文書中のキーワード（企業名）を検索する

import pandas as pd
import re

df_meigara = pd.read_csv('xxx.csv')
df_news = pd.read_csv('xxx.csv')

stock_code_list = df_meigara['STOCK_CODE'].values
news_id_list = df_news['ID'].values

news_stock = {}

for news_id in news_id_list:
    news_text = df_news[df_news['ID']==news_id]['NEWS_DETAIL'].values[0]
    
    stock_list = set()
    for stock_code in stock_code_list:
        stock_name = df_meigara[df_meigara['STOCK_CODE']==stock_code]['STOCK_NAME'].values[0]            
        
        s = re.search(stock_name, news_text)
        if s:
            stock_list.add(stock_name)
    
    news_stock[news_id] = stock_list    

print('news_id', '    stock names')
for news_id in news_id_list[:30]:
    if news_stock[news_id]:
        print(news_id, news_stock[news_id])

news_id = 438663
text = df_news[df_news['ID']==news_id]['NEWS_DETAIL'].values[0]

print(text)

Word2Vec で類似語リストを作成し、文書中に含まれる類似語の数をカウントする

import pandas as pd
import numpy as np

import MeCab
from gensim.models import word2vec

# コーパスの作成
df_news = pd.read_csv('xxx.csv')
tagger = MeCab.Tagger("-Ochasen")

news_list = df_news['NEWS_DETAIL'].values

corpus = []
for text in news_list:
    word_list = []
    node = tagger.parseToNode(text)
    while node:
        word = node.surface
        hinshi = node.feature.split(",")[0]    
        if hinshi=='名詞':
            word_list.append(word)        
        node = node.next
    corpus.append(word_list)

# 学習
model = word2vec.Word2Vec(corpus, size=100, min_count=1, window=10, 
                          seed=1, workers=1, sg=1, hs=0, negative=5, iter=1)
vocab = model.wv.vocab
index2word = model.wv.index2word

# 各単語に対する類似度リストの作成
similar_words_dict = {}
for word in index2word:
    temp = model.wv.most_similar(positive=[word], topn=10)
    similar_words = [t[0] for t in temp]
    similar_words_dict[word] = similar_words


# 使用例
keyword = '日本'
similar_words = np.concatenate(([keyword] ,similar_words_dict[keyword]))
text = news_list[0]

word_list = set()
node = tagger.parseToNode(text)
while node:
    word = node.surface
    hinshi = node.feature.split(",")[0]    
    if hinshi=='名詞':
        word_list.add(word)
        word_list2.append(word)
    node = node.next 
word_list = list(word_list)

count = 0
for word in word_list:
    if word in similar_words:
        count += 1

print('# of similar words in the text')        
print(count)
print()
print('similar words')
print(similar_words)
print()
print('text')
print(text)

Gensim の Poincaré Embeddings を試す

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

import MeCab
from gensim.models.poincare import PoincareModel, PoincareRelations
from gensim.viz.poincare import poincare_2d_visualization

# 階層データの作成
df_news = pd.read_csv('xxx.csv')
news_list = df_news['NEWS_DETAIL'].values

tagger = MeCab.Tagger("-Ochasen")

# relations = set()
relations = []   # 重複があってもよい
for text in news_list[:10]:
    word_list = []
    node = tagger.parseToNode(text)
    while node:
        word = node.surface
        hinshi = node.feature.split(",")[0]    
        if hinshi=='名詞':
            word_list.append(word)        
        node = node.next
        
    for i in range(len(word_list)-1):
        #relations.add((word_list[i], word_list[i+1]))
        relations.append((word_list[i], word_list[i+1]))

# relations = list(relations)

# 　学習
model = PoincareModel(relations, size=2, negative=2, burn_in=10, seed=1)
model.train(epochs=100)

# 作図
relations_set = set(relations)
figure_title = ''
show_node_labels = model.kv.vocab.keys()
vis = poincare_2d_visualization(model, relations_set,
                                figure_title, num_nodes=None,
                                show_node_labels=show_node_labels)

x = vis['data'][1]['x']
y = vis['data'][1]['y']
text = vis['data'][1]['text']

plt.figure(figsize=(7, 7))
plt.scatter(x, y)
# for i, _ in enumerate(x):
#    plt.text(x[i], y[i], text[i])
plt.xticks(np.arange(-1.0, 1.2, 0.2))
plt.yticks(np.arange(-1.0, 1.2, 0.2))
plt.grid('on')
plt.show()

keyword = '日本'
similar_words = model.kv.most_similar(keyword)

print('keyword: ', keyword)
print(len(similar_words))
print(similar_words)

print(model.kv.norm('日本'))
print(model.kv.norm('政府'))

[参照論文・記事]

Poincaré Embeddings for Learning Hierarchical Representations

異空間への埋め込み！Poincare Embeddingsが拓く表現学習の新展開

Poincaré Embeddings による職種の類似度計算とその利用

双曲空間ではじめるレコメンデーション