-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathpolitical_views_recogniser.py
184 lines (154 loc) · 8.01 KB
/
political_views_recogniser.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
import sys
import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import twitter_scraper
import re
import spacy
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import Ridge
from sklearn.base import clone
from stop_words import get_stop_words
from joblib import dump, load
# parameters
dpi = mpl.rcParams['figure.dpi']
plt.rcParams['figure.facecolor'] = "#eeeeee"
img = plt.imread("compass.png")
height, width, depth = img.shape
figsize = width / float(dpi), height / float(dpi)
horizontal_max = 1200
horizontal_min = 200
vertical_max = 150
vertical_min = 1150
third_max = 1150
third_min = 150
nlp = spacy.load('pl_core_news_sm')
# drawing 3-axis political compass, all values should be in range [-1, 1]
def draw_compass(username, horizontal, vertical, third):
if horizontal < -1 or horizontal > 1:
horizontal_value = horizontal_min + int(np.round(float(horizontal_max - horizontal_min) / 2.0, 0))
print(f"ERROR - wrong horizontal value: {horizontal}")
else:
horizontal_value = horizontal_min + int(
np.round((float(horizontal + 1) / 2.0) * float(horizontal_max - horizontal_min), 0))
if vertical < -1 or vertical > 1:
vertical_value = vertical_max + int(np.round(float(vertical_min - vertical_max) / 2.0, 0))
print(f"ERROR - wrong vertical value: {vertical}")
else:
vertical_value = vertical_max + int(
np.round((float(-vertical + 1) / 2.0) * float(vertical_min - vertical_max), 0))
if third < -1 or third > 1:
third_value = third_min + int(np.round(float(third_max - third_min) / 2.0, 0))
print(f"ERROR - wrong third value: {third}")
else:
third_value = third_min + int(np.round((float(third + 1) / 2.0) * float(third_max - third_min), 0))
plt.figure(figsize=figsize)
plt.scatter(x=[horizontal_value], y=[vertical_value], c="red", s=750, linewidths=1, edgecolors="black")
plt.scatter(x=[1574], y=[third_value], c="red", s=11750, marker="_", linewidths=7)
plt.imshow(img)
plt.title(f"Political views compass for Twitter user @{username}", fontsize=24)
plt.axis("off")
plt.tight_layout(pad=1)
plt.show()
plt.close()
def print_statistics(name, values):
print(name.ljust(12), str(len(values)).ljust(6), str(np.round(np.mean(values), 2)).ljust(6),
str(np.round(np.median(values), 2)).ljust(6), str(np.round(np.std(values), 2)).ljust(6),
str(np.round(np.max(values), 2)).ljust(6), str(np.round(np.min(values), 2)).ljust(6), sep="\t")
def load_texts(names, base_path="tweets\\"):
result = []
for name in names:
with open(base_path + name + ".csv", encoding="utf8") as data:
result.append("".join(data.readlines()[1:]))
return result
def clean_text(text):
result = re.sub(r"\"", "", text) # remove quotation marks (especially around tweets)
result = re.sub(r"(^|[^@\w])@(\w{1,15})\b", "", result) # remove twitter usernames
result = re.sub(r"http\S+", "", result) # remove twitter URLs
result = re.sub(r"[＃#](\w+)", "", result) # remove hashtags
result = re.sub(r"\d+", "", result) # remove digits
# lemmatization & punctuation removal
result = " ".join([token.lemma_ for token in nlp(result) if not token.is_punct])
return result
def train_model():
usernames, horizontals, verticals, thirds = twitter_scraper.load_mypolitics_data()
print("Axis \tCount \tMean \tMedian\tStd \tMax \tMin ")
print_statistics("horizontal", horizontals)
print_statistics("vertical", verticals)
print_statistics("third", thirds)
original_texts = load_texts(usernames)
texts = np.array([clean_text(text) for text in original_texts])
np.random.seed(1)
train_indices = np.random.rand(len(texts)) < 0.85 # 85% for training, rest for testing
train_texts = texts[train_indices]
train_horizontals = np.array(horizontals)[train_indices]
train_verticals = np.array(verticals)[train_indices]
train_thirds = np.array(thirds)[train_indices]
test_texts = texts[~train_indices]
test_horizontals = np.array(horizontals)[~train_indices]
test_verticals = np.array(verticals)[~train_indices]
test_thirds = np.array(thirds)[~train_indices]
pipeline = Pipeline([
('tfidf', TfidfVectorizer(encoding='utf-8', stop_words=get_stop_words('polish'),
min_df=0.1, max_df=0.9)),
('clf', Ridge(alpha=0.013, solver='auto', random_state=1))
])
model_horizontal = clone(pipeline).fit(train_texts, train_horizontals)
model_vertical = clone(pipeline).fit(train_texts, train_verticals)
model_third = clone(pipeline).fit(train_texts, train_thirds)
print("Mean score:", np.round(np.mean([model_horizontal.score(test_texts, test_horizontals),
model_vertical.score(test_texts, test_verticals),
model_third.score(test_texts, test_thirds)]), 3))
print("Horizontal score:", np.round(model_horizontal.score(test_texts, test_horizontals), 3))
print("Vertical score: ", np.round(model_vertical.score(test_texts, test_verticals), 3))
print("Third score: ", np.round(model_third.score(test_texts, test_thirds), 3))
print("Prediction / Actual comparison in test subset:")
print("Case\tHorizontal\t\tVertical\t\tThird")
for i, text in enumerate(test_texts):
print(
str(i).ljust(5),
str(np.round(*model_horizontal.predict([text]), 2)).ljust(5),
str(test_horizontals[i]).ljust(5),
str(np.round(*model_vertical.predict([text]), 2)).ljust(5),
str(test_verticals[i]).ljust(5),
str(np.round(*model_third.predict([text]), 2)).ljust(5),
str(test_thirds[i]).ljust(5),
sep="\t"
)
dump(model_horizontal, "model_horizontal.joblib")
dump(model_vertical, "model_vertical.joblib")
dump(model_third, "model_third.joblib")
def most_influential_words_of_model(model, n):
model_names = model['tfidf'].get_feature_names()
model_coefs = model['clf'].coef_.flatten()
pairs = [(name, np.round(abs(coef), 3)) for name, coef in zip(model_names, model_coefs)]
return sorted(pairs, key=lambda tup: tup[1])[-n:]
if __name__ == "__main__":
if len(sys.argv) < 2 or sys.argv[1] == "-help":
print("Wrong arguments!",
f"To recognise political views use: python {sys.argv[0]} [twitter_username]",
f"To train the models use: python {sys.argv[0]} -train",
f"To print the most influential words in the model use: python {sys.argv[0]} -most",
f"To get help use: python {sys.argv[0]} -help",
sep="\n")
elif sys.argv[1] == "-train":
train_model()
elif sys.argv[1] == "-most":
prepared_model_horizontal = load("model_horizontal.joblib")
prepared_model_vertical = load("model_vertical.joblib")
prepared_model_third = load("model_third.joblib")
print("Most influential words in each model (word, abs(coef))")
print("\nHorizontal", *most_influential_words_of_model(prepared_model_horizontal, 10), sep="\n")
print("\nVertical", *most_influential_words_of_model(prepared_model_vertical, 10), sep="\n")
print("\nThird", *most_influential_words_of_model(prepared_model_third, 10), sep="\n")
else:
prepared_model_horizontal = load("model_horizontal.joblib")
prepared_model_vertical = load("model_vertical.joblib")
prepared_model_third = load("model_third.joblib")
name = sys.argv[1]
data = clean_text(twitter_scraper.get_tweets_to_str(name, 1000))
draw_compass(name,
prepared_model_horizontal.predict([data]),
prepared_model_vertical.predict([data]),
prepared_model_third.predict([data]))