(item_id: str, caption: str)
| 35 | |
| 36 | |
| 37 | def classify_caption_kind(item_id: str, caption: str) -> tuple[str, str, str]: |
| 38 | text = f"{item_id} {caption}".lower() |
| 39 | if any( |
| 40 | token in text |
| 41 | for token in [ |
| 42 | "accuracy", |
| 43 | "score", |
| 44 | "performance", |
| 45 | "comparison", |
| 46 | "win-rate", |
| 47 | "results", |
| 48 | "recall", |
| 49 | "latency", |
| 50 | "throughput", |
| 51 | "request rate", |
| 52 | "batched request", |
| 53 | "batch size", |
| 54 | "memory saving", |
| 55 | "ablation", |
| 56 | "ablated", |
| 57 | "overhead", |
| 58 | "microbenchmark", |
| 59 | "block size", |
| 60 | "single sequence generation", |
| 61 | "parallel generation", |
| 62 | ] |
| 63 | ) or re.search( |
| 64 | r"\b(?:model|system|architecture|method|approach)\s+" |
| 65 | r"(?:produces|achieves|outperforms|improves|reduces|increases)\b", |
| 66 | text, |
| 67 | ): |
| 68 | return "main_result", "关键结果", "这张图或表直接承载主结果，适合放在关键结果部分。" |
| 69 | if any( |
| 70 | token in text |
| 71 | for token in [ |
| 72 | "prisma", |
| 73 | "literature flow", |
| 74 | "flow diagram", |
| 75 | "study selection", |
| 76 | "screening flow", |
| 77 | "screened records", |
| 78 | "records screened", |
| 79 | "identification of studies", |
| 80 | ] |
| 81 | ): |
| 82 | return "data_or_task_overview", "数据与任务定义", "这张图解释文献筛选或纳入流程；如果候选图质量足够，适合放在数据与任务定义部分帮助读者理解综述证据来源。" |
| 83 | if any( |
| 84 | token in text |
| 85 | for token in [ |
| 86 | "pipeline", |
| 87 | "framework", |
| 88 | "overview", |
| 89 | "architecture", |
| 90 | "system", |
| 91 | "workflow", |
| 92 | "stage", |
| 93 | "procedure", |
| 94 | "process", |
no outgoing calls
no test coverage detected