File size: 16,002 Bytes
3c332d2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 | """Build standardized ResearchBench JSONL data from the original materials."""
from __future__ import annotations
from collections import Counter
from pathlib import Path
from typing import Any
from .data_io import (
best_title_match,
clean_text,
contains_forbidden_text,
normalize_doi,
paper_pair,
read_json,
write_csv,
write_json,
write_jsonl,
)
from .schemas import DEFAULT_GENERATION_PARAMS, DEFAULT_RETRIEVE_PARAMS, LABEL_BY_FILE, LABEL_PRIORITY, SUBJECTS
def _partial_python_root(source_root: Path) -> Path:
return source_root / "researchbench的部分源代码" / "python"
def _distance_root(source_root: Path) -> Path:
return _partial_python_root(source_root) / "spider" / "distance"
def _ranking_roots(source_root: Path) -> list[Path]:
return [
source_root / "researchbench" / "python" / "ranking_fan" / "ranking",
_partial_python_root(source_root) / "ranking_fan" / "ranking",
]
def _sample_id(discipline: str, doi: str) -> str:
return f"{discipline}/{normalize_doi(doi)}"
def load_papers(source_root: str | Path) -> list[dict[str, Any]]:
source_root = Path(source_root)
rows: list[dict[str, Any]] = []
for discipline in SUBJECTS:
path = _partial_python_root(source_root) / discipline / "result.json"
data = read_json(path)
for idx, item in enumerate(data):
if clean_text(item.get("sufficiency_tag")).lower() != "yes":
continue
doi = clean_text(item.get("doi"))
inspirations = []
for insp in item.get("inspiration", []) or []:
inspirations.append(
{
"title": clean_text(insp.get("title")),
"abstract": clean_text(insp.get("abstract")),
"relation": clean_text(insp.get("relation")),
"insp": clean_text(insp.get("insp")),
}
)
rows.append(
{
"sample_id": _sample_id(discipline, doi),
"discipline": discipline,
"doi": doi,
"title": clean_text(item.get("title")),
"abstract": clean_text(item.get("abstract")),
"research_question": clean_text(item.get("research_question")),
"background_survey": clean_text(item.get("background_little_survey")),
"main_hypothesis": clean_text(item.get("main_hypothesis")),
"fine_grained_hypothesis": clean_text(item.get("fine_grained_hypothesis")),
"experiments_details": clean_text(item.get("experiments_details")),
"inspirations": inspirations,
"source_index": idx,
}
)
rows.sort(key=lambda row: row["sample_id"])
return rows
def _folder_map(root: Path) -> dict[tuple[str, str], Path]:
mapping: dict[tuple[str, str], Path] = {}
for discipline in SUBJECTS:
discipline_root = root / discipline
if not discipline_root.exists():
continue
for folder in discipline_root.iterdir():
if folder.is_dir():
mapping[(discipline, normalize_doi(folder.name))] = folder
return mapping
def _load_pairs(path: Path) -> list[dict[str, Any]]:
if not path.exists():
return []
pairs: list[dict[str, Any]] = []
for item in read_json(path):
parsed = paper_pair(item)
if parsed is None:
continue
title, abstract = parsed
pairs.append({"title": title, "abstract": abstract})
return pairs
def _label_candidates(folder: Path, candidates: list[dict[str, Any]]) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
label_title_maps: dict[str, list[dict[str, Any]]] = {}
for file_name, label in LABEL_BY_FILE.items():
label_title_maps[label] = _load_pairs(folder / file_name)
labeled = []
issues: list[dict[str, Any]] = []
for idx, candidate in enumerate(candidates):
label = "negative_unknown"
matched_file = ""
match_score = 0.0
for candidate_label in LABEL_PRIORITY:
label_candidates = label_title_maps.get(candidate_label, [])
match_idx, score = best_title_match(candidate["title"], label_candidates, threshold=0.72)
if match_idx is not None:
label = candidate_label
matched_file = next(k for k, v in LABEL_BY_FILE.items() if v == candidate_label)
match_score = score
break
match_score = max(match_score, score)
row = {
"index": idx,
"title": candidate["title"],
"abstract": candidate["abstract"],
"label": label,
}
if matched_file:
row["source_file"] = matched_file
labeled.append(row)
if label == "negative_unknown":
issues.append(
{
"candidate_index": idx,
"title": candidate["title"],
"best_match_score": f"{match_score:.3f}",
}
)
return labeled, issues
def build_retrieve_and_generation(
papers: list[dict[str, Any]],
source_root: str | Path,
) -> tuple[list[dict[str, Any]], list[dict[str, Any]], list[dict[str, Any]]]:
source_root = Path(source_root)
folders = _folder_map(_distance_root(source_root))
retrieve_rows: list[dict[str, Any]] = []
generation_rows: list[dict[str, Any]] = []
issues: list[dict[str, Any]] = []
for paper in papers:
key = (paper["discipline"], normalize_doi(paper["doi"]))
folder = folders.get(key)
if folder is None:
issues.append({"sample_id": paper["sample_id"], "issue": "missing_distance_folder"})
continue
merge = _load_pairs(folder / "merge.json")
if len(merge) != 75:
issues.append(
{
"sample_id": paper["sample_id"],
"issue": "candidate_count_not_75",
"value": len(merge),
}
)
candidates, candidate_issues = _label_candidates(folder, merge)
label_counts = Counter(candidate["label"] for candidate in candidates)
for issue in candidate_issues:
issues.append({"sample_id": paper["sample_id"], "issue": "candidate_unknown_label", **issue})
gold = [candidate for candidate in candidates if candidate["label"] == "gold"]
if not gold:
issues.append({"sample_id": paper["sample_id"], "issue": "no_gold_in_candidates"})
retrieve_rows.append(
{
"sample_id": paper["sample_id"],
"discipline": paper["discipline"],
"doi": paper["doi"],
"doi_folder": folder.name,
"research_question": paper["research_question"],
"background_survey": paper["background_survey"],
"candidates": candidates,
"gold_titles": [candidate["title"] for candidate in gold],
"label_counts": dict(label_counts),
"default_params": dict(DEFAULT_RETRIEVE_PARAMS),
}
)
generation_rows.append(
{
"sample_id": paper["sample_id"],
"discipline": paper["discipline"],
"doi": paper["doi"],
"research_question": paper["research_question"],
"background_survey": paper["background_survey"],
"gold_inspirations": [
{"title": candidate["title"], "abstract": candidate["abstract"]} for candidate in gold
],
"true_retrieve": [
{"title": candidate["title"], "abstract": candidate["abstract"]} for candidate in gold
],
"gold_hypothesis": paper["main_hypothesis"],
"fine_grained_hypothesis": paper["fine_grained_hypothesis"],
"experiments_details": paper["experiments_details"],
"default_params": dict(DEFAULT_GENERATION_PARAMS),
}
)
return retrieve_rows, generation_rows, issues
def build_ranking(
papers: list[dict[str, Any]],
source_root: str | Path,
) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
source_root = Path(source_root)
roots = [root for root in _ranking_roots(source_root) if root.exists()]
rows: list[dict[str, Any]] = []
issues: list[dict[str, Any]] = []
def find_ranking_file(paper: dict[str, Any]) -> Path | None:
for root in roots:
discipline_root = root / paper["discipline"]
if not discipline_root.exists():
continue
for prefix in ("random_", ""):
expected = discipline_root / f"{prefix}{paper['doi'].replace('/', '-')}.json"
if expected.exists():
return expected
candidates = []
for path in discipline_root.glob("*.json"):
stem = path.stem.removeprefix("random_")
if normalize_doi(stem) == normalize_doi(paper["doi"]):
candidates.append(path)
if candidates:
candidates.sort(key=lambda path: (0 if path.name.startswith("random_") else 1, path.name))
return candidates[0]
return None
for paper in papers:
path = find_ranking_file(paper)
if path is None:
issues.append({"sample_id": paper["sample_id"], "issue": "missing_ranking_task"})
continue
data = read_json(path)
fake = [clean_text(x) for x in data.get("fake generate hypothesis", []) if clean_text(x)]
model = [clean_text(x) for x in data.get("model generate hypothesis", []) if clean_text(x)]
if len(fake) < 5 or len(model) < 10:
issues.append(
{
"sample_id": paper["sample_id"],
"issue": "ranking_negative_count_too_small",
"fake_count": len(fake),
"model_count": len(model),
}
)
continue
rows.append(
{
"sample_id": paper["sample_id"],
"discipline": paper["discipline"],
"doi": paper["doi"],
"research_question": clean_text(data.get("Background Question")) or paper["research_question"],
"gold_hypothesis": clean_text(data.get("Main hypothesis")) or paper["main_hypothesis"],
"fake_negative_hypotheses": fake[:5],
"model_negative_hypotheses": model[:10],
"source_variant": "random" if path.name.startswith("random_") else "direct",
}
)
rows.sort(key=lambda row: row["sample_id"])
return rows, issues
def _tiny_rows(rows: list[dict[str, Any]], limit: int = 12) -> list[dict[str, Any]]:
by_subject: dict[str, dict[str, Any]] = {}
for row in rows:
by_subject.setdefault(row["discipline"], row)
selected = list(by_subject.values())
if len(selected) >= limit:
return selected[:limit]
selected_ids = {row["sample_id"] for row in selected}
for row in rows:
if row["sample_id"] not in selected_ids:
selected.append(row)
selected_ids.add(row["sample_id"])
if len(selected) >= limit:
break
return selected
def build_all(source_root: str | Path, out: str | Path) -> dict[str, int]:
out = Path(out)
source_root = Path(source_root)
papers = load_papers(source_root)
retrieve, generation, retrieve_issues = build_retrieve_and_generation(papers, source_root)
ranking, ranking_issues = build_ranking(papers, source_root)
write_jsonl(out / "papers.jsonl", papers)
write_jsonl(out / "tasks" / "retrieve.jsonl", retrieve)
write_jsonl(out / "tasks" / "generation.jsonl", generation)
write_jsonl(out / "tasks" / "ranking.jsonl", ranking)
write_jsonl(out / "tiny" / "papers.jsonl", _tiny_rows(papers))
write_jsonl(out / "tiny" / "retrieve.jsonl", _tiny_rows(retrieve))
write_jsonl(out / "tiny" / "generation.jsonl", _tiny_rows(generation))
write_jsonl(out / "tiny" / "ranking.jsonl", _tiny_rows(ranking))
write_csv(
out / "validation" / "retrieve_build_issues.csv",
retrieve_issues,
["sample_id", "issue", "candidate_index", "title", "best_match_score", "value"],
)
write_csv(
out / "validation" / "ranking_build_issues.csv",
ranking_issues,
["sample_id", "issue", "fake_count", "model_count"],
)
summary = {
"papers": len(papers),
"retrieve": len(retrieve),
"generation": len(generation),
"ranking": len(ranking),
"retrieve_issues": len(retrieve_issues),
"ranking_issues": len(ranking_issues),
}
write_json(out / "validation" / "build_summary.json", summary)
return summary
def validate_data_dir(data_dir: str | Path) -> tuple[bool, list[dict[str, Any]]]:
data_dir = Path(data_dir)
issues: list[dict[str, Any]] = []
def add(file_name: str, sample_id: str, issue: str, value: Any = "") -> None:
issues.append({"file": file_name, "sample_id": sample_id, "issue": issue, "value": value})
from .data_io import iter_jsonl
for relative in [
"papers.jsonl",
"tasks/retrieve.jsonl",
"tasks/generation.jsonl",
"tasks/ranking.jsonl",
]:
path = data_dir / relative
if not path.exists():
add(relative, "", "missing_file")
continue
seen: set[str] = set()
for row in iter_jsonl(path):
sample_id = clean_text(row.get("sample_id"))
if not sample_id:
add(relative, "", "missing_sample_id")
elif sample_id in seen:
add(relative, sample_id, "duplicate_sample_id")
seen.add(sample_id)
if contains_forbidden_text(row):
add(relative, sample_id, "forbidden_local_path_or_api_key")
if relative == "papers.jsonl":
for key in ["research_question", "main_hypothesis", "doi", "discipline"]:
if not clean_text(row.get(key)):
add(relative, sample_id, f"empty_{key}")
elif relative == "tasks/retrieve.jsonl":
candidates = row.get("candidates", [])
if len(candidates) != 75:
add(relative, sample_id, "candidate_count_not_75", len(candidates))
if not row.get("gold_titles"):
add(relative, sample_id, "empty_gold_titles")
if not any(candidate.get("label") == "gold" for candidate in candidates):
add(relative, sample_id, "no_gold_candidate")
if not clean_text(row.get("research_question")):
add(relative, sample_id, "empty_research_question")
elif relative == "tasks/generation.jsonl":
if not row.get("gold_inspirations"):
add(relative, sample_id, "empty_gold_inspirations")
if not clean_text(row.get("gold_hypothesis")):
add(relative, sample_id, "empty_gold_hypothesis")
elif relative == "tasks/ranking.jsonl":
if len(row.get("fake_negative_hypotheses", [])) != 5:
add(relative, sample_id, "fake_negative_count_not_5")
if len(row.get("model_negative_hypotheses", [])) != 10:
add(relative, sample_id, "model_negative_count_not_10")
write_csv(
data_dir / "validation" / "validation_issues.csv",
issues,
["file", "sample_id", "issue", "value"],
)
return not issues, issues
|