Sentiment analysis with Python (word2vec)

The other day, I participated in a Python study session sponsored by Team Zet Co., Ltd. The theme this time is "text emotions using word2vec" "Analysis". To be honest, it was a crazy theme for me who first touched Python a week ago, but I wonder if it's possible to experience how the grammar I'm studying is being put to good use. I rushed in and boarded one day before the event.

By the way, I will leave the introduction to this extent and get into the main subject.

What is Word2Vec in the first place? ?? ??

Neural network model (machine learning) that analyzes words. Simply put, it seems that words can be vectorized and weighted. (For more information, refer to here)

This time White Goat Corporation I used the word2vec model of.

How to install word2vec is here

First of all Make sure the words are vectorized by word2vec. Let's type this code with word2vec implemented.

sample.py

import gensim.models.word2vec.Word2Vec as wv 

print(len(model.wv["Love"]))
model.wv["Love"]

Then

50 array([ 0.09289702, -0.16302316, -0.08176763, -0.29827002, 0.05170078, 0.07736144, -0.06452437, 0.19822665, -0.11941547, -0.11159643, 0.03224859, 0.03042056, -0.09065174, -0.1677992 , -0.19054233, 0.10354111, 0.02630192, -0.06666993, -0.06296805, 0.00500843, 0.26934028, 0.05273635, 0.0192258 , 0.2924312 , -0.23919497, 0.02317964, -0.21278766, -0.01392282, 0.24962738, 0.11264788, 0.05772769, 0.20941015, -0.01239212, -0.1256235 , -0.19794041, 0.1267719 , -0.12306885, 0.01006295, 0.08548331, -0.08936502, -0.05429656, -0.09757583, 0.10338967, 0.13714872, 0.23966707, 0.02216845, 0.02270923, 0.32569838, -0.0311841 , -0.00150117], dtype=float32)

The result will be returned. This is the word "love" made up of 50 dimensions. It shows that the component is composed of the above elements.

next

sample.py

#Extract words similar to keyword
sim_do = model.wv.most_similar(positive=["Girlfriend"], topn=30)
#Since it is listed, it is shaped for easy viewing
print(*[" ".join([v, str("{:.5f}".format(s))]) for v, s in sim_do], sep="\n")

When you hit Herself 0.82959 Molly 0.82547 He 0.82406 Sylvia 0.80452 Charlie 0.80336 Lover 0.80197 You can extract words with similar meanings such as. The number to the right of the word is a quantification of how much you are with the word "she".

Also, when you want to know how long the two words are

similarity = model.wv.similarity(w1="Apple", w2="Strawberry")
print(similarity)

similarity = model.wv.similarity(w1="Apple", w2="Aomori")
print(similarity)

similarity = model.wv.similarity(w1="Apple", w2="Anpanman")
print(similarity)

Then 0.79041845 0.30861858 0.45321244 Will be returned. We quantified how similar the words w1 and w2 are. If you say apples, Aomori! I think that many people associate it with, but since I have decided that Anpanman is more similar than Aomori, I understand that this model is not perfect yet.

Well, here

"King"-"Man" + "Woman" = "Queen" ???

sim_do = model.wv.most_similar(positive = ["King", "Female"], negative=["male"], topn=5)
print(*[" ".join([v, str("{:.5f}".format(s))]) for v, s in sim_do], sep="\n")

#Words in positive compare the degree of similarity, words in negative compare the degree of dissimilarity

import numpy as np
t = Tokenizer()
s = '
# Enter your favorite sentences.
'
output_data=[]
x = np.empty((0,4), float)
for token in t.tokenize(s):
  if token.part_of_speech.split(',')[0]=="noun" or token.part_of_speech.split(',')[0]=="adjective":
    print(token.surface)
    similarity1 = model.wv.similarity(w1=token.surface, w2="happy")
    #print("joy:{0}".format(similarity1))
    similarity2 = model.wv.similarity(w1=token.surface, w2="pleasant")
    #print("sorrow:{0}".format(similarity2))
    similarity3 = model.wv.similarity(w1=token.surface, w2="sad")
    #print("anxiety:{0}".format(similarity3))
    similarity4 = model.wv.similarity(w1=token.surface, w2="excitement")
    #print("Interest:{0}".format(similarity4))
    x = np.append(x, np.array([[similarity1, similarity2, similarity3, similarity4]]), axis=0)

print("-"*30)
print(np.mean(x, axis=0))
print("Happy:{0}".format(np.mean(x, axis=0)[0]))
print("easy:{0}".format(np.mean(x, axis=0)[1]))
print("Sadness:{0}".format(np.mean(x, axis=0)[2]))
print("Xing:{0}".format(np.mean(x, axis=0)[3]))