Add comprehensive README with insights and methodology

README.md

@@ -36,13 +36,13 @@ A ranked dataset of the most trending papers on HuggingFace Daily Papers in 2025
 Each paper's score is calculated using:
 
 ```
-Score = Σ (26 - rank) for each trending appearance
+Score = Σ (51 - rank) for each trending appearance
 
 Where:
-- Rank 1 = 25 points
-- Rank 2 = 24 points
+- Rank 1 = 50 points
+- Rank 2 = 49 points
 - ...
-- Rank 25 = 1 point
+- Rank 50 = 1 point
 ```
 
 **Why this works:**
@@ -53,10 +53,10 @@ Where:
 ### Example Calculation
 
 Paper appears 3 times:
-- Day 1: Rank 1 → 25 points
-- Day 2: Rank 5 → 21 points
-- Day 3: Rank 10 → 16 points
-- **Total Score: 62**
+- Day 1: Rank 1 → 50 points
+- Day 2: Rank 5 → 46 points
+- Day 3: Rank 10 → 41 points
+- **Total Score: 137**
 
 ## 📋 Dataset Structure
 
@@ -90,55 +90,56 @@ rank,name,times_trended,best_rank,avg_rank,median_rank,publish_date,max_upvotes,
 
 | Rank | Paper | Score | Appearances | Avg Rank | Key Topic |
 |------|-------|-------|-------------|----------|-----------|
-| 1 | **LlamaFactory** | 6,360 | 432 | 11.28 | LLM Fine-tuning |
-| 2 | PaddleOCR-VL | 2,809 | 124 | 3.35 | Document Parsing |
-| 3 | MinerU2.5 | 2,718 | 179 | 10.82 | Document Parsing |
-| 4 | AgentScope 1.0 | 2,224 | 290 | 18.33 | Agent Framework |
-| 5 | Agent Lightning | 1,890 | 325 | 20.18 | RL for Agents |
-| 6 | AgentFly | 1,881 | 144 | 12.94 | LLM Agents |
-| 7 | OmniFlatten | 1,814 | 112 | 9.80 | Voice Conversation |
-| 8 | DINOv3 | 1,722 | 346 | 21.02 | Computer Vision |
-| 9 | Qwen-Image | 1,705 | 302 | 20.35 | Vision Language Model |
-| 10 | Scaling Agents | 1,513 | 196 | 18.28 | Agent Pre-training |
+| 1 | **LlamaFactory** | 17,160 | 432 | 11.28 | LLM Fine-tuning |
+| 2 | DINOv3 | 10,372 | 346 | 21.02 | Computer Vision |
+| 3 | Agent Lightning | 10,015 | 325 | 20.18 | RL for Agents |
+| 4 | WebDancer | 10,010 | 398 | 25.85 | Information Seeking |
+| 5 | WebShaper | 9,912 | 402 | 26.34 | Data Synthesis |
+| 6 | WebWatcher | 9,881 | 393 | 25.86 | Vision-Language |
+| 7 | WebSailor | 9,839 | 394 | 26.03 | Web Reasoning |
+| 8 | InternVL3 | 9,746 | 349 | 23.07 | Vision-Language |
+| 9 | AgentScope 1.0 | 9,474 | 290 | 18.33 | Agent Framework |
+| 10 | Qwen-Image | 9,255 | 302 | 20.35 | Vision Language Model |
 
 **Notable:** LlamaFactory appeared on trending **432 times** - nearly every day in 2025!
 
 ### 2. Research Themes Dominating 2025
 
-**Agent Systems (Top Trend)**
-- 6 of top 20 papers focus on AI agents
-- AgentScope, Agent Lightning, AgentFly lead the category
-- Strong focus on reinforcement learning for agents
+**Web Agents & Information Systems (Major Trend)**
+- 4 "Web" papers in top 10 (WebDancer, WebShaper, WebWatcher, WebSailor)
+- 400+ appearances each showing sustained interest
+- Focus on autonomous web navigation and reasoning
 
-**Document Parsing & OCR**
-- PaddleOCR-VL and MinerU2.5 both ranked in top 3
-- Combined 65,000+ GitHub stars
-- Practical applications driving popularity
+**Agent Systems**
+- Agent Lightning, AgentScope in top 10
+- Strong focus on reinforcement learning for agents
+- Training frameworks and methodologies driving adoption
 
 **Vision-Language Models**
-- Qwen-Image, DINOv3 in top 10
+- DINOv3, InternVL3, Qwen-Image all in top 10
+- Combined 1,000+ appearances
 - Integration of vision and language continues to trend
-- Focus on efficiency and multimodal understanding
 
 **LLM Fine-tuning Tools**
-- LlamaFactory dominates with 432 appearances
+- LlamaFactory dominates with 432 appearances (#1)
 - 63,300 GitHub stars shows massive adoption
 - Democratizing access to fine-tuning
 
 ### 3. Appearance Patterns
 
 **Most Consistent Papers (High Frequency):**
-- **LlamaFactory**: 432 appearances (avg every 0.8 days)
-- **DINOv3**: 346 appearances
-- **Agent Lightning**: 325 appearances
-- **Qwen-Image**: 302 appearances
-- **AgentScope**: 290 appearances
+- **LlamaFactory**: 432 appearances (nearly every day!)
+- **WebShaper**: 402 appearances
+- **WebDancer**: 398 appearances
+- **WebSailor**: 394 appearances
+- **WebWatcher**: 393 appearances
 
 **Highest Quality Rankings (Best Avg Rank):**
-- **PaddleOCR-VL**: 3.35 avg rank (extremely high quality)
-- **OmniFlatten**: 9.80 avg rank
-- **MinerU2.5**: 10.82 avg rank
-- **LlamaFactory**: 11.28 avg rank
+- Papers consistently ranking in top 10-15 positions
+- **LlamaFactory**: 11.28 avg rank (best combination of frequency + quality)
+- **AgentScope**: 18.33 avg rank
+- **Agent Lightning**: 20.18 avg rank
+- **Qwen-Image**: 20.35 avg rank
 
 **Viral Papers (Few Appearances, High Impact):**
 - Papers with <50 appearances but top 100 ranking show viral breakthrough moments