆撃,健一郎,なぎ倒す,友好,ごい,仏教,顕在,ﾒﾃﾞｨｱ,芝生,ﾐｯｷｰ,こむ,要する,信者,尊い,ｼｬﾝ,毎週,民度,特集,いじる,ﾐﾆ,堅苦しい,海辺,付け替える,下調べ,ﾃﾗｽ,くびれる,四方,企画,扇子,病みつき,茂木,繁栄,秋葉,慎太郎,樹齢,武術,異彩,和室,親交,色直し,ﾀｵﾙ,そびえる,仁義,正念場,ﾌｰ,子宝,みどころ,引き続く,お好み焼き,進出,昆布,ｽﾍﾟﾙ,くらう

Topic # 3 :
する,なる,ある,できる,映画,思う,日本,話題,女性,ない,記事,発売,見る,いう,ｽﾏｰﾄ,機能,対応,ﾌｫﾝ,関連,発表,言う,情報,いる,多い,ﾈｯﾄ,写真,公開,使う,ﾓﾃﾞﾙ,結婚,利用,搭載,ﾄﾞｺﾓ,画面,世界,韓国,行う,女子,監督,ｻｰﾋﾞｽ,男性,更新,紹介,仕事,作品,ｻｲﾄ,いい,可能,番組,人気,登場,撮影,ﾕｰｻﾞｰ,ﾁｪｯｸ,持つ,表示,製品,放送,ｶﾒﾗ,選手,ﾃﾚﾋﾞ,やる,ｹﾞｰﾑ,知る,向け,動画,開始,代表,東京,価格,好き,映像,ｿﾌﾄｳｪｱ,相手,予定,高い,考える,ｼﾘｰｽﾞ,設定,ﾆｭｰｽ,端末,ﾌｧﾝ,ﾋﾞﾃﾞｵ,語る,出る,ｺﾒﾝﾄ,応募,販売,必要,感じる,電話,提供,開発,掲示板,恋愛,聞く,ｷｬﾝﾍﾟｰﾝ,ﾒｰﾙ,最大,ｲﾍﾞﾝﾄ,行く,試合,機種,現在,購入,通信,注目,公式,問題,開催,会社,使用,わかる,しれる,本体,良い,ﾘﾝｸ,商品,転職,結果,今年,ｻｲｽﾞ,新しい,内容,ﾌﾟﾚｾﾞﾝﾄ,ﾓﾊﾞｲﾙ,ﾄﾞﾗﾏ,選ぶ,作る,時代,ﾊﾞｯﾃﾘｰ,充電,ﾊﾟｿｺﾝ,ﾀﾌﾞﾚｯﾄ,出演,入る,最新,無料,ｻｯｶｰ,簡単,読む,気持ち,編集,操作,受ける,演じる,強い,生活,ﾃﾞｻﾞｲﾝ,楽しむ,容量,採用,ﾃﾞｼﾞﾀﾙ,特集,携帯,ﾍﾟｰｼﾞ,ﾃﾞｨｽﾌﾟﾚｲ,ｱｯﾌﾟ,接続,友達,ﾀﾞｳﾝﾛｰﾄﾞ,ｶｰﾄﾞ,自身,視聴,ﾃﾞｰﾀ,当選,売れ筋,ｸﾘｽﾏｽ,野球,すごい,ﾁｰﾑ,ｲﾝﾁ,理由,ｿﾌﾄﾊﾞﾝｸ,話す,電子,連続,発言,子供,便利,入れる,使える,ｼｰﾝ,ｱﾒﾘｶ,全国,食べる,料理,変わる,言葉,執筆,限定,決定,画像,状態,分かる,違う,ﾌｧｲﾙ,大きい,始める,配信,増える,一番,海外,意見,部分,ｾｯﾄ,最近,家族,魅力,美しい,楽しい,確認,五輪,参加,ｹｰｽ,ﾌﾟﾛ,描く,ﾎﾟｲﾝﾄ,社長,批判,人間,液晶,見える,記録,ﾀｲﾌﾟ,今後,伝える,登録,音楽,ｱｯﾌﾟﾃﾞｰﾄ,入力,期待,一緒,実施,観る,寄せる,ｱｲﾃﾑ,経験,呼ぶ,続く,調査,企業,ｲﾒｰｼﾞ,種類,期間,報道,変更,おすすめ,楽しめる,見せる,よる,ｿﾆｰ,最後,出す,書く,ｽﾎﾟｰﾂ,ｺﾞﾙﾌ,ﾌﾞﾗﾝﾄﾞ,画素,場所,効果,活躍,悪い,機器,ｾﾞﾛ,ﾒｰｶｰ,人生,悩み,活動,書籍,環境,ｺﾝﾃﾝﾂ,ｶﾗｰ,展開,意味,女優,ｼｮｯﾌﾟ,ﾗﾝｷﾝｸﾞ,国内,ｲﾝﾀｰﾈｯﾄ,つく,ﾌｧｯｼｮﾝ,大切,家電,技術,かける,防水,主演,早い,時期,ﾒﾃﾞｨｱ,買う,ｻﾎﾟｰﾄ,起動,飲む,様子,ﾒﾝﾊﾞｰ,実現,質問,昨年,皆さん,社会,予約,終了,存在,世代,つける,かわいい,ｱｯﾌﾟﾙ,専用,難しい,大人,会場,注意,言える,出来る,撮る,合わせる,ﾎﾞﾀﾝ,一部,変える,幸せ,嬉しい,印象,状況,ﾊﾟﾅｿﾆｯｸ,ｽﾀｰﾄ,ｺｱ,解説,株式会社,よい,ﾊﾞｰｼﾞｮﾝ,特別,高速,決める,通り,友人,ｸﾘｯｸ,店舗,ﾃｰﾏ,ﾀｯﾌﾟ,ｷｰ,明かす,今日,企画,答える,市場,日本人,男子,お知らせ,追加,体験,取る,来る,ｷｬﾗｸﾀｰ,投稿,中国,過去,対象,音声,契約,ﾄｯﾌﾟ,検索,以外,記念,ﾌｼﾞﾃﾚﾋﾞ,ｲﾝﾀﾋﾞｭｰ,掲載,舞台,展示,選択,ｼｽﾃﾑ,ｺｰﾄﾞ,続ける,ｽﾏ,少ない,集まる,働く,求人,報じる,部屋,最初,無い,ｵｽｽﾒ,原因,いく,ｵｰﾌﾟﾝ,説明,ｼﾞｬﾊﾟﾝ,主人公,意識,ﾒﾓﾘｰ,年収,かかる,獲得,若い,物語,ｱｸｼｮﾝ,無線,集める,用意,子ども,面白い,個人,みる,ﾒﾆｭｰ,求める,成長,影響,加える,電池,ｼｰｽﾞﾝ,連絡,送る,ｶﾃｺﾞﾘ,ﾉｰﾄ,詳細,向ける,再生,一般,ﾓｰﾄﾞ,活用,上映,劇場,その他,与える,終わる,ﾛｺﾞ,ｹｱ,相談,行動,対策,生きる,基本,特徴,事件,ｽﾄｰﾘｰ,挑戦,いかが,文字,大会,ﾛﾝﾄﾞﾝ,探す,芸能,ﾚﾎﾟｰﾄ,中心,練習,お気に入り,優勝,最終,始まる,香川,快適,ｷｰﾎﾞｰﾄﾞ,ｽﾀｲﾙ,なでしこ,務める,ﾋｯﾄ,ｼｬｰﾌﾟ,ﾋﾞｼﾞﾈｽ,出場,募集,ｱｶｳﾝﾄ,通話,電源,従来,頑張る,営業,ｿﾌﾄ,絶対,野村,ﾎｰﾑ,自然,お金,社員,迎える,ｵﾘｼﾞﾅﾙ,含む,非常,管理,ｱｸｾｽ,長い,最高,上がる,予想,振り返る,最強,くる,受賞,回答,占い,ﾘﾘｰｽ,ﾌﾞｯｸ,押す,年間,異なる,連載,教える,国際,漫画
(bachelor) user@MacBook-Pro bachelor % 

lunar123456.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar5.py header
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

#lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
# print(tfidf_data.toarray()) 
# print(tfidf_data.shape)

# lunar5.py
def gridsearch_best_model(tfidf_data, plot_enabled=True):
    # Define Search Param
    n_topics = [4,5,6,7,8,9]
    search_params = {'n_components': n_topics}
    
    # Init the Model
    lda = LDA(max_iter=25,               # Max learning iterations
              learning_method='batch',   
              random_state=0,            # Random state
              n_jobs = -1,               # Use all available CPUs)
              )
    
    # Init Grid Search Class
    model = GridSearchCV(lda, param_grid=search_params)
    
    # Do the Grid Search
    model.fit(tfidf_data)
    
    # Best Model
    best_lda_model = model.best_estimator_
    
    # Model Parameters
    print("Best Model's Params: ", model.best_params_)
    
    # Log Likelihood Score
    print("Best Log Likelihood Score: ", model.best_score_)
    
    # Perplexity
    print("Model Perplexity: ", best_lda_model.perplexity(tfidf_data))
    
    # Get Log Likelyhoods from Grid Search Output
    log_likelyhoods_score = [round(score) for score in model.cv_results_["mean_test_score"]]
    
    if plot_enabled:
        # Show graph
        plt.figure(figsize=(12, 8))
        plt.plot(n_topics, log_likelyhoods_score)
        plt.title("Choosing Optimal LDA Model")
        plt.xlabel("Numer of Topics")
        plt.ylabel("Log Likelyhood Scores")
        plt.show()

    return best_lda_model
best_lda_model = gridsearch_best_model(tfidf_data)

# lunar6.py
def print_topics(model, count_vectorizer, n_top_words):
    fig = plt.figure(figsize=(15,8))
    words = count_vectorizer.get_feature_names()
    for topic_idx, topic in enumerate(model.components_):
        print("\nTopic #", topic_idx, ":")
        long_string = ','.join([words[i] for i in topic.argsort()[:-n_top_words - 1:-1]])
        print(long_string)
        topic_wordcloud(topic_idx, fig, long_string)
    # show plots
    fig.tight_layout()
    fig.show()

def topic_wordcloud(topic_idx, fig, long_string):
    ax = fig.add_subplot(2, 3, topic_idx + 1)
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    wordcloud.generate(long_string)
    ax.imshow(wordcloud)
    ax.set_title('Topic '+str(topic_idx))

number_words = 500
print_topics(best_lda_model, count_vectorizer, number_words) 

lunar1234567.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar5.py header
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

#lunar1.py
text_paths = glob.glob('data/text/**/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    text = text.split('\n')
    title = text[2]
    text = ' '.join(text[3:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
# print(tfidf_data.toarray()) 
# print(tfidf_data.shape)

# lunar5.py
# def gridsearch_best_model(tfidf_data, plot_enabled=True):
def gridsearch_best_model(tfidf_data, plot_enabled=False):
    # Define Search Param
    n_topics = [4,5,6,7,8,9]
    search_params = {'n_components': n_topics}
    
    # Init the Model
    lda = LDA(max_iter=25,               # Max learning iterations
              learning_method='batch',   
              random_state=0,            # Random state
              n_jobs = -1,               # Use all available CPUs)
              )
    
    # Init Grid Search Class
    model = GridSearchCV(lda, param_grid=search_params)
    
    # Do the Grid Search
    model.fit(tfidf_data)
    
    # Best Model
    best_lda_model = model.best_estimator_
    
    # Model Parameters
    print("Best Model's Params: ", model.best_params_)
    
    # Log Likelihood Score
    print("Best Log Likelihood Score: ", model.best_score_)
    
    # Perplexity
    print("Model Perplexity: ", best_lda_model.perplexity(tfidf_data))
    
    # Get Log Likelyhoods from Grid Search Output
    log_likelyhoods_score = [round(score) for score in model.cv_results_["mean_test_score"]]
    
    if plot_enabled:
        # Show graph
        plt.figure(figsize=(12, 8))
        plt.plot(n_topics, log_likelyhoods_score)
        plt.title("Choosing Optimal LDA Model")
        plt.xlabel("Number of Topics")
        plt.ylabel("Log Likelyhood Scores")
        plt.show()

    return best_lda_model
best_lda_model = gridsearch_best_model(tfidf_data)

# lunar6.py
def print_topics(model, count_vectorizer, n_top_words):
    fig = plt.figure(figsize=(15,8))
    words = count_vectorizer.get_feature_names()
    for topic_idx, topic in enumerate(model.components_):
        print("\nTopic #", topic_idx, ":")
        long_string = ','.join([words[i] for i in topic.argsort()[:-n_top_words - 1:-1]])
        print(long_string)
        topic_wordcloud(topic_idx, fig, long_string)
    # show plots
    fig.tight_layout()
    fig.show()

def topic_wordcloud(topic_idx, fig, long_string):
    ax = fig.add_subplot(2, 3, topic_idx + 1)
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    wordcloud.generate(long_string)
    ax.imshow(wordcloud)
    ax.set_title('Topic '+str(topic_idx))

number_words = 500
# print_topics(best_lda_model, count_vectorizer, number_words)

# lunar7.py
test_data_ss = pd.Series(['このクリームの成分', 'このパソコンの性能が良い'])
processed_test_data_ss = preprocess_jp(test_data_ss)
test_count_data = count_vectorizer.transform(processed_test_data_ss)
test_tfidf_data = tfidf_vectorizer.transform(test_count_data)
doc_topic_mat = best_lda_model.transform(test_tfidf_data)
dominant_topic = np.argmax(doc_topic_mat, axis=1)

test_data_df = pd.DataFrame(test_data_ss, columns=['text'])
test_data_df['topic_id'] = dominant_topic
display(test_data_df) 

ocu1.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar1.py
text_paths = glob.glob('data/ocu2/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    # text = text.split('\n') # modified
    text = text.split(',')
    title = text[3] # added
    # title = text[2] # modified
    text = ' '.join(text[8:])
    texts.append(text)
    
news_ss = pd.Series(texts)
display(news_ss.head()) 
# display(news_ss) 

ocu12.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob
# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar1.py
text_paths = glob.glob('data/ocu2/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    # text = text.split('\n') # modified
    text = text.split(',')
    title = text[3] # added
    # title = text[2] # modified
    text = ' '.join(text[8:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 
# display(news_ss) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
display(processed_news_ss.head())
# display(processed_news_ss)

ocu123.py

# フォントの入手元 https://moji.or.jp/ipafont/ipafontdownload/
# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar1.py
text_paths = glob.glob('data/ocu2/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    # text = text.split('\n') # modified
    text = text.split(',')
    title = text[3] # added
    # title = text[2] # modified
    text = ' '.join(text[8:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 


# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

show_wordcloud(processed_news_ss) 

ocu1234.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar1.py
text_paths = glob.glob('data/ocu2/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    # text = text.split('\n') # modified
    text = text.split(',')
    title = text[3] # added
    # title = text[2] # modified
    text = ' '.join(text[8:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
print(tfidf_data.toarray()) 
print(tfidf_data.shape)

ocu12345.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar5.py header
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

#lunar1.py
text_paths = glob.glob('data/ocu2/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    # text = text.split('\n') # modified
    text = text.split(',')
    title = text[3] # added
    # title = text[2] # modified
    text = ' '.join(text[8:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
# print(tfidf_data.toarray()) 
# print(tfidf_data.shape)

# lunar5.py
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

def gridsearch_best_model(tfidf_data, plot_enabled=True):
    # Define Search Param
    # n_topics = [4,5,6,7,8,9]
    n_topics = [8,9,10,11,12,13]
    search_params = {'n_components': n_topics}
    
    # Init the Model
    lda = LDA(max_iter=25,               # Max learning iterations
              learning_method='batch',   
              random_state=0,            # Random state
              n_jobs = -1,               # Use all available CPUs)
              )
    
    # Init Grid Search Class
    model = GridSearchCV(lda, param_grid=search_params)
    
    # Do the Grid Search
    model.fit(tfidf_data)
    
    # Best Model
    best_lda_model = model.best_estimator_
    
    # Model Parameters
    print("Best Model's Params: ", model.best_params_)
    
    # Log Likelihood Score
    print("Best Log Likelihood Score: ", model.best_score_)
    
    # Perplexity
    print("Model Perplexity: ", best_lda_model.perplexity(tfidf_data))
    
    # Get Log Likelyhoods from Grid Search Output
    log_likelyhoods_score = [round(score) for score in model.cv_results_["mean_test_score"]]
    
    if plot_enabled:
        # Show graph
        plt.figure(figsize=(12, 8))
        plt.plot(n_topics, log_likelyhoods_score)
        plt.title("Choosing Optimal LDA Model")
        plt.xlabel("Number of Topics")
        plt.ylabel("Log Likelyhood Scores")
        plt.show()

    return best_lda_model
best_lda_model = gridsearch_best_model(tfidf_data) 

ocu123456-run.txt

(bachelor) user@MacBook-Pro bachelor % python3 ocu123456.py
File already exists.
Best Model's Params:  {'n_components': 10}
Best Log Likelihood Score:  -57286.863089259525
Model Perplexity:  3495.2774341479862

Topic # 0 :
ﾙｰﾌﾟ,検知,泉井,ﾀﾞｳﾝ,閉塞,ｼｬｯﾄ,配下,眞佐子,くん,森本,外科,歯科,口腔,大黒,自動,泰成,幸田,佐賀,筑波大,ｱｰｶｲﾌﾞ,赴任,ｾﾐﾅｰ,毎朝,部内,高井,滋賀大学,飛鳥,ﾄｯﾌﾟﾍﾟｰｼﾞ,解除,数学,予備,切替える,余裕,ｽｷｬﾅ,附属,不備,ﾍﾟｲﾝ,ﾄﾞｸﾀｰ,つかえる,柳本,井口,復帰,溝上,記事,ﾒﾝﾊﾞ,はいれる,不正,ｸﾛｰﾑ,古賀,ｿﾌﾄﾊﾞﾝｸ,後援,達也,問合せ,静愛,直樹,もとづく,ﾊｲﾌﾞﾘｯﾄ,ｹﾞｽﾄﾊｳｽ,石神,給付,正式,光成,買い替える,亀田,小川,ﾔﾏﾓﾄ,ﾏｷﾀ,生里,始業,平素,偏る,聞き出す,植松,たける,分ける,心当たり,好ましい,ｳｲﾝﾄﾞｳ,小塚,無給,授与,模様,倫理,去年,ゆるい,関田,ｽﾄﾚｰｼﾞ,幹線,赤色,ﾏｯｸ,思い違い,高塚,いじる,さし,折れる,少数,ﾋﾟｸｾﾙ,正方形,枝里子,中台,あむ,免疫,配付,遅れる,割り込み,国外,直前,ｹﾞﾉﾑ,気づく,ｲﾝﾁ,きたす,引ける,西浦,企業,しれる,勤怠,中島,洋司,浅井,作動,剛史,所定,もつ,ずれ込む,取り込める,ﾌｧｲﾙｻｲｽﾞ,湯口,大樹,裕朗,玉井,美香,競合,返送,ｳｴﾀﾞ,知幸,とれる,ども,宮本,左記,かなう,邪魔,おこなう,推定,お気に入り,抜染,まずい,刺す,ﾊﾟｰﾄ,ﾏﾝﾂｰﾏﾝ,冨澤,あおい,ﾎﾞｽ,ﾋﾟｰｸ,基板,未帆,ｽﾀﾝｽ,望ましい,晴恵,果たす,部門,臨地,結合,ｶﾜｻｷ,肩書,迅速,埋める,差し込む,伏せる,勝手,解る,受け持つ,ｶﾗ,ともなう,ﾀｲ,お陰様,ﾎﾟｲﾝﾀｰ,ﾍﾟﾝ,貰える,紘子,老年,逼迫,立て込む,特筆,悩み,ｺﾞｰﾙﾃﾞﾝｳｨｰｸ,みあたる,繋がり,出せる,加工,生産,直接ｱｸｾｽ,ｴﾝｼﾞﾆｱ,給湯,同居,正行,公共,無知,慎重,一室,組み換え,早朝,弊社,退社,重視,時差,全課,国文学,糀谷,多恵子,代打,混雑,三宅,件数,国語,端子,枠組み,開放,取り消し,高齢,新調,及ぶ,旅行,鹿毛,まち,手間取る,引用,右往左往,ﾍﾞｽﾄ,夏期,近畿大学,作り,うる,加筆,なおこ,うの,書き直し,ﾊﾞｯｸｽ,合わさる,先だって,潜在,中止,明記,ﾍﾟｰｽ,余計,遮断,存じ上げる,締め切り,明細,ｵﾝ,目標,病院,受入れ,弘之,支社,ﾛｹｰｼｮﾝ,備わる,統括,曰く,2月,お断り,図れる,達成,学び,関わり,実践,4つ,ｱｶ,ｺﾐｭﾆｹｰｼｮﾝ,最低限,日数,目安,開講,独立,構成,うえ,探し出す,回数,観察,我慢,取り外す,取り除く,今津,しなす,途切れる,転記,たどり着く,立てる,繰り返し,申告,広報,普段,分かれる,とる,問いあわせる,八千代,日時,ﾀｸﾞ,静的,ご多忙,連続,切り替える,学位,ｽﾙｰ,立て続け,不可欠,掃除,9月,浮く,とまる,上位,整理,益田,受け入れる,松浦,取りまとめ,抜き,石原,ｽﾄｰﾑ,項目,持ち帰り,番号,理系,のる,専任,深刻,文系,山尾,重度,株式会社,ﾌﾞﾛｰﾄﾞｷｬｽﾄ,許可,重なる,ﾎｰﾙ,防ぐ,時刻,お手伝い,満つ,向かう,えり,ｾｯｼｮﾝ,人間,研究所,答える,福祉,台風,川合,再発,むける,広岡,連動,文教,兼ねる,上限,山岡,下さる,換え,ﾘｱﾙﾀｲﾑ,事情,抜く,受入,宜しい,医学部,かす,速やか,羽下,由紀子,上昇,打ち合わせる,ｷｭｰ,言う,埋め込む,形跡,組む,規模,信頼,頻度,半分,脅威,不便,実習,低下,ｻｲｽﾞ,構造,当課,変える,石橋,指名,亮輔,多発,差す,ｼﾞｬｰﾅﾙ,則る,中央,ご存知,解消,相当,止まる,疑う,無題,あわせる,要因,主催,ﾊﾞｯｸ,保障,終了,いる,ｶｰｿﾙ,休暇,ｽｲｯﾁ,超過,添付,ﾌﾛｱ,安定,操作,早々,弾く,整形,購読,超える,ﾃﾞｰﾀ,ﾙｰﾑ,時半,おりる,けんか,だける,忠雄,差し替える,振替,残念,池端,ｺﾝ,引継ぎ,ﾐｽ,環境,判断,引き続く,保持,上述,ﾀﾞｳﾝﾛｰﾄﾞ,点く,差し上げる,講座,充てる,富士ｾﾞﾛｯｸｽ,親子,笠川,配線,佐藤,掴む,書き込み,ｺﾝｾﾝﾄ,申し上げる,差替え,ｸｲｯｸ,電話,拒否,制度,控える,お世話,産業,ﾒﾓﾘ,返る,詳細,先程,なし,西口,ﾎﾟｰﾄ,大変,女性,裏手,谷川,客員,過ぎ,含める,川上,合わせ,ﾄﾞﾗｯｸﾞ,やり取り,物理,どなた,ｷｰﾎﾞｰﾄﾞ,ﾊﾞｸﾞ,程度,監視,営業,阿部,先生,張る,ｼｮｰﾄｶｯﾄ,ﾀﾞｲﾊﾞｰ,酒井,ﾏｳｽ,分配,支援,上田,直後,厳しい,目的,途中,ﾒﾃﾞｨｱ

Topic # 1 :
記念,聡美,原田,晴美,中野,村上,固まる,植物,数式,英数字,高原,全角,白い,美緒,選考,ﾊｲﾌﾝ,ﾎｰﾙ,小文字,広岡,京都大,発信,株式会社,渡辺,持田,大文字,附属,ｴﾃﾞｨﾀｰ,半年,一志,江川,ﾌﾞﾗｯｸ,論文,ｱｯﾌﾟﾙ,ﾒﾘｯﾄ,折返し,中田,貸出,ｽﾎﾟｰﾂ,森下,小畑,理紗,森内,こみあう,敏之,混雑,黒文字,茶畑,里奈,神山,森原,小泉,題名,ｴﾝﾄﾞ,分類,各社,ﾛｰﾏ字,耕一,ﾍｯﾀﾞ,絞る,ﾃﾞﾒﾘｯﾄ,渡し,県立,兵庫,思い出す,代行,保留,ﾋｯﾄ,前日,中原,課外,立ち会い,和人,ｲﾍﾞﾝﾄ,菅原,流出,ついで,海渡,一昨年,対面,払出し,澄子,反対,蓄積,直接ｱｸｾｽ,結線,誕生,上位,古澤,ｽｲｯﾁ,橋本,委譲,地名,礼実,簡易,廣川,法医学,集線,依子,西脇,岩熊,譲り受ける,職人,ﾋﾞﾙ,駅前,祐加,奈緒,見え,紛失,切れ,透過,取り,宏子,英子,いづみ,おちる,まちがい,とたん,検出,立ち会う,寄る,出かける,ﾉｳﾊｳ,ﾊﾟｯﾁ,雨宮,打つ,ﾀｲﾌﾟ,号機,角南,貴司,江原,省一,協定,藤原,臨む,帳票,ﾀｰﾑ,淑子,12月,貯まる,折れる,ｷｭｰ,しなす,差し,配慮,つめ,統括,ひさし,助ける,日経新聞,ﾘﾀｰﾝ,ﾊﾞｯﾌｧ,高所,ｴｱ,ｲﾝﾃﾘｼﾞｪﾝﾄ,ｽﾃｰｼｮﾝ,集計,部門,秘匿,ｺﾂ,範囲,まわす,浩志,問診,屋外,郁朗,確度,訪れる,おしらせ,高校生,差し込み,佳奈,坪田,救出,岩見,達也,哲治,火曜日,数個,ｼﾓ,ﾅｶ,森脇,簡素,寛人,間に合う,労使,乗り換え,だいたい,高校,来客,次週,学術情報ｾﾝﾀｰ,振興,昭子,寺岡,話し合い,江梨,半角,離任,新設,渡す,来年,ｺﾐｭﾆｹｰｼｮﾝ,ゆくゆく,水曜,文句,関する,困窮,要項,狩俣,手つかず,募集,みなす,千博,選任,日比野,ﾎﾜｲﾄ,老朽,終業,間際,今日,給付,ｺｽﾄ,分野,秀之,退出,競合,放つ,高齢,読み取る,桃子,統一,行宏,玉造元,天王寺,共栄,つとめる,弁理,商標,特許,十分,最下,大全,あま,対抗,見当,甘い,適う,忘れ物,落とし物,ﾊﾞﾗﾊﾞﾗ,真弓,阿呆,見極め,増田,ﾌﾞﾗｯｸﾘｽﾄ,市川,安価,編入,はな,絢子,半月,在住,騒ぎ,関東,お手上げ,離宮,ﾊﾞｽ,ﾄﾞﾚｽ,いか,おそい,性格,持ち運び,実態,釈然,疑う,ﾒｲﾙ,まちまち,用事,充分,兼用,ﾍﾟｰｽ,直人,閉館,入館,折る,半日,労働,置き場,取り組む,広場,付近,脱字,約束,ｻﾝﾄﾞ,裁量,要す,受け取り,知らせる,誤字,篤い,うかがう,皆さん,学ぶ,痛感,学業,学友,例える,測定,宇田,専門,中央大学,持ち出し,お陰様,究明,ｶｰｿﾙ,拒絶,まじえる,位置づけ,迅速,逸脱,国内外,思いつく,空港,正史,押せる,在学,話し合う,専修,防げる,誤判,下回る,ﾄｰﾀﾙｼｽﾃﾑ,当てはまる,小野,隠蔽,容認,取り掛かる,ｶｽﾀﾏｲｽﾞ,的確,深謝,空欄,保証,ける,保守,基準,妨害,内外,書き直す,煩雑,知識,ﾄﾞｷｭﾒﾝﾄ,面す,右手,新旧,ｶﾞﾗｽ張り,正門,企業,いっさい,入口,同等,地図,ひとこと,示唆,黙認,再考,ていねい,まかる,抱える,ゆえ,全国,止む,粘る,いきる,負える,並行,いえる,予測,ﾕﾆ,検疫,ﾛﾌﾄ,顧客,同席,細長い,挟む,地区,去年,部室,事務所,ﾄﾚｰｽ,予防,嬉しい,強度,発表,入り口,安心,開く,価格,記号,試みる,なおす,建築,うける,ﾙｰﾄ,裏口,乗り換える,通行,在る,矢印,金額,赤字,試用,休日,根拠,万全,実状,ﾃﾞｰ,自室,なくす,洋一,変化,動的,ﾓﾃﾞﾙ,義成,遮断,騙る,ｴﾃﾞｨｼｮﾝ,緊急,消せる,詐称,一点,平成,走る,つかえる,等々,模索,しな,辺り,投げる,転記,願い,書架,ｹｶﾞ,引っ掛ける,横切る,ざま,うに,騒がせる,再送,取り直す,現行,上部,隔離,判定,漏洩,私用,出力,菜々,年末,移る,今週,松岡,制約,ｼｪｱ,個々,竹内,もしか,外部,回線,うつす,最長,受け取れる,置き換える,中山,皆様,回収,破損,有無,待ち,呼ぶ,ﾌﾟﾛ,救命,ﾎﾟｯﾌﾟｱｯﾌﾟ,つなげる,お詫び,起こす,になう,不手際,まんが,ｶﾞｲﾀﾞﾝｽ,ﾒﾝﾊﾞ,診断,来れる,映す,図書館,木曜日,搭載,久しぶり,名奈,取り扱う,健康

Topic # 2 :
回収,無線,希望,ｺﾐｭﾆﾃｨ,返却,代替え,不調,篠原,週間,ﾌｧｰﾑｳｪｱ,小久保,芝野,伺い,豊田,坂田,順番,程度,留意,機種,ｲﾉﾍﾞｰｼｮﾝ,緊急,除く,山中,前後,法学部,眞田,場所,玉巻,清美,貸出,健太,代替,太朗,最新,諸々,明日香,竜馬,視聴,兼田,永友,頂戴,真衣,防犯,ｴﾃﾞｨｼｮﾝ,要因,知子,玲奈,品切れ,大友,設置,適応,職種,盛田,事項,板谷,ﾒｰｶｰ,福谷,増設,附属,万が一,記入,早智子,部屋,速度,水野,大江,恵子,進捗,上位,たずねる,美幸,限り,ﾊﾝｸﾞ,心理,っぱなし,試験,送れる,信一郎,野本,脆弱,ﾚﾝｼﾞ,荻尾,阪大,発表,ﾊﾞｰｼﾞｮﾝ,弘美,適用,久保,松浦,断る,小関,彰一,広瀬,ﾊﾟｯﾁ,臨床,ｱｯﾌﾟﾃﾞｰﾄ,光永,松井,正治,業務,主張,始め,和文,石崎,唐沢,特殊,川井,つき,光成,つけ,選べる,熊田,文書,分配,本部,広美,富士ｾﾞﾛｯｸｽ,誰か,ﾌﾛｱ,ﾌﾞﾛｰﾄﾞｷｬｽﾄ,取り換える,私用,知香,ｲﾝﾀｰﾝ,倫子,不要,前半,ｸﾗ,ｳﾄﾞ,福田,経験,巻き,ｵｰﾌﾟﾝ,時刻,周り,持ち合わせる,巻く,美智,規程,改正,過日,後々,弓削,良子,米国,学長,くん,二郎,ﾅｶﾞｻｷ,点く,赤字,森山,越し,ｽﾄｰﾑ,鳥丸,うかがう,ｼｮｰ,同居,付近,櫻田,初期,村田,欲しい,第三者,高石,飛ぶ,木曜,ﾌｧﾝ,分岐,同等,話し合い,応答,実咲,柏山,質疑,放つ,ﾄﾞｷｭﾒﾝﾄ,やり方,止める,彰宏,就職,波及,組む,便覧,弊害,日数,続く,つたえる,真未,ﾙｰﾃｨﾝｸﾞ,ﾌﾟﾛｼﾞｪｸﾀ,事務所,ﾌﾟﾗｸﾞ,かえる,短時間,展子,太田,増強,壊れる,途切れる,調整,張り替える,武井,小畠,同行,拝聴,評価,汐見,石丸,政策,学期,遅れる,在外,小田中,引き渡し,ｶｸ,文献,祐太郎,疑い,借りる,如何,祥行,福島,松下,従事,踏襲,見当,ます,来年,小生,割り当て,社員,開講,知宏,id,手近,澤田,健康,智子,達哉,上村,ﾃﾞｨｽﾌﾟﾚｲ,滞在,かざす,成績,向ける,ﾚｼﾞｽﾄ,ｺﾐｭﾆﾃｨｰ,むら,復帰,ﾓﾄ,さす,上席,峰子,正則,ﾘｺｰ,拠点,追伸,康平,ｺﾝﾀｸﾄ,ｽﾛｯﾄ,つかめる,立ち会い,最短,基子,崩れる,採点,和木,流量,働きかける,並べる,またがる,車両,ｺﾝ,受け,止まる,思い込む,ng,色々,ｶｳﾝﾀ,過程,数多い,安川,眼科,ｷｬﾋﾞﾈｯﾄ,櫛野,ｵﾘｼﾞﾅﾙ,もどる,遷移,直之,頼む,差異,ﾅｵ,北山,萩尾,休止,変る,市川,破損,ｽﾋﾟｰｶ,乖離,現状,お断り,問合せ,香恵,ﾉｯｸ,ｽｲｯﾁﾝｸﾞ,理緒,渡慶次,持ち出せる,宇田,まとめる,育休,推薦,湯浅,切り離し,内蔵,最低限,引く,収集,中原,ｼｭｳ,ｼﾞﾝ,価格,ｱﾙﾌｧﾍﾞｯﾄ,向かう,常男,美咲,浮く,いえる,保留,上嶋,富田林,津々山台,なを,小さい,片言,小松,登美子,映像,解明,ｽﾄｯﾌﾟｳｫｯﾁ,ｱｸｼｮﾝ,公衆,三苫,岩見,将来,運動,綴り,金澤,元々,久裕,ﾋﾟｸﾁｬ,ﾐｭｰｼﾞｯｸ,ごみ箱,多分,最寄り,電車,辻田,秋本,徹底,気が付く,向上,紙面,栞里,西端,正誤,入社,形態,おる,笹島,未菜,大仁田,開示,干渉,火曜,把握,佐和子,文庫,大宅,けんか,雅美,取り外す,上長,だす,乗越,拓海,似る,保つ,経由,美佳,中前,数回,ｼﾌﾄ,通番,整形,欠落,医員,ﾌｼﾞｲ,ﾗｲﾌｻｲｸﾙ,せん,三菱,宮地,善彦,東條,閉鎖,あらわれる,普段,お返し,そもそも,本務,かう,締め切り,しぼれる,原本,いそぐ,鮮明,見やすい,同席,別々,迅速,ご無沙汰,章浩,靖子,池野,裁量,水原,施錠,迷う,折る,立ち会う,森井,ﾄｯﾌﾟﾍﾟｰｼﾞ,ｼｰﾄ,静か,天井,端子,眞奈美,英雄,市販,天候,適宜,いたる,なくす,地下,大生,帳票,転記,実態,講読,知らせる,強度,左記,転職,ﾕﾆ,規定,やらかす,ﾊﾟｯｸ,電池,手すき,特有,やう,問い,見積もり,予測,王様,司書,伴う,任せる,お世話,立て続け,破綻,ﾁｬﾈﾙ,製造元,ささやか,着く,帰路,時折,ｽﾗｲﾄﾞ,受け取り,流れる,4つ,火曜日,僭越,解凍,離れる,ｶｰｿﾙ,拝復,通勤,代々,気軽,ばば,ｹﾝ,ｶﾂ,知らす,各所,先着,当選,発達,足らず

Topic # 3 :
する,認証,障害,接続,ｼｽﾃﾑ,教員,できる,無線,復旧,設定,ｱﾄﾞﾚｽ,相談,ﾌﾟﾘﾝﾀ,起動,問い合わせ,職員,不明,ｹｰﾌﾞﾙ,ﾈｯﾄﾜｰｸ,ﾛｸﾞｲﾝ,画面,取得,繋がる,なる,表示,使用,ﾎﾟｰﾄ,部屋,登録,ﾛｸﾞｱｳﾄ,出来る,変更,学生,ｻｰﾊﾞ,お伝え,確認,ﾒｰﾙ,問題,不具合,通信,実施,電源,有線,ｲﾝﾀｰﾈｯﾄ,機器,ｺﾝｾﾝﾄ,解除,依頼,情報,端末,ﾓｰﾄﾞ,正しい,原因,発生,不良,現地,動作,正常,先生,連絡,失敗,ない,交換,受信,ある,配線,対応,教える,調査,つながる,ﾊﾟｽﾜｰﾄﾞ,説明,ﾊﾌﾞ,ﾌﾟﾘﾝﾀｰ,ﾌﾞﾘｯｼﾞ,誤る,管理,送信,可能,申請,頂く,固定,設置,思う,ｴﾗｰ,印刷,案内,無効,ﾛｸﾞ,削除,ｻﾎﾟｰﾄ,行う,状態,出る,切れる,事象,不安定,利用,様子見,有効,ｾｸﾞﾒﾝﾄ,訪問,送付,故障,使える,ｸﾞﾛｰﾊﾞﾙ,新規,回答,成功,停電,入力,方式,構成,財務,抜き差し,今朝,購入,居室,理学部,済み,ﾊﾟｿｺﾝ,ﾌﾟﾛﾄｺﾙ,試す,工学部,ﾌﾟﾛﾌｧｲﾙ,消える,いただく,同様,お願い,結果,会計,手動,ｻｰﾊﾞｰ,ﾈｯﾄ,伝える,作業,わかる,影響,ﾏﾆｭｱﾙ,再発,手順,ｱｸｾｽ,抜ける,現象,残る,研究,払い出す,特定,入る,範囲,切り分け,見る,紹介,ｶﾒﾗ,修正,改善,解決,届く,検証,ﾎﾜｲﾄ,先方,ｲﾝｽﾄｰﾙ,阿る,変わる,昨日,ﾕｰｻﾞ,ﾐｽ,挿す,業者,ﾁｪｯｸ,経過,外部,ｺﾝ,隔離,複数,間違い,様子,機能,おかしい,その他,判明,質問,かかる,ｽﾌﾟｰﾙ,貸し出す,あり,会議,ﾍﾟｰｼﾞ,格納,理由,自然,誤り,無し,ｽﾊﾟﾑﾒｰﾙ,収容,上手い,物理,戻す,互換,処理,形跡,でる,詳細,到着,送れる,移動,室内,起こす,新しい,払い出し,不可,撤去,ｻｲﾄ,ｱｶｳﾝﾄ,ﾌｨﾙﾀ,外す,最近,うまい,本体,暫定,断線,ﾌﾟﾗｲﾍﾞｰﾄ,ﾃｽﾄ,切り替え,電話,大量,ｱｲﾃﾑ,古い,ﾀｲﾐﾝｸﾞ,再現,ｶｽｹｰﾄﾞ,ｾﾝﾀｰ,ｱﾀﾞﾌﾟﾀ,肥大,解放,ﾋｱﾘﾝｸﾞ,全体,飛ぶ,無い,ﾃﾞﾌｫﾙﾄ,ﾄﾚｲ,ﾂｰﾙ,必要,環境,直接,買い替え,落ちる,ｱﾀﾞﾌﾟﾀｰ,完了,推測,範疇,実行,要す,遅延,ﾄﾞﾒｲﾝ,本日,西口,現在,不適切,赴任,切り替わる,追加,ﾛｯｸ,ﾒｰｶｰ,種別,回避,今後,不在,状況,買い換える,報告,いる,疎通,解消,負荷,問い合わせる,入れる,起こる,通る,頻繁,遮断,全学,無事,繋ぐ,お気に入り,繰り返す,広瀬,更新,ﾘｽﾄ,振る,理解,ﾃｰﾌﾞﾙ,現状,開放,新名,立つ,安定,外れる,辻川,抜く,ｱｯﾌﾟ,購読,相性,一部,取れる,異なる,齋藤,振り替える,該当,ﾊﾟｹｯﾄ,ﾕｰｻﾞｰ,学科,一端,無くなる,実験,仮想,変える,先週,差し替える,消せる,判定,検討,拒否,ﾏｼﾝ,ｽｸﾘﾌﾟﾄ,対象,導入,もらえる,回収,接触,事務,自動的,見える,頻発,直後,おすすめ,守衛,保持,配属,替える,小笠原,挙動,当該,悪い,組み合わせ,柏木,操作,ﾊﾞｰｼﾞｮﾝｱｯﾌﾟ,出力,下書き,正樹,高い,作成,ﾘﾝｸ,起きる,間違う,載る,ﾊｰﾄﾞ,点灯,勧める,工場,出荷,過去,見直す,初期,ｲﾝｼﾃﾞﾝﾄ,工事,差す,配送,いく,空き,了承,空く,坪田,経済,経路,予定,変換,ﾒｰﾙｱﾄﾞﾚｽ,伸びる,川合,次回,遠隔,みる,行く,2つ,通常,提案,ｽﾀﾃｨｯｸ,ﾒｯｾｰｼﾞ,ﾜｲﾔﾚｽ,分類,同一,ﾓﾆﾀｰ,映る,考える,復活,転送,場所,図面,試行,程度,研究所,入替る,貰う,最後,来る,引き込む,対向,科学,やる,藤田,振り分ける,指定,明け,廊下,ｴｽｶﾚｰｼｮﾝ,根本,黒松,増やす,地区,三田,推進,移転,臨時,修理,やり直す,組織,推奨,見れる,大学,言う,開ける,空欄,かわく,けいすけ,割り振る,増設,保守,立ち上がる,学内,かける,件数,怪しい,ﾊﾞｯｸｱｯﾌﾟ,調子,部局,経つ,ｻｰﾋﾞｽ,もらう,亜紀,後日,助手,印刷物,関数,ｵﾌﾟｼｮﾝ,求める,つま,久井,違う,通電,施す,ﾙｰﾃｨﾝｸﾞ,大文字,ﾒｰﾙﾎﾞｯｸｽ,ﾌﾟﾛｸﾞﾗﾑ,出す,判断,認識,対処,覚え,裕光,宛て,関わる,両方,全員,ﾘﾝｸﾞ,分かる,分散,詐称,光る,しる,超える,閲覧,生活,構造,本館,対策,割り,根拠,ﾘﾘｰｽ,発する,一存,施策,参照,見直し,教務,学外,警報,完全

Topic # 4 :
依頼,物質,ｺﾐｭﾆﾃｨ,登録,化学,訓練,向け,受講,教職員,取扱,年度,ﾛｸﾞｲﾝ,ｸﾘｯｸ,氏名,作業,各自,衛生,安全,ｺｰｽ,研究,教育,大阪市立大学,ﾎｰﾑﾍﾟｰｼﾞ,納品,支援,短縮,基づく,田邉,設定,ｱﾅｳﾝｽ,地球,理学,分子,ｱｶｳﾝﾄ,哲朗,技術,回収,小原,動作,無線,持つ,担当,おく,大橋,全学,ｱｸｾｽ,大きい,住吉,職員,いただく,希望,足立,ﾊﾞｰｼﾞｮﾝｱｯﾌﾟ,生物,公式,経緯,ﾌｧｰﾑｳｪｱ,申し込み,専攻,所属,持ち,ﾊﾟｽﾜｰﾄﾞ,大阪,参照,収まる,杉本,ｴﾐ,完了,ｵﾘｼﾞﾝ,野村,別物,実験,堀井,弥生,久末,奈津子,買い替える,認証,ﾌｫﾛｰ,行う,代香,宍戸,付け替える,設置,一孝,ﾀﾌﾞﾚｯﾄ,定型,裕子,生体,電話機,管理,ﾌｫﾝﾄ,渡邊,検査,剛史,招聘,基礎,分野,ｷｬﾉﾝ,恭代,教員,吉里,学生,先端,部門,減少,連休,ためし,いただける,ｺﾝﾃｷｽﾄ,皆谷,榎並,新調,設立,する,競合,停滞,内藤,押せる,新設,取り違える,宣行,もろもろ,軽い,意表,貴子,守矢,里佳,補佐,ｷﾔﾉﾝ,増設,上位,真宏,坂本,昼休み,だす,にる,止まる,英雄,構造,ｵｰﾃﾞｨｵ,ﾒｲﾙ,切り離せる,ｵﾘｼﾞﾅﾙ,引き継ぎ,宇田,島村,満里子,ap,販売元,製図,不通,特有,確認,設計,直通,千恵,改行,勝利,千代美,手書き,ある,川倉,紀彦,防災,領域,田丸,植松,お世話,増田,利用,浩志,撤退,ﾍﾙﾌﾟ,飽和,専属,美香,約束,寿一,黒板,中島,背面,当時,敏也,幸隆,ﾃﾞｽｸ,灰色,記号,製品,ひとまず,おしえる,募金,ﾔﾅｷﾞ,元年,基金,発言,特徴,ｼｰﾝ,委ねる,吉川,藤森,呼び出し,受話器,理想,ｸﾘｯﾌﾟ,平成,裕紀子,ﾎﾞｰﾄﾞ,入り組む,鳴る,ｴﾝﾀｰﾌﾟﾗｲｽﾞ,卒業生,池上,統一,年明け,歯車,宮下,神経,具合,内科,間近,兼務,なる,医局,識別,しがらみ,転換,奈美子,発信,拝復,なぜだか,拍子抜け,12月,関山,受発信,たかこ,しずか,健一,ｲﾗｽﾄﾚｰﾀｰ,困惑,ｶﾞｲﾄﾞﾌﾞｯｸ,包含,吉良,破損,性能,もみ,ふじ,あて,増加,発令,中心,手こずる,固まる,山家,書架,ｹｶﾞ,横切る,引っ掛ける,お願い,拝借,頼む,田邊,玉井,ﾚｰｻﾞｰ,量子,ｵｰﾌﾟﾝ,代物,しょうが,かなう,地盤,走る,ﾁｬｯﾄ,対話,中辻,いか,議員,つけ,監視,除外,陽一郎,南部,お陰様,測定,三宅,映像,在学,古谷,読み替える,安心,民間,声明,隣接,松美,悪化,思わしくない,案内,清二,智美,貴様,宗一,芳紀,堀越,前川,解析,佳い,幹事,由紀,中沢,促進,本道,まえる,波動,b5,福永,村田,利夫,佳彦,生理学,繁行,加工,生産,一案,先頭,ﾊﾞｽ,実態,関田,鹿毛,長澤,元素,差し戻し,懇談,講習,藤巻,集団,分析,迷う,谷本,橋本,幸一,益田,晴恵,お互い,名簿,高等,委員,高齢,美輪子,ｽｸﾘﾌﾟﾄ,佐伯,洋司,ｱﾚﾝｼﾞ,人口,高分子,後援,石川,代用,茂喜,逓送,貼付,亜紀,内線,黒松,生里,つながり,小宮,関西,正樹,かおり,松山,知幸,中台,枝里子,ﾍﾘｳﾑ,液化,小池,ｶﾞﾗｽ,気がつく,状況,点く,植物,村島,下中,工作,重たい,紛らわしい,慣れる,足す,文教,付箋,神谷,昇格,美紗,泰子,附属,事業,複合,除去,厳しい,宇野,理恵,出力,問う,出口,施設,亀田,食品,細胞,同一,和美,空欄,ｻﾝﾌﾟﾙ,済み,博士,思える,記す,湯浅,理学部,物理,保持,寺田,ﾃﾞｨﾚｸﾄﾘ,数字,家庭,受ける,敏之,点滅,市立,森内,間に合う,代表,武田,期待,切り替える,長倉,講座,川上,工学,都市,本学,うごく,秘書,機構,広い,ｼｰﾙ,栄養,早め,和田,ご多忙,下位,大学院,除く,取り出す,動物,日出雄,光合成,直子,要因,割当てる,解凍,比べる,葉子,故障,同室,お話し,機能,人工,健康,公立大,岡崎,日々,学科,申告,ﾖｼﾀﾞ,最大,電子,田中,ついで,かた,価格,連携,趣旨,相談,協議,取り扱い,盗む,お詫び,問い合わせ,絶対,ﾌﾛｰ,ﾃﾝﾌﾟﾚｰﾄ,携帯,漏れ,別室,貸し,付属,宜しい,聡美,稼働,さま,数学

Topic # 5 :
ﾌﾟﾘﾝﾀ,複合,印刷,する,ﾌﾟﾘﾝﾀｰ,ﾈｯﾄﾜｰｸ,仮想,登録,申請,ｱﾄﾞﾚｽ,ｼｽﾃﾑ,問い合わせ,教員,できる,相談,ｸﾞﾛｰﾊﾞﾙ,ﾛｸﾞｲﾝ,説明,追加,ﾎｰﾑ,使用,ｾﾞﾛｯｸｽ,職員,移行,許可,学生,出力,先生,人事,認証,研究,ﾄﾞﾗｲﾊﾞｰ,管理,ﾌﾟﾗｲﾍﾞｰﾄ,設定,新しい,設置,依頼,石本,支援,自動,事務,指定,教える,変更,村岡,ﾒﾝﾊﾞｰ,案内,ｵﾌﾗｲﾝ,移設,ない,切替,経緯,号館,松本,なる,作業,割り当てる,お伝え,ﾃﾚﾋﾞ,障害,ｱｸｾｽ,院生,つける,送付,小林,端末,ｺﾋﾟｰ,故障,必要,通信,疎通,入れ替え,ｲﾝｽﾄｰﾙ,会議,付き,回答,異動,参照,本体,お願い,権限,梅田,入れ替える,共同,入試,就職,更新,九鬼,確認,保全,正しい,雅嗣,運用,切り替える,いただく,哲子,接続,機器,真由美,ある,変わる,広岡,人工,友代,電話,給与,山上,効果,居室,工学部,一郎,問題,移動,地球,ﾃｰﾏ,先端,使い方,付ける,舟川,ｻﾃﾗｲﾄ,新規,手順,生物,よい,機構,藤原,橋本,理学部,図書,阪本,購入,ﾃｽﾄ,ﾛｸﾞｱｳﾄ,川井,ﾋｱﾘﾝｸﾞ,番号,瀬川,佐官,思う,無し,いける,可能,緒方,部屋,入れ替わる,ｽﾄﾚｰｼﾞ,本部,伝える,ﾊﾟｿｺﾝ,裕光,富士ｾﾞﾛｯｸｽ,共有,謙一,操作,見える,上述,美津江,ﾉｰﾄ,学修,野口,所属,環境,ﾌﾟﾛ,付与,質問,受け入れ,とれる,ﾌﾟﾘﾝﾄ,与える,ｷｼ,印字,はじく,発信,機種,制限,ｻｰﾊﾞ,誤り,羽生,かける,適切,学籍,家族,受け入れる,異なる,いらっしゃる,実施,伴う,法学部,学部,神田,普段,検索,着任,否認,統合,あやまる,五十嵐,りう,ﾙｰﾄ,口頭,属する,最終,配置,使える,持つ,以外,買い替える,おしえる,午前,任せる,金森,吉野,ｺﾆｶ,取得,作成,個々,解析,渡す,機械,研修,土肥,独立,利用,ご存じ,嶋本,属す,成松,大塚,聞く,ﾊﾞｲｵ,孝一,三洋,山崎,大学入試ｾﾝﾀｰ,一輝,順子,情報,お答え,ﾏﾆｭｱﾙ,形態,ﾓﾉｸﾛ,個人,事前,ｽｹｼﾞｭｰﾙ,間違う,水口,留学生,所望,厚生,ｻﾎﾟｰﾄ,場所,返送,出来る,来る,記憶,佐竹,切り離す,装置,未登録,委譲,雅人,所在地,正一,自己,欲しい,暗い,生田,ﾕｰｻﾞ,生徒,対応,両方,知る,引越し,監視,成美,安心,まとめる,湯浅,資料,皆様,修理,ふつう,現在,取り外す,常勤,化学,長崎,連絡,失念,高い,温度,実央,在学,用途,返る,英輔,了解,該当,深水,ｽﾃｰﾀｽ,通る,非常勤,分ける,文学部,振る,停電,今年度,光合成,職種,理乃,整える,寺澤,弾く,河村,複数,ﾎﾟｰﾄ,大城,置く,調査,今日,克昭,最新,起票,目安,撮る,会員,基幹,使う,ﾌｧｲﾙ,あきらめる,予定,大阪大学,理事,絵理子,ﾒｰﾙ,内容,行う,建屋,久美子,受入れる,ﾜｰﾄﾞ,いる,勝手,不審,固定,健史,対象,中台,生える,水谷,導入,推進,完了,違う,横山,所管,博史,ﾓﾆﾀ,探せる,中旬,無い,代わる,識別,不一致,付随,奨学,森山,設備,山口,回数,基子,おこなう,使い分け,植村,宮下,厚志,加奈,宇野,理恵,明日,お陰様,ｱｻｲﾝ,存在,発注,松野,横浜市大,英一,ﾄﾞｷｭﾒﾝﾄ,遅い,迎える,ｻｻｷ,経由,大学,同室,克明,付箋,対向,今更,教授,自前,破損,ｱｸｾｻﾘ,期間,年度,映像,来客,気がつく,認識,解像度,智也,不要,不十分,ﾄﾞｯﾄ,杉本,出る,ﾙｰﾑ,変動,かかる,望ましい,大型,入れる,みつける,主催,文字,綾乃,正晃,ｽｷｬﾅｰ,号機,金曜,川嶋,後日,当該,顔ぶれ,昇子,英喜,うまい,櫻木,ｸｽ,集計,単体,尚夫,工作,含める,由紀子,解釈,部長,恐れ,ｵﾘｼﾞﾅﾙ,究明,ｱﾄﾞﾎｯｸ,ｲﾝﾌﾗｽﾄﾗｸﾁｬｰ,来年度,お気づき,ｾｸﾞﾒﾝﾄ,代理,2月,明細,ｲﾝﾌﾟｯﾄ,正式,実務,楠本,株式会社,流用,神谷,お断り,増える,会う,受入,必須,取りまとめ,書き,藤本,続報,設計,改革,客員,通知,ﾌﾟﾘﾝ,平等,ｼﾏ,ﾌﾞﾗｯｸﾘｽﾄ,ﾊｲﾌﾝ,ｽﾀﾝｽ,後援,律子,多田,部署,中村,宇高,標題,余計,技能,要請,園枝,切り替わる,ﾀｰﾐﾅﾙ,構造,康子,早い

Topic # 6 :
中尾,画面,憲一,する,ｱﾌﾟﾘｹｰｼｮﾝ,動画,ｱｲｺ,渡邊,ｾｷｭﾘﾃｨｰ,ｽｲｯﾁ,建築,ﾍﾙﾌﾟ,表示,証明,写真,おかけいたす,ﾌﾟﾗﾝ,ﾃﾞｽｸ,西野,簡単,なる,安否,お願い,ﾍﾟｰｼﾞ,小檜山,いただける,許可,ｽｸﾘｰﾝ,ｼｮｯﾄ,ｻｲﾝ,閲覧,お世話,問い合わせ,続行,石堂,ｱｳﾄ,お送り,お手数,ﾌﾟﾗｲﾊﾞｼｰ,川田,ﾊﾞｰ,左側,書く,状況,ｾｷｭﾘﾃｨ,理学,一般,上部,赤丸,選択肢,雄一郎,選択,女性,ｴﾗｰ,工学部,ﾏｰｸ,松倉,ｶﾞｲﾄﾞ,尾島,衛生,本学,みれる,設定,学科,教員,和恵,ﾒﾆｭｰ,弓場,現時点,ﾊﾞｰｼﾞｮﾝ,ｿﾌﾄ,確認,俊子,警告,ｲﾝｽﾄｰﾙ,良い,添える,しれる,塩川,環境,伊原,高等,対処,宮田,下崎,黒い,よし,止まる,ﾌｫｰﾑ,雅美,三船,奈々,辿る,安全,すすむ,時刻,沿う,題名,保護,治代,個人,できる,いただき,裕子,教示,その他,研究,ﾒｰﾙ,原木,昨夜,下田,利用,付属,ﾏﾆｭｱﾙ,文科,英里,ｽﾗｲﾄﾞ,取り直す,手順,ｸﾘｯｸ,身分,小林,項目,ｼｮｰ,遠隔,明記,内線,情報,企業,直子,でる,検査,即答,専有,記録,伊藤,森田,津田,江村,服部,名古屋大学,画像,由紀子,上長,ﾌﾛｱ,恐れ入る,全額,佳代子,薮崎,頂ける,随時,差出人,進める,信昭,ござる,立てる,ｼｽﾃﾑ,変動,浦野,丸山,ﾌﾗｸﾞ,将来,中島,佐知子,所属,真人,つくる,亜希子,長倉,取り消し,曽我,宜しい,先ほど,提案,ﾀｯﾌﾟ,行う,移管,ｻｲﾄ,選ぶ,ない,内容,存する,指導,追伸,左記,相談,貸出,読み替える,さゆり,ﾊﾞｰｼﾞｮﾝｱｯﾌﾟ,常勤,ふる,昇一,脇本,竹田,重要,負荷,ﾛｸﾞｲﾝ,寺西,存じる,ﾙｯｸ,使い方,右端,夕葉,云々,半分,使用,番号,下さる,ﾊﾞｯﾃｨﾝｸﾞ,ﾌﾟﾛｼﾞｪｸﾄ,水田,地図,通す,細胞,送信,当方,俊也,足す,ｶｳﾝｾﾗｰ,掲示板,とのこ,増井,ﾎﾟｯﾌﾟｱｯﾌﾟ,最新,華子,日原,若森,換え,地区,戴く,仁郎,騒がせる,飛ぶ,間に合う,ﾀｲﾄﾙ,裕己,生物,図書館,康一,含む,わかる,ﾃﾞｨﾚｸﾄﾘ,包括,植物,竹谷,矢田,清重,解決,ﾗｲｾﾝｽ,標題,授業,経営,久保,書き,中橋,千代子,医局,別府,保証,よろしい,きょう,孝昌,ﾀﾞｳﾝﾛｰﾄﾞ,杉原,ﾘﾝｸ,高梨,分類,主幹,払う,該当,貴校,分館,満つ,職員,適当,兼ねる,検収,連絡,立花,ｽﾀｰﾄ,介す,動物,和也,藤川,可能,脇田,ｾﾞﾛ,破棄,大きい,扱う,康夫,畠中,主催,ｴﾝ,ﾄﾗﾝｽ,論文,容量,健司,忘れ,初期,おき,返信,上司,他人,無視,村田,否認,なりすます,草生,信二,委任,実行,赤字,新人,曽田,保有,助ける,結構,病態,辰郎,打診,入力,末尾,名波,章江,生理学,通り,届け出,詳しい,適正,千尋,流れる,富澤,gb,hp,web,家族,就職,荒井,検討,ｲﾝ,ｲﾝﾀｰﾈｯﾄ,ｱｸｾｽ,石橋,買い替える,ﾘｱﾙﾀｲﾑ,成績,締め切り,ﾅﾝﾊﾞｰ,差し替え,渡す,付け加える,理学部,ゆえ,ｵﾌｨｽ,佳世,区域,取り消す,連動,ﾛｰﾄﾞ,移譲,入居,流す,動き,当人,施錠,小宮,些細,久嗣,各人,差し替える,熊本,ﾘｺｰ,契約,手引き,良治,産婦人科,全文,ﾊﾟｽﾜｰﾄﾞ,ﾀｰﾐﾅﾙ,蓄積,日野,別々,辻岡,ｱﾝﾀﾞｰﾗｲﾝ,疎い,張り付ける,ｱｯﾌﾟ,自宅,奈央子,竹下,白田,担当,内山,ある,解る,ｴﾝﾄﾞｳ,書店,肝胆,志津子,白藤,答える,修了,ｺﾝｿｰﾙ,ﾌﾞﾗｳｻﾞｰ,徳山,ﾄﾞﾛｯﾌﾟ,鍵谷,くっつく,ﾄﾞｲﾂ,文字数,響子,川崎,つきあい,東海林,医療,光希,ﾏｽｷﾝｸﾞ,卒業生,知識,未読,ﾌﾟﾗｯﾄﾌｫｰﾑ,まじる,払い出し,圧迫,医学,退室,ﾚｽﾎﾟﾝｽ,ﾐﾆ,連続,改善,遠い,延びる,ｽﾀｲﾙ,今日,劣化,化け,ｴﾗｰｺｰﾄﾞ,岩川,良浩,茂喜,教材,文系,新谷,ｳｨﾝﾄﾞｳ,不明,ｱｸｼｮﾝ,撮影,駆除,すい,なれる,小西,満たせる,火曜日,奥村,短時間,実施,八千代,ｱﾏｿﾞﾝ,手前,規約,理系,山村,補完,機材,英子,文書,渡辺,ﾎﾟｲﾝﾀ,吉川,手伝う,大まか,ｵﾝ,動的,学内,押下,望み,充彦,参照,不安定,ﾌｫｰﾗﾑ,納品,ｱｯﾌﾟﾃﾞｰﾄ,英数字,幸美,発生,英文,ﾛｼｱ,諒子,浦上,案内,添付,黒岩,ｳｲﾝﾄﾞｳ,吉良,兼務,受け入れる,品質,曽部
Traceback (most recent call last):
  File "/Users/user/bachelor/ocu123456.py", line 187, in <module>
    print_topics(best_lda_model, count_vectorizer, number_words) 
  File "/Users/user/bachelor/ocu123456.py", line 174, in print_topics
    topic_wordcloud(topic_idx, fig, long_string)
  File "/Users/user/bachelor/ocu123456.py", line 180, in topic_wordcloud
    ax = fig.add_subplot(2, 3, topic_idx + 1)
  File "/Users/user/bachelor/lib/python3.9/site-packages/matplotlib/figure.py", line 1402, in add_subplot
    ax = subplot_class_factory(projection_class)(self, *args, **kwargs)
  File "/Users/user/bachelor/lib/python3.9/site-packages/matplotlib/axes/_subplots.py", line 39, in __init__
    self._subplotspec = SubplotSpec._from_subplot_args(fig, args)
  File "/Users/user/bachelor/lib/python3.9/site-packages/matplotlib/gridspec.py", line 689, in _from_subplot_args
    raise ValueError(
ValueError: num must be 1 <= num <= 6, not 7

ocu123456.py

# import IPython
from IPython.display import display

# lunar1.py header
import pandas as pd
import glob

# lunar2.py header
import MeCab
tagger = MeCab.Tagger("-Ochasen")
import mojimoji
import os
import urllib

# lunar3.py header
from wordcloud import WordCloud
import matplotlib
import matplotlib.pyplot as plt
import numpy as np

#lunar4.py header
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer

# lunar5.py header
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import LatentDirichletAllocation as LDA

#lunar1.py
text_paths = glob.glob('data/ocu2/*.txt')
texts = []
for text_path in text_paths:
    text = open(text_path, 'r').read()
    # text = text.split('\n') # modified
    text = text.split(',')
    title = text[3] # added
    # title = text[2] # modified
    text = ' '.join(text[8:])
    texts.append(text)
    
news_ss = pd.Series(texts)
# display(news_ss.head()) 

# lunar2.py
def load_jp_stopwords(path="data/jp_stop_words.txt"):
    url = 'http://svn.sourceforge.jp/svnroot/slothlib/CSharp/Version1/SlothLib/NLP/Filter/StopWord/word/Japanese.txt'
    if os.path.exists(path):
        print('File already exists.')
    else:
        print('Downloading...')
        urllib.request.urlretrieve(url, path)
    return pd.read_csv(path, header=None)[0].tolist()

def preprocess_jp(series):
    stop_words = load_jp_stopwords()
    def tokenizer_func(text):
        tokens = []
        node = tagger.parseToNode(str(text))
        while node:
            features = node.feature.split(',')
            surface = features[6]
            if (surface == '*') or (len(surface) < 2) or (surface in stop_words):
                node = node.next
                continue
            noun_flag = (features[0] == '名詞')
            proper_noun_flag = (features[0] == '名詞') & (features[1] == '固有名詞')
            verb_flag = (features[0] == '動詞') & (features[1] == '自立')
            adjective_flag = (features[0] == '形容詞') & (features[1] == '自立')
            if proper_noun_flag:
                tokens.append(surface)
            elif noun_flag:
                tokens.append(surface)
            elif verb_flag:
                tokens.append(surface)
            elif adjective_flag:
                tokens.append(surface)
            node = node.next
        return " ".join(tokens)

    series = series.map(tokenizer_func)
    
    #---------------Normalization-----------#
    series = series.map(lambda x: x.lower())
    series = series.map(mojimoji.zen_to_han)

    return series
    
processed_news_ss = preprocess_jp(news_ss)
# display(processed_news_ss.head()) 

# lunar3.py
font_path="/Library/Fonts/ipaexg.ttf"
font_property = matplotlib.font_manager.FontProperties(fname=font_path, size=24)
# font_property = matplotlib.font_manager.FontProperties(size=24)

def show_wordcloud(series):
    long_string = ','.join(list(series.values))
    
    # Create a WordCloud object
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    # wordcloud = WordCloud( background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    
    # Generate a word cloud
    wordcloud.generate(long_string)
    
    # Visualize the word cloud
    plt.imshow(wordcloud)
    plt.show()

# show_wordcloud(processed_news_ss) 

# lunar4.py
count_vectorizer = CountVectorizer()
count_data = count_vectorizer.fit_transform(processed_news_ss)
tfidf_vectorizer = TfidfTransformer()
tfidf_data = tfidf_vectorizer.fit_transform(count_data)
# print(tfidf_data.toarray()) 
# print(tfidf_data.shape)

# lunar5.py
def gridsearch_best_model(tfidf_data, plot_enabled=False):
    # Define Search Param
    n_topics = [8,9,10,11,12,13]
    # n_topics = [4,5,6,7,8,9]
    search_params = {'n_components': n_topics}
    
    # Init the Model
    lda = LDA(max_iter=25,               # Max learning iterations
              learning_method='batch',   
              random_state=0,            # Random state
              n_jobs = -1,               # Use all available CPUs)
              )
    
    # Init Grid Search Class
    model = GridSearchCV(lda, param_grid=search_params)
    
    # Do the Grid Search
    model.fit(tfidf_data)
    
    # Best Model
    best_lda_model = model.best_estimator_
    
    # Model Parameters
    print("Best Model's Params: ", model.best_params_)
    
    # Log Likelihood Score
    print("Best Log Likelihood Score: ", model.best_score_)
    
    # Perplexity
    print("Model Perplexity: ", best_lda_model.perplexity(tfidf_data))
    
    # Get Log Likelyhoods from Grid Search Output
    log_likelyhoods_score = [round(score) for score in model.cv_results_["mean_test_score"]]
    
    if plot_enabled:
        # Show graph
        plt.figure(figsize=(12, 8))
        plt.plot(n_topics, log_likelyhoods_score)
        plt.title("Choosing Optimal LDA Model")
        plt.xlabel("Number of Topics")
        plt.ylabel("Log Likelyhood Scores")
        plt.show()

    return best_lda_model
best_lda_model = gridsearch_best_model(tfidf_data)

# lunar6.py
def print_topics(model, count_vectorizer, n_top_words):
    fig = plt.figure(figsize=(15,8))
    words = count_vectorizer.get_feature_names()
    for topic_idx, topic in enumerate(model.components_):
        print("\nTopic #", topic_idx, ":")
        long_string = ','.join([words[i] for i in topic.argsort()[:-n_top_words - 1:-1]])
        print(long_string)
        topic_wordcloud(topic_idx, fig, long_string)
    # show plots
    fig.tight_layout()
    fig.show()

def topic_wordcloud(topic_idx, fig, long_string):
    ax = fig.add_subplot(2, 3, topic_idx + 1)
    wordcloud = WordCloud(font_path=font_path, background_color="white", max_words=1000, contour_width=3, contour_color='steelblue')
    wordcloud.generate(long_string)
    ax.imshow(wordcloud)
    ax.set_title('Topic '+str(topic_idx))

number_words = 500
print_topics(best_lda_model, count_vectorizer, number_words)