forked from woctezuma/steam-reviews
-
Notifications
You must be signed in to change notification settings - Fork 0
/
identify_joke_reviews.py
313 lines (237 loc) · 14.3 KB
/
identify_joke_reviews.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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
import sys
from langdetect import detect, DetectorFactory, lang_detect_exception
from textblob import TextBlob, exceptions
from cluster_reviews import print_sentiment_analysis
from compute_wilson_score import compute_wilson_score
from describe_reviews import load_data, describe_data, get_review_content
def detect_language(review_content, blob=None, call_google_translate=False):
try:
if call_google_translate:
# More accurate but requires an active Internet connection, and might result in a slow down of the process.
if blob is None:
blob = TextBlob(review_content)
detected_language = blob.detect_language()
else:
# It is up to the user to decide (trade-off accuracy vs. slower running time + Internet requirement).
DetectorFactory.seed = 0
detected_language = detect(review_content)
except exceptions.TranslatorError:
# This exception can be raised by 'textblob'.
# The error is typically: TranslatorError('Must provide a string with at least 3 characters.')
# Since the review is very short, it is likely a joke review, so we won't dismiss it from our study.
# Example of such a review: http://steamcommunity.com/profiles/76561198169555911/recommended/723090/
detected_language = 'en'
except lang_detect_exception.LangDetectException:
# This exception can be raised by 'langdetect'.
detected_language = 'en'
return detected_language
def get_review_sentiment_dictionary(app_id, accepted_languages=None,
perform_language_detection_with_google_tool=False,
verbose_reviews_wrongly_tagged_as_written_in_english=False):
# A light version of aggregate_reviews() from describe_reviews.py
# NB: Only reviews marked on Steam as being written in English are accepted for sentiment analysis to work properly.
if accepted_languages is None:
accepted_languages = ['english']
review_data = load_data(app_id)
print('\nAppID: ' + app_id)
(_, reviews) = describe_data(review_data)
wrongly_tagged_review_id = []
count_reviews_wrongly_tagged_as_written_in_english = 0
count_reviews_tagged_as_written_in_english = 0
review_dict = dict()
review_dict['positive'] = dict()
review_dict['negative'] = dict()
for review in reviews:
if review['language'] in accepted_languages:
# Review ID
review_id = review["recommendationid"]
# Whether the review is marked on Steam as positive
is_positive_review = bool(review['voted_up'])
# Review text
review_content = review['review']
count_reviews_tagged_as_written_in_english += 1
# Sentiment analysis
blob = TextBlob(review_content)
# Check language with a tool by Google to detect reviews WRONGLY tagged as English
if perform_language_detection_with_google_tool:
detected_language = detect_language(review_content, blob)
else:
detected_language = 'en'
# Hard-coded check for English language
accepted_languages_iso = ['en']
# TODO For generality, one would need to match accepted_languages to accepted_languages_iso (ISO 639-1)
# cf. https://en.wikipedia.org/wiki/ISO_639-1
# cf. https://gist.github.com/carlopires/1262033
if not (detected_language in accepted_languages_iso):
count_reviews_wrongly_tagged_as_written_in_english += 1
wrongly_tagged_review_id.append(review_id)
if verbose_reviews_wrongly_tagged_as_written_in_english:
print('\nReview #' + str(count_reviews_wrongly_tagged_as_written_in_english)
+ ' detected as being written in ' + detected_language + ' instead of English:')
print(review_content + '\n')
continue
if is_positive_review:
keyword = 'positive'
else:
keyword = 'negative'
review_dict[keyword][review_id] = dict()
review_dict[keyword][review_id]['polarity'] = blob.sentiment.polarity
review_dict[keyword][review_id]['subjectivity'] = blob.sentiment.subjectivity
review_dict['language_tag'] = dict()
review_dict['language_tag']['reviews_wrongly_tagged_English'] = wrongly_tagged_review_id
review_dict['language_tag'][
'num_reviews_wrongly_tagged_English'] = count_reviews_wrongly_tagged_as_written_in_english
review_dict['language_tag']['num_reviews_tagged_English'] = count_reviews_tagged_as_written_in_english
review_dict['language_tag']['num_reviews'] = len(reviews)
# Mostly display
try:
percentage_reviews_tagged_as_written_in_english = review_dict['language_tag']['num_reviews_tagged_English'] / \
review_dict['language_tag']['num_reviews']
end_of_sentence = ' Percentage of English tags: {0:.2f}'.format(percentage_reviews_tagged_as_written_in_english)
except ZeroDivisionError:
percentage_reviews_tagged_as_written_in_english = -1
end_of_sentence = ''
review_dict['language_tag']['prct_English_tags'] = percentage_reviews_tagged_as_written_in_english
sentence = 'Number of reviews: {0} ({1} with English tag ; {2} with another language tag).'
print(sentence.format(review_dict['language_tag']['num_reviews'],
review_dict['language_tag']['num_reviews_tagged_English'],
review_dict['language_tag']['num_reviews'] - review_dict['language_tag'][
'num_reviews_tagged_English'])
+ end_of_sentence)
# Mostly display
# noinspection PyPep8
num_confirmed_english_tags = review_dict['language_tag']['num_reviews_tagged_English'] - \
review_dict['language_tag']['num_reviews_wrongly_tagged_English']
try:
percentage_confirmed_english_tags = num_confirmed_english_tags / review_dict['language_tag'][
'num_reviews_tagged_English']
end_of_sentence = ' Percentage of confirmed English tags: {0:.2f}\n'.format(percentage_confirmed_english_tags)
except ZeroDivisionError:
percentage_confirmed_english_tags = -1
end_of_sentence = '\n'
review_dict['language_tag']['prct_confirmed_English_tags_among_English_tags'] = percentage_confirmed_english_tags
# noinspection PyPep8
review_dict['language_tag']['prct_confirmed_English_tags_among_all_tags'] = max(0,
percentage_confirmed_english_tags) * \
max(0,
percentage_reviews_tagged_as_written_in_english)
accepted_languages_as_concatenated_str = ' '.join(l.capitalize() for l in accepted_languages)
sentence = 'Number of reviews tagged as in ' + accepted_languages_as_concatenated_str + \
': {0} ({1} with dubious tag ; {2} with tag confirmed by language detection).'
print(sentence.format(review_dict['language_tag']['num_reviews_tagged_English'],
review_dict['language_tag']['num_reviews_wrongly_tagged_English'],
num_confirmed_english_tags)
+ end_of_sentence)
return review_dict
def classify_reviews(review_dict, sentiment_threshold=None, verbose=False):
# The variable sentiment_threshold is a Python dictionary, which describes the criterion to distinguish between
# acceptable and joke reviews, based on Sentiment Analysis.
if (sentiment_threshold is None) or bool(len(sentiment_threshold) == 0):
sentiment_threshold = {'polarity': [-0.2, 0.2], 'subjectivity': [0.36, 1]}
# NB: If thresholds for polarity and subjectivity are set to the following values,
# then the classification cannot be performed, i.e. every review is marked as ACCEPTABLE.
# - polarity threshold: [-1, 1] to avoid any polarity-based criterion (therefore solely rely on subjectivity)
# - subjectivity threshold: [ 0, 1] to avoid any subjectivity-based criterion (therefore solely rely on polarity)
acceptable_reviews_dict = dict()
joke_reviews_dict = dict()
for keyword in ['positive', 'negative']:
current_review_ids = review_dict[keyword].keys()
# Polarity set used with OR: joke reviews necessarily have a polarity INSIDE the interval!
# noinspection PyPep8
acceptable_review_ids_wrt_polarity = [reviewID for reviewID in current_review_ids
if bool(
review_dict[keyword][reviewID]['polarity'] < sentiment_threshold['polarity'][0]
or review_dict[keyword][reviewID]['polarity'] > sentiment_threshold['polarity'][1])]
# Subjectivity interval used with AND: joke reviews necessarily have a subjectivity OUTSIDE the interval!
# noinspection PyPep8,PyPep8
acceptable_review_ids_wrt_subjectivity = [reviewID for reviewID in current_review_ids
if bool(
sentiment_threshold['subjectivity'][0] <= review_dict[keyword][reviewID]['subjectivity'] <=
sentiment_threshold['subjectivity'][1])]
acceptable_review_ids = set(acceptable_review_ids_wrt_polarity).union(
set(acceptable_review_ids_wrt_subjectivity))
acceptable_reviews_dict[keyword] = acceptable_review_ids
joke_reviews_dict[keyword] = current_review_ids - acceptable_review_ids
if verbose:
print('A review is acceptable if it is acceptable with respect to either polarity or subjectivity.')
print('Set for being acceptable w.r.t. polarity: [-1.00, {0:.2f}[ U ]{1:.2f}, 1.00]'.format(
sentiment_threshold['polarity'][0],
sentiment_threshold['polarity'][1]))
print('Interval for being acceptable w.r.t. subjectivity: [{0:.2f}, {1:.2f}]'.format(
sentiment_threshold['subjectivity'][0],
sentiment_threshold['subjectivity'][1]))
return acceptable_reviews_dict, joke_reviews_dict
def get_dictionary_wilson_score(review_dict, verbose=False):
num_pos = len(review_dict['positive'])
num_neg = len(review_dict['negative'])
wilson_score = compute_wilson_score(num_pos, num_neg)
if verbose:
num_reviews = num_pos + num_neg
if num_reviews > 0:
sentence = 'Number of reviews: {0} ({1} up ; {2} down) ; Wilson score: {3:.2f}'
print(sentence.format(num_reviews, num_pos, num_neg, wilson_score))
else:
sentence = 'Number of reviews: {0}'
print(sentence.format(num_reviews))
return wilson_score
def show_reviews(app_id, review_id_list, max_num_reviews_to_print=None):
# Adapted from show_fixed_number_of_reviews_from_given_cluster() in cluster_reviews.py
for (review_count, reviewID) in enumerate(review_id_list):
review_content = get_review_content(app_id, reviewID)
if (max_num_reviews_to_print is not None) and (review_count >= max_num_reviews_to_print):
break
# Reference: https://stackoverflow.com/a/18544440
print("\n ==== Review " + str(review_count + 1) + " (#reviews = " + str(len(review_id_list)) + ") ====")
print_sentiment_analysis(review_content)
try:
print(review_content)
except UnicodeEncodeError:
# Reference: https://stackoverflow.com/a/3224300
print(review_content.encode('ascii', 'ignore'))
return
def main(argv):
app_id_list = ["723090", "639780", "573170"]
if len(argv) == 0:
app_id = app_id_list[-1]
print("No input detected. AppID automatically set to " + app_id)
else:
app_id = argv[0]
print("Input appID detected as " + app_id)
accepted_languages = ['english']
perform_language_detection_with_google_tool = True
verbose_reviews_wrongly_tagged_as_written_in_english = True
review_dict = get_review_sentiment_dictionary(app_id, accepted_languages,
perform_language_detection_with_google_tool,
verbose_reviews_wrongly_tagged_as_written_in_english)
sentiment_verbose = True
# noinspection PyTypeChecker
(acceptable_reviews_dict, joke_reviews_dict) = classify_reviews(review_dict,
sentiment_threshold=None,
verbose=sentiment_verbose)
print_wilson_score = True
print('\nStats for all reviews available in ' + ' '.join([l.capitalize() for l in accepted_languages]))
wilson_score_raw = get_dictionary_wilson_score(review_dict, print_wilson_score)
max_num_reviews_to_print = 5
for keyword in ['positive', 'negative']:
review_id_list = acceptable_reviews_dict[keyword]
if len(review_id_list) > 0:
print('\n\t[ ========================================== ]')
print('\t[ ====== Acceptable ' + keyword + ' reviews ======= ]')
show_reviews(app_id, review_id_list, max_num_reviews_to_print)
for keyword in ['positive', 'negative']:
review_id_list = joke_reviews_dict[keyword]
if len(review_id_list) > 0:
print('\n\t[ ==================================== ]')
print('\t[ ====== Joke ' + keyword + ' reviews ======= ]')
show_reviews(app_id, review_id_list, max_num_reviews_to_print)
print('\n\t[ ==================================== ]')
print('\nStats for all acceptable reviews (subjectivity >= threshold)')
wilson_score_acceptable_only = get_dictionary_wilson_score(acceptable_reviews_dict, print_wilson_score)
print('\nStats for detected joke reviews (subjectivity < threshold)')
_ = get_dictionary_wilson_score(joke_reviews_dict, print_wilson_score)
wilson_score_deviation = wilson_score_raw - wilson_score_acceptable_only
print(
'\nConclusion: estimated deviation of Wilson score due to joke reviews: {0:.2f}'.format(wilson_score_deviation))
return True
if __name__ == "__main__":
main(sys.argv[1:])