-
Notifications
You must be signed in to change notification settings - Fork 1
/
datapipeline.py
145 lines (120 loc) · 5.29 KB
/
datapipeline.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
import pandas as pd
from copy import deepcopy
from joblib import load, dump
from statistics import mean
from collections import Counter
import sys
sys.path.append('./utils')
from utils import review_feature
import argparse
import time
rf = review_feature()
start = time.time()
parser = argparse.ArgumentParser(description='Ranking Reviews')
parser.add_argument('--spell_threshold', action='store', type=int, default=0.9, help='Spell checking threshold')
parser.add_argument('--model_path', action='store', type=str, default='randomforest.joblib', help='Model Path')
parser.add_argument('--file_name', action='store', type=str, default='data/test.csv',
help='File to Rank (Product, Review)')
parser.add_argument('--testing', action='store', type=str, default='False', help='Get Ranking Score Test or Not (True/False)')
args = parser.parse_args()
classifier = load(args.model_path)
df = pd.read_csv(args.file_name)
print(df.head(5))
bad_reviews = set()
language_error = set()
gibberish = set()
swear = set()
company_tag = set()
for indx in df.index:
review = df.at[indx, 'answer_option']
## Language Detection
try:
b = rf.language_detection(review)
if b == 'hi' or b == 'mr':
language_error.add(indx)
except:
language_error.add(indx)
## Gibberish Detection
if rf.gibberish_detection(review, prefix_path='utils'):
gibberish.add(indx)
## Swear Words Check
if rf.english_swear_check(review) or rf.hindi_swear_check(review):
swear.add(indx)
## Identify reviews on Competitive Brands
if rf.competitive_brand_tag(review):
company_tag.add(indx)
print("Number of Bad Reviews for Language Error: {} \n Number of Bad Reviews for Gibberish: {} \n Number of Bad Reviews for Swear: {} \n Number of Bad Reviews for Competitive Brand: {}".format(len(language_error), len(gibberish), len(swear), len(company_tag)))
bad_reviews = list(bad_reviews.union(swear, company_tag, gibberish, language_error))
print("DELETED REVIEWS: \n", df[df.index.isin(bad_reviews)])
df = df[~df.index.isin(bad_reviews)].reset_index(drop=True)
# ## Spell Correct
# for indx in df.index:
# review = df.at[indx, 'answer_option']
# try:
# df.at[indx, 'answer_option'] = rf.spell_correct(review, args.spell_threshold)
# except:
# bad_reviews.append(indx)
# df = df[~df.index.isin(bad_reviews)].reset_index(drop=True)
# print("Stage Spell Correction Complete")
df = df.sort_values(by=['product'], ignore_index=True)
df['review_len'] = df['answer_option'].apply(lambda x: len(x.split()))
df['Rn'] = 0.0
df['Rp'] = 0.0
df['Rs'] = 0.0
df['Rc'] = 0.0
df['Rd'] = 0.0
df['Rsc'] = 0.0
product_list = df['product'].unique()
for product in product_list:
data = df[df['product'] == product]
unique_bag = set()
for review in data['answer_option']:
review = review.lower()
words = review.split()
unique_bag = unique_bag.union(set(words))
for indx in data.index:
review = data.at[indx, 'answer_option']
df.at[indx, 'Rp'] = rf.polarity_sentiment(review)
df.at[indx, 'Rs'] = rf.subjectivity_sentiment(review)
df.at[indx, 'Rd'] = rf.service_tag(review)
df.at[indx, 'Rsc'] = rf.slang_emoji_polarity_compoundscore(review)
df.at[indx, 'Rc'] = float(len(set(review.split()))) / float(len(unique_bag))
df.loc[df['product'] == product, 'Rn'] = rf.noun_score(data['answer_option'].values).values
product_list = df['product'].unique()
df['win'] = 0
df['lose'] = 0
df['review_score'] = 0.0
df.reset_index(inplace=True, drop=True)
def score_giver(C, D):
E = pd.merge(C, D, how='outer', on='j')
E.drop(columns=['j'], inplace=True)
q = classifier.predict(E.values)
return Counter(q)
for i, product in enumerate(product_list):
data = df[df['product'] == product]
for indx in data.index:
review = df.loc[indx, ['review_len', 'Rn', 'Rp', 'Rs', 'Rc', 'Rd', 'Rsc']]
review['j'] = 'jn'
C = pd.DataFrame([review])
D = data[data.index != indx].loc[:, ['review_len', 'Rn', 'Rp', 'Rs', 'Rc', 'Rd', 'Rsc']]
D['j'] = 'jn'
E = pd.merge(C, D, how='outer', on='j')
score = score_giver(C, D)
df.at[indx, 'win'] = 0 if score.get(1) is None else score.get(1)
df.at[indx, 'lose'] = 0 if score.get(0) is None else score.get(0)
df.at[indx, 'review_score'] = float(0 if score.get(1) is None else score.get(1)) / len(data) * 1.0
print("Iteration: {} Reviews of Product: {} Ranked".format(i + 1, product))
df = df.sort_values(by=['product', 'review_score'], ascending=False)
if args.testing == 'True':
data_split = pd.crosstab(df['product'], df['label'])
r_accuracy = []
for product in product_list:
x = data_split[data_split.index == product][1][0]
number_of_1_in_x = Counter(df[df['product'] == product].iloc[:x, ]['label']).get(1)
rank_accuracy = float(number_of_1_in_x * 1.0 / x * 1.0)
print("Product: {} | Rank Accuracy: {}".format(product, rank_accuracy))
r_accuracy.append(rank_accuracy)
print("TEST DATA: Mean Rank Accuracy: {}".format(mean(r_accuracy)))
print(df[['product', 'answer_option', 'review_score']])
df[['product', 'answer_option', 'review_score']].to_csv('data/test_ranked_output.csv', index=False)
print('RANKING COMPLETE TIME TAKEN: {}'.format(time.time() - start))