Create compactness_measurements.py
Browse files- compactness_measurements.py +111 -0
compactness_measurements.py
ADDED
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import stanza
|
3 |
+
|
4 |
+
nlp = stanza.Pipeline(lang='en', processors='tokenize,pos,lemma,depparse')
|
5 |
+
articles = ['a', 'an', 'the']
|
6 |
+
|
7 |
+
# input file is the compactIE output extractions on a set of sentences. set this variable accordingly.
|
8 |
+
INPUT_FILE = 'compactIE_predictions.txt'
|
9 |
+
|
10 |
+
|
11 |
+
def verb_count(part, sent):
|
12 |
+
s_doc = nlp(sent)
|
13 |
+
text2token = {}
|
14 |
+
for i in range(len(s_doc.sentences)):
|
15 |
+
tokens = [word.to_dict() for word in s_doc.sentences[i].words]
|
16 |
+
for t in tokens:
|
17 |
+
text2token[t["text"]] = t
|
18 |
+
doc = nlp(part)
|
19 |
+
doc = doc.sentences[0]
|
20 |
+
tokens = [word.to_dict() for word in doc.words]
|
21 |
+
verbs = 0
|
22 |
+
for token in tokens:
|
23 |
+
if (token['upos'] == 'VERB' and (token['deprel'] not in ['xcomp', 'amod', 'case', 'obl'])) or (token['upos'] == "AUX" and token['deprel'] == 'cop'):
|
24 |
+
try:
|
25 |
+
if text2token[token["text"]]["deprel"] == token['deprel']:
|
26 |
+
# print(token["text"], token["deprel"])
|
27 |
+
verbs += 1
|
28 |
+
except:
|
29 |
+
continue
|
30 |
+
return verbs
|
31 |
+
|
32 |
+
|
33 |
+
def clausal_constituents(extraction):
|
34 |
+
clausal_consts = 0
|
35 |
+
if extraction["predicate"].strip() != "":
|
36 |
+
pred_count = verb_count(extraction["predicate"], extraction["sentence"])
|
37 |
+
if pred_count > 1:
|
38 |
+
clausal_consts += pred_count - 1
|
39 |
+
|
40 |
+
if extraction["subject"].strip() != "":
|
41 |
+
clausal_consts += verb_count(extraction["subject"], extraction["sentence"])
|
42 |
+
|
43 |
+
if extraction["object"].strip() != "":
|
44 |
+
clausal_consts += verb_count(extraction["object"], extraction["sentence"])
|
45 |
+
# if clausal_consts > 0:
|
46 |
+
# print("clausal consts within extraction: ", extraction["subject"], extraction["predicate"], extraction["object"], clausal_consts)
|
47 |
+
return clausal_consts
|
48 |
+
|
49 |
+
|
50 |
+
if __name__ == "__main__":
|
51 |
+
extractions_df = pd.DataFrame(columns=["sentence", "subject", "predicate", "object"])
|
52 |
+
|
53 |
+
with open(INPUT_FILE, 'r') as f:
|
54 |
+
lines = f.readlines()
|
55 |
+
|
56 |
+
sentences = set()
|
57 |
+
for line in lines:
|
58 |
+
sentence, ext, score = line.split('\t')
|
59 |
+
sentences.add(sentence)
|
60 |
+
try:
|
61 |
+
arg1 = ext[ext.index('<arg1>') + 6:ext.index('</arg1>')]
|
62 |
+
except:
|
63 |
+
arg1 = ""
|
64 |
+
try:
|
65 |
+
rel = ext[ext.index('<rel>') + 5:ext.index('</rel>')]
|
66 |
+
except:
|
67 |
+
rel = ""
|
68 |
+
try:
|
69 |
+
arg2 = ext[ext.index('<arg2>') + 6:ext.index('</arg2>')]
|
70 |
+
except:
|
71 |
+
arg2 = ""
|
72 |
+
row = pd.DataFrame(
|
73 |
+
{"sentence": [sentence],
|
74 |
+
"subject": [arg1],
|
75 |
+
"predicate": [rel],
|
76 |
+
"object": [arg2]}
|
77 |
+
)
|
78 |
+
extractions_df = pd.concat([extractions_df, row])
|
79 |
+
|
80 |
+
# overlapping arguments
|
81 |
+
grouped_df = extractions_df.groupby("sentence")
|
82 |
+
total_number_of_arguments = 0
|
83 |
+
number_of_unique_arguments = 0
|
84 |
+
num_of_sentences = len(grouped_df.groups.keys())
|
85 |
+
for sent in grouped_df.groups.keys():
|
86 |
+
per_sentence_argument_set = set()
|
87 |
+
sen_group = grouped_df.get_group(sent).reset_index(drop=True)
|
88 |
+
extractions_list = list(sen_group.T.to_dict().values())
|
89 |
+
for extr in extractions_list:
|
90 |
+
if extr["subject"] not in ['', 'nan']:
|
91 |
+
total_number_of_arguments += 1
|
92 |
+
per_sentence_argument_set.add(extr["subject"])
|
93 |
+
if extr["object"] not in ['', 'nan']:
|
94 |
+
total_number_of_arguments += 1
|
95 |
+
per_sentence_argument_set.add(extr["object"])
|
96 |
+
number_of_unique_arguments += len(per_sentence_argument_set)
|
97 |
+
|
98 |
+
print("average # repetitions per argument: {}".format(total_number_of_arguments/number_of_unique_arguments))
|
99 |
+
print("average # extractions per sentence: {}".format(extractions_df.shape[0]/len(sentences)))
|
100 |
+
avg_arguments_size = 0.0
|
101 |
+
for sent in sentences:
|
102 |
+
extractions_per_sent = extractions_df[extractions_df["sentence"] == sent]
|
103 |
+
|
104 |
+
sent_extractions = extractions_per_sent.shape[0]
|
105 |
+
extractions_per_sent["avg_arg_length"] = extractions_per_sent.apply(lambda r: (len(str(r["subject"]).split(' ')) + len(str(r["predicate"]).split(' ')) + len(str(r["object"]).split(' ')))/3, axis=1)
|
106 |
+
avg_arguments_size += sum(extractions_per_sent["avg_arg_length"].values.tolist()) / extractions_per_sent.shape[0]
|
107 |
+
|
108 |
+
print("average length of constituents (per sentence, per extraction): ", avg_arguments_size/len(sentences))
|
109 |
+
extractions_df["clause_counts"] = extractions_df.apply(lambda r: clausal_constituents(r), axis=1)
|
110 |
+
avg_clause_count = sum(extractions_df["clause_counts"].values.tolist()) / len(sentences)
|
111 |
+
print("number of verbal clauses within arguments: ", avg_clause_count)
|