File size: 8,793 Bytes
60ded96
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import pandas as pd
import numpy as np

def has_vectors(doc):
  return np.all([token.has_vector for token in doc])

def extract_doc_aspects(doc):

    prod_pronouns = ['it','this','they','these']

    rule1_pairs = []
    rule2_pairs = []
    rule3_pairs = []
    rule4_pairs = []
    rule5_pairs = []
    rule6_pairs = []
    rule7_pairs = []

    for token in doc:
        if token.text == 'product':
          continue

        ## FIRST RULE OF DEPENDANCY PARSE -
        ## M - Sentiment modifier || A - Aspect
        ## RULE = M is child of A with a relationship of amod
        A = "999999"
        M = "999999"
        if token.dep_ == "amod" and not token.is_stop:
            M = token.text
            A = token.head.text

            # add adverbial modifier of adjective (e.g. 'most comfortable headphones')
            M_children = token.children
            for child_m in M_children:
                if(child_m.dep_ == "advmod"):
                    M_hash = child_m.text
                    M = M_hash + " " + M
                    break

            # negation in adjective, the "no" keyword is a 'det' of the noun (e.g. no interesting characters)
            A_children = token.head.children
            for child_a in A_children:
                if(child_a.dep_ == "det" and child_a.text == 'no'):
                    neg_prefix = 'not'
                    M = neg_prefix + " " + M
                    break

        if(A != "999999" and M != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict1 = {"noun" : A, "adj" : M, "rule" : 1}
            rule1_pairs.append(dict1)


        # # SECOND RULE OF DEPENDANCY PARSE -
        # # M - Sentiment modifier || A - Aspect
        # Direct Object - A is a child of something with relationship of nsubj, while
        # M is a child of the same something with relationship of dobj
        # Assumption - A verb will have only one NSUBJ and DOBJ
        children = token.children
        A = "999999"
        M = "999999"
        add_neg_pfx = False
        for child in children :
            if(child.dep_ == "nsubj" and not child.is_stop):
                A = child.text

            if((child.dep_ == "dobj" and child.pos_ == "ADJ") and not child.is_stop):
                M = child.text

            if(child.dep_ == "neg"):
                neg_prefix = child.text
                add_neg_pfx = True

        if (add_neg_pfx and M != "999999"):
            M = neg_prefix + " " + M

        if(A != "999999" and M != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict2 = {"noun" : A, "adj" : M, "rule" : 2}
            rule2_pairs.append(dict2)


        ## THIRD RULE OF DEPENDANCY PARSE -
        ## M - Sentiment modifier || A - Aspect
        ## Adjectival Complement - A is a child of something with relationship of nsubj, while
        ## M is a child of the same something with relationship of acomp
        ## Assumption - A verb will have only one NSUBJ and DOBJ
        ## "The sound of the speakers would be better. The sound of the speakers could be better" - handled using AUX dependency

        children = token.children
        A = "999999"
        M = "999999"
        add_neg_pfx = False
        for child in children :
            if(child.dep_ == "nsubj" and not child.is_stop):
                A = child.text

            if(child.dep_ == "acomp" and not child.is_stop):
                M = child.text

            # example - 'this could have been better' -> (this, not better)
            if(child.dep_ == "aux" and child.tag_ == "MD"):
                neg_prefix = "not"
                add_neg_pfx = True

            if(child.dep_ == "neg"):
                neg_prefix = child.text
                add_neg_pfx = True

        if (add_neg_pfx and M != "999999"):
            M = neg_prefix + " " + M

        if(A != "999999" and M != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict3 = {"noun" : A, "adj" : M, "rule" : 3}
            rule3_pairs.append(dict3)


        ## FOURTH RULE OF DEPENDANCY PARSE -
        ## M - Sentiment modifier || A - Aspect

        #Adverbial modifier to a passive verb - A is a child of something with relationship of nsubjpass, while
        # M is a child of the same something with relationship of advmod

        #Assumption - A verb will have only one NSUBJ and DOBJ

        children = token.children
        A = "999999"
        M = "999999"
        add_neg_pfx = False
        for child in children :
            if((child.dep_ == "nsubjpass" or child.dep_ == "nsubj") and not child.is_stop):
                A = child.text

            if(child.dep_ == "advmod" and not child.is_stop):
                M = child.text
                M_children = child.children
                for child_m in M_children:
                    if(child_m.dep_ == "advmod"):
                        M_hash = child_m.text
                        M = M_hash + " " + child.text
                        break

            if(child.dep_ == "neg"):
                neg_prefix = child.text
                add_neg_pfx = True

        if (add_neg_pfx and M != "999999"):
            M = neg_prefix + " " + M

        if(A != "999999" and M != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict4 = {"noun" : A, "adj" : M, "rule" : 4}
            rule4_pairs.append(dict4)

        ## FIFTH RULE OF DEPENDANCY PARSE -
        ## M - Sentiment modifier || A - Aspect

        #Complement of a copular verb - A is a child of M with relationship of nsubj, while
        # M has a child with relationship of cop

        #Assumption - A verb will have only one NSUBJ and DOBJ

        children = token.children
        A = "999999"
        buf_var = "999999"
        for child in children :
            if(child.dep_ == "nsubj" and not child.is_stop):
                A = child.text

            if(child.dep_ == "cop" and not child.is_stop):
                buf_var = child.text

        if(A != "999999" and buf_var != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict5 = {"noun" : A, "adj" : token.text, "rule" : 5}
            rule5_pairs.append(dict5)


        ## SIXTH RULE OF DEPENDANCY PARSE -
        ## M - Sentiment modifier || A - Aspect
        ## Example - "It ok", "ok" is INTJ (interjections like bravo, great etc)

        children = token.children
        A = "999999"
        M = "999999"
        if(token.pos_ == "INTJ" and not token.is_stop):
            for child in children :
                if(child.dep_ == "nsubj" and not child.is_stop):
                    A = child.text
                    M = token.text

        if(A != "999999" and M != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict6 = {"noun" : A, "adj" : M, "rule" : 6}
            rule6_pairs.append(dict6)

        ## SEVENTH RULE OF DEPENDANCY PARSE -
        ## M - Sentiment modifier || A - Aspect
        ## ATTR - link between a verb like 'be/seem/appear' and its complement
        ## Example: 'this is garbage' -> (this, garbage)

        children = token.children
        A = "999999"
        M = "999999"
        add_neg_pfx = False
        for child in children :
            if(child.dep_ == "nsubj" and not child.is_stop):
                A = child.text

            if((child.dep_ == "attr") and not child.is_stop):
                M = child.text

            if(child.dep_ == "neg"):
                neg_prefix = child.text
                add_neg_pfx = True

        if (add_neg_pfx and M != "999999"):
            M = neg_prefix + " " + M

        if(A != "999999" and M != "999999"):
            if A in prod_pronouns :
                A = "product"
            dict7 = {"noun" : A, "adj" : M, "rule" : 7}
            rule7_pairs.append(dict7)

    aspects = []

    aspects = rule1_pairs + rule2_pairs + rule3_pairs +rule4_pairs +rule5_pairs + rule6_pairs + rule7_pairs

    return aspects

def extract_aspects(nlp, reviews):
  aspects = []

  data = ([
    (x[1], x[0]) for x in reviews['text_cleaned'].reset_index().to_numpy()
  ])

  for doc, review_id in nlp.pipe(data, as_tuples=True):
    doc_aspects = extract_doc_aspects(doc)
    doc_aspects = [
        [review_id, aspect['noun'], aspect['adj'], aspect['rule']] 
        for aspect in doc_aspects if not aspect['noun'].lower().startswith('product')
    ]
    # filter aspects with out of vocubalary nouns
    doc_aspects = [
        doc_aspect for doc_aspect in doc_aspects 
        if has_vectors(nlp(doc_aspect[1]))
    ]
    aspects.extend(doc_aspects)

  aspects = pd.DataFrame(aspects, columns=['review_id', 'aspect', 'opinion', 'rule'])

  return aspects