Spaces:
Sleeping
Sleeping
Commit
·
3c5755f
1
Parent(s):
b3179c4
experiment5
Browse files- absa_evaluator.py +113 -1
absa_evaluator.py
CHANGED
@@ -3,8 +3,10 @@ from typing import Dict, List
|
|
3 |
import evaluate
|
4 |
from datasets import Features, Sequence, Value
|
5 |
from sklearn.metrics import accuracy_score
|
|
|
|
|
|
|
6 |
|
7 |
-
from preprocessing import absa_term_preprocess
|
8 |
|
9 |
_CITATION = """
|
10 |
"""
|
@@ -164,3 +166,113 @@ class AbsaEvaluatorTest(evaluate.Metric):
|
|
164 |
"retrieved": retrieved,
|
165 |
"relevant": relevant,
|
166 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
import evaluate
|
4 |
from datasets import Features, Sequence, Value
|
5 |
from sklearn.metrics import accuracy_score
|
6 |
+
from itertools import chain
|
7 |
+
from random import choice
|
8 |
+
from typing import Any, Dict, List, Optional, Tuple
|
9 |
|
|
|
10 |
|
11 |
_CITATION = """
|
12 |
"""
|
|
|
166 |
"retrieved": retrieved,
|
167 |
"relevant": relevant,
|
168 |
}
|
169 |
+
|
170 |
+
def adjust_predictions(refs, preds, choices):
|
171 |
+
"""Adjust predictions to match the length of references with either a special token or random choice."""
|
172 |
+
adjusted_preds = []
|
173 |
+
for ref, pred in zip(refs, preds):
|
174 |
+
if len(pred) < len(ref):
|
175 |
+
missing_count = len(ref) - len(pred)
|
176 |
+
pred.extend([choice(choices) for _ in range(missing_count)])
|
177 |
+
adjusted_preds.append(pred)
|
178 |
+
return adjusted_preds
|
179 |
+
|
180 |
+
|
181 |
+
def extract_aspects(data, specific_key, specific_val):
|
182 |
+
"""Extracts and returns a list of specified aspect details from the nested 'aspects' data."""
|
183 |
+
return [item[specific_key][specific_val] for item in data]
|
184 |
+
|
185 |
+
|
186 |
+
def absa_term_preprocess(references, predictions, subtask_key, subtask_value):
|
187 |
+
"""
|
188 |
+
Preprocess the terms and polarities for aspect-based sentiment analysis.
|
189 |
+
|
190 |
+
Args:
|
191 |
+
references (List[Dict]): A list of dictionaries containing the actual terms and polarities under 'aspects'.
|
192 |
+
predictions (List[Dict]): A list of dictionaries containing predicted aspect categories to terms and their sentiments.
|
193 |
+
|
194 |
+
Returns:
|
195 |
+
Tuple[List[str], List[str], List[str], List[str]]: A tuple containing lists of true aspect terms,
|
196 |
+
adjusted predicted aspect terms, true polarities, and adjusted predicted polarities.
|
197 |
+
"""
|
198 |
+
|
199 |
+
# Extract aspect terms and polarities
|
200 |
+
truth_aspect_terms = extract_aspects(references, subtask_key, subtask_value)
|
201 |
+
pred_aspect_terms = extract_aspects(predictions, subtask_key, subtask_value)
|
202 |
+
truth_polarities = extract_aspects(references, subtask_key, "polarity")
|
203 |
+
pred_polarities = extract_aspects(predictions, subtask_key, "polarity")
|
204 |
+
|
205 |
+
# Define adjustment parameters
|
206 |
+
special_token = "NONE" # For missing aspect terms
|
207 |
+
sentiment_choices = [
|
208 |
+
"positive",
|
209 |
+
"negative",
|
210 |
+
"neutral",
|
211 |
+
"conflict",
|
212 |
+
] # For missing polarities
|
213 |
+
|
214 |
+
# Adjust the predictions to match the length of references
|
215 |
+
adjusted_pred_terms = adjust_predictions(
|
216 |
+
truth_aspect_terms, pred_aspect_terms, [special_token]
|
217 |
+
)
|
218 |
+
adjusted_pred_polarities = adjust_predictions(
|
219 |
+
truth_polarities, pred_polarities, sentiment_choices
|
220 |
+
)
|
221 |
+
|
222 |
+
return (
|
223 |
+
flatten_list(truth_aspect_terms),
|
224 |
+
flatten_list(adjusted_pred_terms),
|
225 |
+
flatten_list(truth_polarities),
|
226 |
+
flatten_list(adjusted_pred_polarities),
|
227 |
+
)
|
228 |
+
|
229 |
+
|
230 |
+
def flatten_list(nested_list):
|
231 |
+
"""Flatten a nested list into a single-level list."""
|
232 |
+
return list(chain.from_iterable(nested_list))
|
233 |
+
|
234 |
+
|
235 |
+
def extract_pred_terms(
|
236 |
+
all_predictions: List[Dict[str, Dict[str, str]]]
|
237 |
+
) -> List[List]:
|
238 |
+
"""Extract and organize predicted terms from the sentiment analysis results."""
|
239 |
+
pred_aspect_terms = []
|
240 |
+
for pred in all_predictions:
|
241 |
+
terms = [term for cat in pred.values() for term in cat.keys()]
|
242 |
+
pred_aspect_terms.append(terms)
|
243 |
+
return pred_aspect_terms
|
244 |
+
|
245 |
+
|
246 |
+
def merge_aspects_and_categories(aspects, categories):
|
247 |
+
result = []
|
248 |
+
|
249 |
+
# Assuming both lists are of the same length and corresponding indices match
|
250 |
+
for aspect, category in zip(aspects, categories):
|
251 |
+
combined_entry = {
|
252 |
+
"aspects": {"term": [], "polarity": []},
|
253 |
+
"category": {"category": [], "polarity": []},
|
254 |
+
}
|
255 |
+
|
256 |
+
# Process aspect entries
|
257 |
+
for cat_key, terms_dict in aspect.items():
|
258 |
+
for term, polarity in terms_dict.items():
|
259 |
+
combined_entry["aspects"]["term"].append(term)
|
260 |
+
combined_entry["aspects"]["polarity"].append(polarity)
|
261 |
+
|
262 |
+
# Add category details based on the aspect's key if available in categories
|
263 |
+
if cat_key in category:
|
264 |
+
combined_entry["category"]["category"].append(cat_key)
|
265 |
+
combined_entry["category"]["polarity"].append(
|
266 |
+
category[cat_key]
|
267 |
+
)
|
268 |
+
|
269 |
+
# Ensure all keys in category are accounted for
|
270 |
+
for cat_key, polarity in category.items():
|
271 |
+
if cat_key not in combined_entry["category"]["category"]:
|
272 |
+
combined_entry["category"]["category"].append(cat_key)
|
273 |
+
combined_entry["category"]["polarity"].append(polarity)
|
274 |
+
|
275 |
+
result.append(combined_entry)
|
276 |
+
|
277 |
+
return result
|
278 |
+
|