scan-v5.json (26824B)
1 { 2 "scan_version": 5, 3 "paper_type": "empirical", 4 "paper": { 5 "title": "DEAR: A Novel Deep Learning-based Approach for Automated Program Repair", 6 "authors": [ 7 "Yi Li", 8 "Shaohua Wang", 9 "Tien N. Nguyen" 10 ], 11 "year": 2022, 12 "venue": "International Conference on Software Engineering", 13 "arxiv_id": "2205.01859", 14 "doi": "10.1145/3510003.3510177" 15 }, 16 "checklist": { 17 "claims_and_evidence": { 18 "abstract_claims_supported": { 19 "applies": true, 20 "answer": true, 21 "justification": "All major abstract claims (42%-683% improvement on Defects4J, 31-145 more bugs on BigFix, 169 multi-hunk bugs among 667 fixed on CPatMiner) are directly supported by Tables 1, 4, and 6.", 22 "source": "haiku" 23 }, 24 "causal_claims_justified": { 25 "applies": true, 26 "answer": true, 27 "justification": "Causal claims about each component's contribution are backed by the ablation study in RQ4/Table 9, which systematically removes hunk detection, expansion, and attention+cycle training to measure individual impact.", 28 "source": "haiku" 29 }, 30 "generalization_bounded": { 31 "applies": true, 32 "answer": true, 33 "justification": "The paper explicitly bounds results to Java code and lists this in the Limitations section, noting that only the FL and post-processing modules are language-dependent.", 34 "source": "haiku" 35 }, 36 "alternative_explanations_discussed": { 37 "applies": true, 38 "answer": false, 39 "justification": "The paper does not discuss whether training data volume differences, hyperparameter search advantages, or benchmark-specific factors (rather than architectural innovations) could explain the observed gains over baselines.", 40 "source": "haiku" 41 }, 42 "proxy_outcome_distinction": { 43 "applies": true, 44 "answer": true, 45 "justification": "The paper explicitly distinguishes 'correct patches' (matching developer fixes) from 'plausible patches' (passing all tests but not necessarily the ground truth fix), reporting both counts separately in all tables.", 46 "source": "haiku" 47 } 48 }, 49 "limitations_and_scope": { 50 "limitations_section_present": { 51 "applies": true, 52 "answer": true, 53 "justification": "The paper contains both a 'Threats to Validity' paragraph and a dedicated 'Limitations' section with five enumerated specific limitations.", 54 "source": "haiku" 55 }, 56 "threats_to_validity_specific": { 57 "applies": true, 58 "answer": true, 59 "justification": "Specific threats named include: Java-only evaluation, the need to reimplement CURE (unavailable), and use of a 5-hour time limit for pattern-based tools; these are specific, not boilerplate.", 60 "source": "haiku" 61 }, 62 "scope_boundaries_stated": { 63 "applies": true, 64 "answer": true, 65 "justification": "Explicit scope boundaries state: Java only, only bugs causing failing tests, cannot fix bugs requiring many new inserted statements, and expansion may incorrectly include non-buggy statements.", 66 "source": "haiku" 67 } 68 }, 69 "conflicts_of_interest": { 70 "funding_disclosed": { 71 "applies": true, 72 "answer": true, 73 "justification": "NSF funding disclosed in Acknowledgments with specific grant numbers: CNS-2120386, CCF-1723215, CCF-1723432, TWC-1723198, CCF-1518897, CNS-1513263.", 74 "source": "haiku" 75 }, 76 "affiliations_disclosed": { 77 "applies": true, 78 "answer": true, 79 "justification": "Author affiliations (NJIT and UT Dallas) are listed on the title page with email addresses.", 80 "source": "haiku" 81 }, 82 "funder_independent_of_outcome": { 83 "applies": true, 84 "answer": true, 85 "justification": "NSF is a government funding agency with no commercial stake in the outcome; authors are not evaluating a product they have financial ties to.", 86 "source": "haiku" 87 }, 88 "financial_interests_declared": { 89 "applies": true, 90 "answer": false, 91 "justification": "No competing interests statement or declaration of financial interests (patents, equity, consulting) appears anywhere in the paper.", 92 "source": "haiku" 93 } 94 }, 95 "scope_and_framing": { 96 "key_terms_defined": { 97 "applies": true, 98 "answer": true, 99 "justification": "APR, multi-hunk/multi-statement bugs, spectrum-based FL, plausible vs. correct patches, and bug types (Types 1-5) are all defined with examples in Sections 1-2.", 100 "source": "haiku" 101 }, 102 "intended_contribution_clear": { 103 "applies": true, 104 "answer": true, 105 "justification": "Three key contributions are explicitly enumerated in the introduction: FL for multi-hunk bugs, compositional divide-and-conquer fixing, and the enhanced two-layer LSTM with attention and cycle training.", 106 "source": "haiku" 107 }, 108 "engagement_with_prior_work": { 109 "applies": true, 110 "answer": true, 111 "justification": "Section 7 extensively covers both DL-based and pattern-based APR prior work, and the introduction situates DEAR by identifying the specific limitation (individual-statement-only fixing) in existing DL approaches.", 112 "source": "haiku" 113 } 114 } 115 }, 116 "type_checklist": { 117 "empirical": { 118 "artifacts": { 119 "code_released": { 120 "applies": true, 121 "answer": true, 122 "justification": "GitHub repository explicitly cited: https://github.com/AutomatedProgramRepair-2021/dear-auto-fix, with the contribution statement 'Our data and tool are publicly available.'", 123 "source": "haiku" 124 }, 125 "data_released": { 126 "applies": true, 127 "answer": true, 128 "justification": "All three evaluation datasets (Defects4J, BigFix, CPatMiner) are publicly available benchmarks from cited prior work.", 129 "source": "haiku" 130 }, 131 "environment_specified": { 132 "applies": true, 133 "answer": false, 134 "justification": "Only hardware specs are provided (8-core Intel CPU, GTX Titan GPU); no requirements.txt, Dockerfile, or software dependency list appears in the paper.", 135 "source": "haiku" 136 }, 137 "reproduction_instructions": { 138 "applies": true, 139 "answer": false, 140 "justification": "The paper describes the algorithm in detail but provides no step-by-step reproduction instructions; a reader would need to infer setup from the GitHub repo, which is not described in the paper itself.", 141 "source": "haiku" 142 } 143 }, 144 "statistical_methodology": { 145 "confidence_intervals_or_error_bars": { 146 "applies": true, 147 "answer": false, 148 "justification": "No confidence intervals or error bars are reported for any results; all comparative claims are based on raw counts and percentages without statistical uncertainty.", 149 "source": "haiku" 150 }, 151 "significance_tests": { 152 "applies": true, 153 "answer": false, 154 "justification": "No statistical significance tests (Wilcoxon, t-test, etc.) are applied to any of the comparative claims across all five RQs.", 155 "source": "haiku" 156 }, 157 "effect_sizes_reported": { 158 "applies": true, 159 "answer": true, 160 "justification": "Relative improvements are reported throughout (42%-683% over baselines, 31-145 more bugs fixed on BigFix) with baseline counts providing context for interpreting magnitude.", 161 "source": "haiku" 162 }, 163 "sample_size_justified": { 164 "applies": true, 165 "answer": false, 166 "justification": "The paper uses established benchmarks without justifying whether the 395-bug Defects4J set or the 80/10/10 split ratios are adequate for the specific claims made.", 167 "source": "haiku" 168 }, 169 "variance_reported": { 170 "applies": true, 171 "answer": false, 172 "justification": "No variance, standard deviation, or results across multiple runs are reported; it is unclear whether experiments were run once or repeatedly.", 173 "source": "haiku" 174 } 175 }, 176 "evaluation_design": { 177 "baselines_included": { 178 "applies": true, 179 "answer": true, 180 "justification": "Six DL-based APR baselines (DLFix, CoCoNuT, SequenceR, Tufano19, CODIT, CURE) and eight pattern-based APR tools are included across the RQs.", 181 "source": "haiku" 182 }, 183 "baselines_contemporary": { 184 "applies": true, 185 "answer": true, 186 "justification": "Baselines include CURE (ICSE 2021) and CoCoNuT (ISSTA 2020), which are the most recent competitive DL-based APR approaches at time of publication.", 187 "source": "haiku" 188 }, 189 "ablation_study": { 190 "applies": true, 191 "answer": true, 192 "justification": "RQ4/Table 9 removes three components individually (hunk detection, multi-statement expansion, attention+cycle training) and measures the performance drop on Defects4J.", 193 "source": "haiku" 194 }, 195 "multiple_metrics": { 196 "applies": true, 197 "answer": true, 198 "justification": "Correct patches, plausible patches, top-K accuracy (K=1,3,5), per-bug-type breakdowns (Types 1-5), and training parameter counts are all reported.", 199 "source": "haiku" 200 }, 201 "human_evaluation": { 202 "applies": false, 203 "answer": false, 204 "justification": "Human evaluation is not applicable for an APR tool benchmarked against ground-truth developer fixes and validated by automated test cases.", 205 "source": "haiku" 206 }, 207 "held_out_test_set": { 208 "applies": true, 209 "answer": true, 210 "justification": "For BigFix/CPatMiner an 80%/10%/10% train/tune/test split is used; Defects4J serves as a held-out test with explicit no-overlap guarantee against the CPatMiner training set.", 211 "source": "haiku" 212 }, 213 "per_category_breakdown": { 214 "applies": true, 215 "answer": true, 216 "justification": "Results broken down by five bug types (Type 1: one-hunk/one-statement through Type 5: multi-hunk/mix-statements) in Tables 2, 3, 6, 8, and 9.", 217 "source": "haiku" 218 }, 219 "failure_cases_discussed": { 220 "applies": true, 221 "answer": true, 222 "justification": "The Limitations section enumerates specific failure modes: rare/out-of-vocabulary names, security/non-failing-test bugs, fixes requiring many new statements, and incorrect multi-statement expansion.", 223 "source": "haiku" 224 }, 225 "negative_results_reported": { 226 "applies": true, 227 "answer": false, 228 "justification": "The ablation shows performance drops when components are removed, but no design decisions that failed or approaches tried and abandoned are reported.", 229 "source": "haiku" 230 } 231 }, 232 "setup_transparency": { 233 "model_versions_specified": { 234 "applies": true, 235 "answer": false, 236 "justification": "The paper refers to 'Google's pre-trained BERT model' without specifying the checkpoint (BERT-base vs. BERT-large) or version used for fine-tuning.", 237 "source": "haiku" 238 }, 239 "prompts_provided": { 240 "applies": false, 241 "answer": false, 242 "justification": "DEAR is a non-LLM deep learning model with no prompts; this criterion is not applicable.", 243 "source": "haiku" 244 }, 245 "hyperparameters_reported": { 246 "applies": true, 247 "answer": true, 248 "justification": "Hyperparameters are reported with full search ranges and best values: epoch size, batch size, learning rate for BERT and LSTM; vector size, learning rate, batch size, epoch size for GloVe.", 249 "source": "haiku" 250 }, 251 "scaffolding_described": { 252 "applies": false, 253 "answer": false, 254 "justification": "No agentic scaffolding used; DEAR is a traditional DL pipeline, not an agent-based system.", 255 "source": "haiku" 256 }, 257 "data_preprocessing_documented": { 258 "applies": true, 259 "answer": true, 260 "justification": "Preprocessing steps are documented in detail: AST parsing, alpha-renaming, GloVe encoding, TreeCaps summarization, CPatMiner-based subtree pairing rules, and the four pairing rules for buggy/fixed subtrees.", 261 "source": "haiku" 262 } 263 }, 264 "data_integrity": { 265 "raw_data_available": { 266 "applies": true, 267 "answer": true, 268 "justification": "Public benchmarks (Defects4J, BigFix, CPatMiner) are used; Defects4J in particular is a well-established independently reproducible benchmark with versioned releases.", 269 "source": "haiku" 270 }, 271 "data_collection_described": { 272 "applies": true, 273 "answer": true, 274 "justification": "Section 5.2 describes the data sources (44k+ bugs from 5,832 Java projects in CPatMiner, 26k+ in BigFix, 395 in Defects4J) and the 80/10/10 split procedure.", 275 "source": "haiku" 276 }, 277 "recruitment_methods_described": { 278 "applies": false, 279 "answer": false, 280 "justification": "Standard public benchmarks are used with no human participant recruitment.", 281 "source": "haiku" 282 }, 283 "data_pipeline_documented": { 284 "applies": true, 285 "answer": true, 286 "justification": "The full pipeline from source code → AST parsing → subtree pairing → context building → model training → patch generation → test validation is documented across Sections 3-4.", 287 "source": "haiku" 288 } 289 }, 290 "contamination": { 291 "training_cutoff_stated": { 292 "applies": false, 293 "answer": false, 294 "justification": "DEAR trains its own models from scratch on specific datasets; there is no pre-trained LLM with a training cutoff being evaluated.", 295 "source": "haiku" 296 }, 297 "train_test_overlap_discussed": { 298 "applies": true, 299 "answer": true, 300 "justification": "The paper explicitly states 'no overlap between the two datasets' for CPatMiner (training) and Defects4J (testing), and uses separate held-out splits for large dataset evaluation.", 301 "source": "haiku" 302 }, 303 "benchmark_contamination_addressed": { 304 "applies": false, 305 "answer": false, 306 "justification": "DEAR is not an LLM evaluated on pre-existing benchmarks; the model is trained from scratch with explicit train/test separation.", 307 "source": "haiku" 308 } 309 }, 310 "human_studies": { 311 "pre_registered": { 312 "applies": false, 313 "answer": false, 314 "justification": "No human participants in this study.", 315 "source": "haiku" 316 }, 317 "irb_or_ethics_approval": { 318 "applies": false, 319 "answer": false, 320 "justification": "No human participants in this study.", 321 "source": "haiku" 322 }, 323 "demographics_reported": { 324 "applies": false, 325 "answer": false, 326 "justification": "No human participants in this study.", 327 "source": "haiku" 328 }, 329 "inclusion_exclusion_criteria": { 330 "applies": false, 331 "answer": false, 332 "justification": "No human participants in this study.", 333 "source": "haiku" 334 }, 335 "randomization_described": { 336 "applies": false, 337 "answer": false, 338 "justification": "No human participants in this study.", 339 "source": "haiku" 340 }, 341 "blinding_described": { 342 "applies": false, 343 "answer": false, 344 "justification": "No human participants in this study.", 345 "source": "haiku" 346 }, 347 "attrition_reported": { 348 "applies": false, 349 "answer": false, 350 "justification": "No human participants in this study.", 351 "source": "haiku" 352 } 353 }, 354 "cost_and_practicality": { 355 "inference_cost_reported": { 356 "applies": true, 357 "answer": true, 358 "justification": "Prediction time per candidate patch is reported: 2.4-3.1 seconds for CPatMiner, 3.6-4.2 for BigFix, 2.1 for Defects4J.", 359 "source": "haiku" 360 }, 361 "compute_budget_stated": { 362 "applies": true, 363 "answer": true, 364 "justification": "Training time stated for each dataset (22+ hours for CPatMiner, 18-19 hours for BigFix) along with hardware (8-core Intel CPU, GTX Titan GPU).", 365 "source": "haiku" 366 } 367 } 368 } 369 }, 370 "claims": [ 371 { 372 "claim": "DEAR outperforms all DL-based APR baselines on Defects4J by 42%-683% in terms of auto-fixed bugs using top-1 patches.", 373 "evidence": "Table 1: DEAR fixes 47 bugs vs. CURE 36, CoCoNuT 33, DLFix 30, Tufano19 14, Sequencer 15, CODIT 6.", 374 "supported": "strong" 375 }, 376 { 377 "claim": "DEAR is the first DL-based APR model to fix multi-hunk/multi-statement bugs; all prior DL approaches fix zero such bugs.", 378 "evidence": "Table 2: DLFix, CoCoNuT, and CURE each fix 0 bugs of Types 2-5 on Defects4J, while DEAR fixes 18.", 379 "supported": "strong" 380 }, 381 { 382 "claim": "DEAR achieves comparable and complementary results to the best pattern-based APR tools on Defects4J.", 383 "evidence": "Table 7: DEAR fixes 47 bugs vs. Hercules 49 and Tbar 43; Table 8 shows DEAR fixes 7 multi-statement bugs Hercules misses entirely.", 384 "supported": "strong" 385 }, 386 { 387 "claim": "DEAR requires 7x fewer training parameters than CURE while achieving better repair performance.", 388 "evidence": "RQ5: DEAR requires 0.39M parameters vs. CURE's 3.1M on CPatMiner (0.42M vs. 3.5M on BigFix), while outperforming CURE on all datasets.", 389 "supported": "strong" 390 }, 391 { 392 "claim": "Multi-statement expansion contributes to fixing more uniquely challenging bugs than hunk detection.", 393 "evidence": "Table 9: Without expansion, 7 multi-statement bugs (Types 2,4,5) are lost; without hunk detection, 14 multi-hunk bugs are lost but only 3 of those are of the hardest types (4-5).", 394 "supported": "moderate" 395 }, 396 { 397 "claim": "DEAR generalizes across datasets, consistently outperforming baselines in cross-dataset evaluation.", 398 "evidence": "Table 5: DEAR achieves best top-1 results in both cross-dataset directions (7.5% for CPatMiner→BigFix, 9.6% for BigFix→CPatMiner).", 399 "supported": "strong" 400 } 401 ], 402 "methodology_tags": [ 403 "benchmark-eval" 404 ], 405 "key_findings": "DEAR is the first deep learning-based APR approach to fix multi-hunk, multi-statement bugs through a pipeline combining BERT-based hunk detection, RNN+data-flow multi-statement expansion, and a two-tier tree LSTM with attention and cycle training. On Defects4J it fixes 47 bugs — 31% more than the best DL baseline (CURE at 36) and comparable to top pattern-based tools — while uniquely fixing 18 multi-hunk/multi-statement bugs that all prior DL approaches fail on entirely. The approach requires 7x fewer training parameters than CURE while achieving better results, and performance scales monotonically with training data size across the 70/80/90% split experiments.", 406 "red_flags": [ 407 { 408 "flag": "No statistical tests", 409 "detail": "All comparative claims rely on raw counts without statistical significance tests, confidence intervals, or p-values. On Defects4J the margin is 11 bugs (47 vs 36), which could plausibly be within chance variation with no significance established." 410 }, 411 { 412 "flag": "CURE reimplemented by authors", 413 "detail": "The top-performing baseline CURE was 'unavailable' and reimplemented by the DEAR authors, introducing potential implementation bias; a reimplementation may underperform the original." 414 }, 415 { 416 "flag": "No variance across runs", 417 "detail": "No information on whether experiments were run multiple times; with stochastic training, variance across seeds is unknown and results may not be reproducible." 418 }, 419 { 420 "flag": "Plausible patch ratio", 421 "detail": "DEAR generates 91 plausible patches vs. 47 correct ones on Defects4J — nearly a 2:1 ratio — raising questions about patch quality and whether test suites provide adequate validation signal." 422 }, 423 { 424 "flag": "Java-only generalization claims", 425 "detail": "The paper claims key modules are 'language-independent' but all evaluation is Java-only; no evidence supports the generalization claim." 426 } 427 ], 428 "cited_papers": [ 429 { 430 "title": "DLFix: Context-Based Code Transformation Learning for Automated Program Repair", 431 "relevance": "Direct predecessor and primary baseline; DEAR extends DLFix's two-layer tree LSTM with attention, cycle training, and multi-hunk/multi-statement support." 432 }, 433 { 434 "title": "CURE: Code-Aware Neural Machine Translation for Automatic Program Repair", 435 "relevance": "Top-performing DL APR baseline (ICSE 2021); DEAR outperforms it with 7x fewer training parameters." 436 }, 437 { 438 "title": "CoCoNuT: Combining Context-Aware Neural Translation Models Using Ensemble for Program Repair", 439 "relevance": "Key DL APR baseline compared across all three datasets; fixes 0 multi-hunk/multi-statement bugs vs. DEAR's 18." 440 }, 441 { 442 "title": "Graph-Based Mining of in-the-Wild, Fine-Grained, Semantic Code Change Patterns (CPatMiner)", 443 "relevance": "Provides the primary training dataset (44k+ bugs) and AST change detection tool used in DEAR's divide-and-conquer strategy." 444 }, 445 { 446 "title": "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", 447 "relevance": "Pre-trained BERT fine-tuned for fixing-together hunk detection — the first component of DEAR's fault localization pipeline." 448 }, 449 { 450 "title": "TreeCaps: Tree-Based Capsule Networks for Source Code Processing", 451 "relevance": "Used for AST subtree summarization in both context building and multi-statement expansion components." 452 }, 453 { 454 "title": "Harnessing Evolution for Multi-Hunk Program Repair (Hercules)", 455 "relevance": "Top pattern-based APR baseline; DEAR reaches comparable total count while fixing 7 multi-statement bugs Hercules misses entirely." 456 }, 457 { 458 "title": "TBar: Revisiting Template-Based Automated Program Repair", 459 "relevance": "Pattern-based APR baseline; DEAR fixes 15 more multi-hunk/multi-statement bugs than TBar on Defects4J." 460 }, 461 { 462 "title": "GloVe: Global Vectors for Word Representation", 463 "relevance": "Used for code token embedding throughout context building, expansion, and transformation learning in DEAR." 464 }, 465 { 466 "title": "Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks", 467 "relevance": "Source of the cycle training methodology adapted by DEAR to improve code transformation learning in the LSTM layers." 468 } 469 ], 470 "engagement_factors": { 471 "practical_relevance": { 472 "score": 2, 473 "justification": "APR tools directly reduce developer effort for bug fixing, though Java-only support and 20+ hour training time limit immediate adoption." 474 }, 475 "surprise_contrarian": { 476 "score": 1, 477 "justification": "Showing DL-based APR can match pattern-based APR is a meaningful milestone but follows the expected arc of capability improvement in the field." 478 }, 479 "fear_safety": { 480 "score": 0, 481 "justification": "No AI safety or risk implications; this is a software engineering automation tool." 482 }, 483 "drama_conflict": { 484 "score": 0, 485 "justification": "Standard competitive benchmark evaluation with no controversy or community conflict." 486 }, 487 "demo_ability": { 488 "score": 2, 489 "justification": "GitHub repository is publicly available, but 20+ hour training time on specialized hardware limits casual experimentation." 490 }, 491 "brand_recognition": { 492 "score": 0, 493 "justification": "Authors from NJIT and UT Dallas; no association with high-profile AI labs or widely known industry products." 494 } 495 }, 496 "hn_data": { 497 "threads": [ 498 { 499 "hn_id": "36018657", 500 "title": "DarkBERT: A Language Model for the Dark Side of the Internet", 501 "points": 142, 502 "comments": 59, 503 "url": "https://news.ycombinator.com/item?id=36018657" 504 }, 505 { 506 "hn_id": "38162779", 507 "title": "Category Theory for Programming", 508 "points": 47, 509 "comments": 6, 510 "url": "https://news.ycombinator.com/item?id=38162779" 511 }, 512 { 513 "hn_id": "30565951", 514 "title": "Improved Approximation Algorithms and Lower Bounds for Search-Diversification", 515 "points": 5, 516 "comments": 0, 517 "url": "https://news.ycombinator.com/item?id=30565951" 518 }, 519 { 520 "hn_id": "35994539", 521 "title": "DarkBERT: A Language Model for the Dark Side of the Internet", 522 "points": 4, 523 "comments": 0, 524 "url": "https://news.ycombinator.com/item?id=35994539" 525 }, 526 { 527 "hn_id": "36013633", 528 "title": "DarkBERT: A Language Model for the Dark Side of the Internet", 529 "points": 3, 530 "comments": 1, 531 "url": "https://news.ycombinator.com/item?id=36013633" 532 }, 533 { 534 "hn_id": "43311058", 535 "title": "A programmable environment for shape optimization and shapeshifting problems", 536 "points": 2, 537 "comments": 0, 538 "url": "https://news.ycombinator.com/item?id=43311058" 539 }, 540 { 541 "hn_id": "46211392", 542 "title": "A Simple Proof of the Riemann Hypothesis", 543 "points": 1, 544 "comments": 0, 545 "url": "https://news.ycombinator.com/item?id=46211392" 546 }, 547 { 548 "hn_id": "44067109", 549 "title": "The effectiveness of Large Language Models in the mechanical design domain", 550 "points": 1, 551 "comments": 0, 552 "url": "https://news.ycombinator.com/item?id=44067109" 553 }, 554 { 555 "hn_id": "40350177", 556 "title": "GPT-4 passes most of the 297 written Polish Board Certification Examinations", 557 "points": 1, 558 "comments": 0, 559 "url": "https://news.ycombinator.com/item?id=40350177" 560 } 561 ], 562 "top_points": 142, 563 "total_points": 206, 564 "total_comments": 66 565 } 566 }