Spaces:
Sleeping
Sleeping
File size: 8,477 Bytes
75bea1c | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 | from __future__ import annotations
"""Quality evaluation for responses."""
from dataclasses import dataclass
from typing import Any
@dataclass
class QualityScore:
"""Quality evaluation scores."""
relevance: float # 0-1: How relevant is the answer
completeness: float # 0-1: How complete is the answer
accuracy: float # 0-1: Estimated accuracy
clarity: float # 0-1: How clear is the answer
sourcing: float # 0-1: Quality of source citations
overall: float # 0-1: Overall quality score
feedback: list[str] # Specific feedback items
class QualityEvaluator:
"""Evaluates the quality of generated responses."""
def __init__(self, min_quality_threshold: float = 0.6):
"""Initialize the evaluator.
Args:
min_quality_threshold: Minimum acceptable quality score
"""
self.min_quality_threshold = min_quality_threshold
def evaluate(
self,
query: str,
answer: str,
sources: list[dict[str, str]] | None = None,
reasoning_steps: list[str] | None = None,
) -> QualityScore:
"""Evaluate the quality of a response.
Args:
query: Original user query
answer: Generated answer
sources: List of source citations
reasoning_steps: Reasoning steps taken
Returns:
QualityScore with detailed evaluation
"""
feedback = []
# Evaluate relevance
relevance = self._evaluate_relevance(query, answer)
if relevance < 0.5:
feedback.append("Answer may not be relevant to the question")
# Evaluate completeness
completeness = self._evaluate_completeness(query, answer)
if completeness < 0.5:
feedback.append("Answer appears incomplete")
# Evaluate accuracy (based on source count and reasoning)
accuracy = self._evaluate_accuracy(sources, reasoning_steps)
if accuracy < 0.5:
feedback.append("Accuracy could not be verified with sources")
# Evaluate clarity
clarity = self._evaluate_clarity(answer)
if clarity < 0.5:
feedback.append("Answer could be clearer")
# Evaluate sourcing
sourcing = self._evaluate_sourcing(answer, sources)
if sourcing < 0.5:
feedback.append("More sources would improve credibility")
# Calculate overall score (weighted average)
overall = (
relevance * 0.25
+ completeness * 0.2
+ accuracy * 0.25
+ clarity * 0.15
+ sourcing * 0.15
)
if overall >= self.min_quality_threshold:
feedback.insert(0, "Response meets quality standards")
else:
feedback.insert(0, "Response may need refinement")
return QualityScore(
relevance=relevance,
completeness=completeness,
accuracy=accuracy,
clarity=clarity,
sourcing=sourcing,
overall=overall,
feedback=feedback,
)
def is_acceptable(self, score: QualityScore) -> bool:
"""Check if quality score is acceptable.
Args:
score: Quality score to check
Returns:
True if acceptable
"""
return score.overall >= self.min_quality_threshold
def _evaluate_relevance(self, query: str, answer: str) -> float:
"""Evaluate answer relevance to query.
Args:
query: User query
answer: Generated answer
Returns:
Relevance score (0-1)
"""
if not answer:
return 0.0
# Simple keyword matching
query_words = set(query.lower().split())
answer_words = set(answer.lower().split())
# Remove common words
stopwords = {"the", "a", "an", "is", "are", "was", "were", "what", "how", "when", "where", "why", "who"}
query_words -= stopwords
answer_words -= stopwords
if not query_words:
return 0.5
overlap = len(query_words & answer_words)
return min(1.0, overlap / len(query_words) + 0.3) # Base score + overlap
def _evaluate_completeness(self, query: str, answer: str) -> float:
"""Evaluate answer completeness.
Args:
query: User query
answer: Generated answer
Returns:
Completeness score (0-1)
"""
if not answer:
return 0.0
# Check answer length relative to query complexity
query_words = len(query.split())
answer_words = len(answer.split())
# Longer queries typically need longer answers
expected_min = max(20, query_words * 3)
if answer_words < expected_min:
return answer_words / expected_min
# Check for explanation patterns
explanation_markers = ["because", "since", "therefore", "this means", "in other words"]
has_explanation = any(marker in answer.lower() for marker in explanation_markers)
score = 0.7
if has_explanation:
score += 0.2
if answer_words > expected_min * 2:
score += 0.1
return min(1.0, score)
def _evaluate_accuracy(
self,
sources: list[dict[str, str]] | None,
reasoning_steps: list[str] | None,
) -> float:
"""Evaluate estimated accuracy.
Args:
sources: List of sources
reasoning_steps: Reasoning steps
Returns:
Accuracy score (0-1)
"""
score = 0.3 # Base score
# More sources = higher potential accuracy
if sources:
score += min(0.3, len(sources) * 0.1)
# Reasoning steps suggest careful analysis
if reasoning_steps:
score += min(0.3, len(reasoning_steps) * 0.1)
# Cap at 0.9 since we can't truly verify accuracy
return min(0.9, score)
def _evaluate_clarity(self, answer: str) -> float:
"""Evaluate answer clarity.
Args:
answer: Generated answer
Returns:
Clarity score (0-1)
"""
if not answer:
return 0.0
score = 0.5
# Check sentence structure (average length)
sentences = answer.split(".")
if sentences:
avg_sentence_length = len(answer.split()) / len(sentences)
# Ideal: 15-25 words per sentence
if 10 <= avg_sentence_length <= 30:
score += 0.2
# Check for structure (paragraphs, lists)
if "\n" in answer:
score += 0.1
if any(marker in answer for marker in ["-", "•", "1.", "2."]):
score += 0.1
# Check for hedge words (too many = less clear)
hedge_words = ["might", "perhaps", "maybe", "possibly", "could"]
hedge_count = sum(1 for word in hedge_words if word in answer.lower())
if hedge_count > 3:
score -= 0.1
return min(1.0, max(0.0, score))
def _evaluate_sourcing(
self,
answer: str,
sources: list[dict[str, str]] | None,
) -> float:
"""Evaluate source quality.
Args:
answer: Generated answer
sources: List of sources
Returns:
Sourcing score (0-1)
"""
if not sources:
return 0.2
score = 0.3
# More sources = better
source_count = len(sources)
score += min(0.3, source_count * 0.1)
# Check for diverse domains
urls = [s.get("url", "") for s in sources]
domains = set()
for url in urls:
if url:
try:
from urllib.parse import urlparse
domain = urlparse(url).netloc
domains.add(domain)
except Exception:
pass
# Domain diversity
if len(domains) > 1:
score += 0.2
# Check for reliable domains
reliable_indicators = [".gov", ".edu", "wikipedia.org"]
for url in urls:
if any(ind in url.lower() for ind in reliable_indicators):
score += 0.1
break
return min(1.0, score)
|