# 按年标题热词（第二轮补充）

- 生成：2026-10-19 13:03 Asia/Shanghai
- 方法：论文 title 英文词干粗分词，去停用词；仅作趋势嗅探，非严格 NLP。

## 2025 Top40

| rank | term | count |
|---:|---|---:|
| 1 | generation | 119 |
| 2 | knowledge | 70 |
| 3 | reasoning | 69 |
| 4 | vision-language | 62 |
| 5 | detection | 58 |
| 6 | question | 55 |
| 7 | answering | 52 |
| 8 | evaluation | 50 |
| 9 | biomedical | 50 |
| 10 | benchmark | 47 |
| 11 | report | 46 |
| 12 | alignment | 44 |
| 13 | task | 43 |
| 14 | enhancing | 42 |
| 15 | summarization | 41 |
| 16 | extraction | 39 |
| 17 | through | 39 |
| 18 | evaluating | 38 |
| 19 | text | 37 |
| 20 | diagnosis | 35 |
| 21 | radiology | 34 |
| 22 | foundation | 34 |
| 23 | retrieval | 33 |
| 24 | retrieval-augmented | 33 |
| 25 | mental | 31 |
| 26 | domain | 30 |
| 27 | visual | 28 |
| 28 | prediction | 27 |
| 29 | disease | 27 |
| 30 | entity | 27 |
| 31 | image | 25 |
| 32 | vision | 24 |
| 33 | multi-agent | 23 |
| 34 | protein | 23 |
| 35 | understanding | 22 |
| 36 | shared | 22 |
| 37 | agents | 22 |
| 38 | graph | 22 |
| 39 | agent | 21 |
| 40 | prompt | 21 |

## 2026 Top40

| rank | term | count |
|---:|---|---:|
| 1 | reasoning | 153 |
| 2 | generation | 129 |
| 3 | biomedical | 78 |
| 4 | detection | 75 |
| 5 | benchmark | 72 |
| 6 | vision-language | 70 |
| 7 | report | 68 |
| 8 | evaluation | 61 |
| 9 | question | 57 |
| 10 | extraction | 56 |
| 11 | answering | 53 |
| 12 | knowledge | 52 |
| 13 | retrieval | 48 |
| 14 | diagnosis | 46 |
| 15 | evaluating | 46 |
| 16 | alignment | 46 |
| 17 | multi-agent | 45 |
| 18 | mental | 45 |
| 19 | benchmarking | 42 |
| 20 | radiology | 41 |
| 21 | text | 40 |
| 22 | agentic | 39 |
| 23 | classification | 39 |
| 24 | prediction | 37 |
| 25 | smm4h-heard | 37 |
| 26 | understanding | 36 |
| 27 | visual | 35 |
| 28 | image | 34 |
| 29 | structured | 34 |
| 30 | evidence | 33 |
| 31 | segmentation | 32 |
| 32 | modeling | 32 |
| 33 | agents | 31 |
| 34 | multilingual | 29 |
| 35 | reinforcement | 28 |
| 36 | retrieval-augmented | 28 |
| 37 | patient | 27 |
| 38 | dynamic | 27 |
| 39 | beyond | 26 |
| 40 | through | 26 |

## 相对上升（按年归一化频率差，合计≥8）

| term | 2025 | 2026 | Δrate |
|---|---:|---:|---:|
| reasoning | 69 | 153 | +0.0513 |
| smm4h-heard | 0 | 37 | +0.0270 |
| agentic | 8 | 39 | +0.0215 |
| evidence | 7 | 33 | +0.0180 |
| segmentation | 10 | 32 | +0.0146 |
| classification | 16 | 39 | +0.0145 |
| pipeline | 3 | 22 | +0.0134 |
| biomedical | 50 | 78 | +0.0132 |
| benchmarking | 20 | 42 | +0.0132 |
| multi-agent | 23 | 45 | +0.0127 |
| reinforcement | 9 | 28 | +0.0126 |
| diagnostic | 5 | 22 | +0.0117 |
| psydefdetect | 0 | 16 | +0.0117 |
| benchmark | 47 | 72 | +0.0114 |
| nlp | 4 | 20 | +0.0111 |
| dynamic | 10 | 27 | +0.0110 |
| structured | 16 | 34 | +0.0108 |
| modeling | 15 | 32 | +0.0102 |
| efficient | 10 | 26 | +0.0102 |
| defense | 0 | 14 | +0.0102 |
| mediqa-synur | 0 | 13 | +0.0095 |
| crf | 0 | 13 | +0.0095 |
| report | 46 | 68 | +0.0094 |
| notes | 8 | 22 | +0.0091 |
| ultrasound | 3 | 16 | +0.0091 |

## 相对下降

| term | 2025 | 2026 | Δrate |
|---|---:|---:|---:|
| summarization | 41 | 17 | -0.0235 |
| knowledge | 70 | 52 | -0.0233 |
| task | 43 | 23 | -0.0209 |
| enhancing | 42 | 22 | -0.0207 |
| domain | 30 | 15 | -0.0153 |
| through | 39 | 26 | -0.0152 |
| leveraging | 19 | 5 | -0.0130 |
| chinese | 21 | 8 | -0.0125 |
| foundation | 34 | 25 | -0.0115 |
| peranssumm | 13 | 0 | -0.0114 |
| protein | 23 | 12 | -0.0114 |
| survey | 14 | 2 | -0.0108 |
| representation | 20 | 10 | -0.0102 |
| generation | 119 | 129 | -0.0100 |
| semeval-2025 | 11 | 0 | -0.0096 |
| rag | 18 | 9 | -0.0092 |
| lay | 12 | 2 | -0.0090 |
| retrieval-augmented | 33 | 28 | -0.0084 |
| shared | 22 | 15 | -0.0083 |
| human | 16 | 8 | -0.0082 |
| generative | 21 | 14 | -0.0082 |
| your | 10 | 1 | -0.0080 |
| vision | 24 | 18 | -0.0079 |
| texts | 13 | 5 | -0.0077 |
| entity | 27 | 22 | -0.0076 |