-
Notifications
You must be signed in to change notification settings - Fork 0
/
qa.py
178 lines (150 loc) · 5.92 KB
/
qa.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
import sys
import os
import re
import time
import itertools
import gensim
import spacy
from spacy.matcher import Matcher
import helpers
# testset2
# RECALL = .5511
# PRECISION = .3849
# F-MEASURE = .4533
def answer_story_questions(path, story_key):
story_file = open(path + story_key + ".story")
story = "".join(story_file.readlines()[6:]).replace('\n', ' ')
question_file = open(path + story_key + ".questions")
questions = list((list(g) for k,g in itertools.groupby(question_file.readlines(), key=lambda x: x != '\n') if k))
doc = nlp(story)
sents = list(doc.sents)
N = len(sents)
out = []
for q_a in questions:
question = [helpers.remove_punc(w) for w in q_a[1].strip().lower().split()[1:] if w not in stopwords]
full_q = " ".join(question)
q_sent = list(nlp(" ".join(q_a[1].strip().split()[1:])).sents)[0]
# find desired named entities based on keywords
desired_ent = set()
for k in helpers.named_entities_by_question.keys():
if k in full_q:
desired_ent.update(helpers.named_entities_by_question[k])
# assemble question vocab
q_vocab = set()
for word in question:
if word in w2v:
q_vocab.add(word)
# vector similarity of each word in sentence with question
scores = []
contains_ent = [False] * N
for i, s in enumerate(sents):
scores.append([0] * len(s))
for j, word in enumerate(s):
if word.ent_type_ in desired_ent:
contains_ent[i] = True
elif word.lower_ not in stopwords and word.lower_ in w2v:
scores[i][j] = max(map(lambda x: w2v.similarity(word.lower_, x), q_vocab))
# average score for each sentence
sent_scores = [0] * N
for i, s in enumerate(scores):
sum_ = 0
len_ = 0
for num in s:
if num != 0:
sum_ += num
len_ += 1
if len_ != 0:
sent_scores[i] = sum_ / len_
# score modifiers
final_scores = [0] * N
for i, s in enumerate(sent_scores):
score = s
if i > 0 and i < len(sent_scores) - 1:
score += 0.25*sent_scores[i-1]
score += 0.25*sent_scores[i+1]
elif i > 0:
score += 0.5*sent_scores[i-1]
elif i < len(sent_scores) - 1:
score += 0.5*sent_scores[i+1]
if contains_ent[i]:
score += .3
if q_sent.root.lemma_ == sents[i].root.lemma_:
score += .2
if "why" in question and "because" in sents[i].lower_:
score += .3
if "when" in question and "when" in sents[i].lower_:
score += .2
final_scores[i] = score
while True:
# max score
max_score = max(final_scores)
max_idx = final_scores.index(max_score)
answer_sent = sents[max_idx]
# if answer is an entity
answer = ''
for w in answer_sent:
if w.ent_type_ in desired_ent and not [ent for ent in doc.ents if ent.start_char <= w.idx <= ent.end_char][0].lower_ in full_q:
if w.ent_type_ == "MONEY" and w.left_edge.is_currency and not w.left_edge.ent_type_ in desired_ent:
answer += w.left_edge.text
answer += w.text + ' '
# otherwise return the whole sentence
if answer == '':
answer = answer_sent.text
if not answer.endswith('?'):
break
final_scores[max_idx] = 0
else:
answer_story_questions.ner_count += 1
break
# hand-written rules
if "why" in question and "because" in answer_sent.lower_:
answer = "because" + answer_sent.lower_.split("because")[1]
if "where" in question and " at " in answer:
answer = "at " + answer.split(" at ")[1]
if "where" in question and " in " in answer:
answer = "in " + answer.split(" in ")[1]
for word in answer.split():
try:
if w2v.vocab[word.lower()].count < 299900 and re.search(r'(\s+|^)' + word.lower() + r'(\s+|$)', q_sent.lower_) is not None:
answer = re.sub(r'(\s+|^)' + word + r'(\s+|$)', ' ', answer)
except:
pass
if os.name == 'nt':
out.append(q_a[0])
out.append(f"Answer: {answer}\n\n")
else:
print(q_a[0])
print(f"Answer: {answer}\n\n")
if os.name == 'nt':
out_file.writelines(out)
# entry point
if os.name == 'nt':
start_time = time.perf_counter()
nlp = spacy.load('en_core_web_lg')
if os.name == 'nt':
stopwords = {w.strip() for w in open("./data/training/stopwords.txt").readlines()}
w2v = gensim.models.KeyedVectors.load("./data/training/word2vec.vectors", mmap='r')
out_file = open("all.answers", "w")
else:
stopwords = {w.strip() for w in open("./data/training/stopwords.txt").readlines()}
w2v = gensim.models.KeyedVectors.load("/home/u0907330/QA_data/word2vec.vectors", mmap='r')
infile_name = sys.argv[1]
infile = open(infile_name)
lines = infile.readlines()
path = lines[0].strip()
if not path.endswith('/'):
path += '/'
if os.name == 'nt':
print(f"Startup time: {time.perf_counter() - start_time}")
start_time = time.perf_counter()
answer_story_questions.ner_count = 0
for story_key in lines[1:]:
if story_key.startswith('#'):
break
answer_story_questions(path, story_key.strip())
if os.name == 'nt':
total = time.perf_counter() - start_time
print(f"Total time: {total}")
print(f"Time per story: {total / (len(lines)-1)}")
print(f"NER: {answer_story_questions.ner_count}")
out_file.close()