Duplicate from alibabagroup/eMCR

.gitattributes

@@ -0,0 +1,61 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text
+data/emcr_benchmark.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,127 @@
+---
+language:
+- zh
+license: apache-2.0
+size_categories:
+- 10K<n<100K
+task_categories:
+- text-retrieval
+- visual-document-retrieval
+tags:
+- e-commerce
+- multi-condition
+- retrieval
+- benchmark
+- chinese
+configs:
+- config_name: default
+  data_files:
+  - split: test
+    path: data/emcr_benchmark.jsonl
+---
+
+# eMCR: A Benchmark for Multi-Condition Product Retrieval in Chinese E-Commerce
+
+This is the official dataset and evaluation code for the paper *"eMCR: A Benchmark for Multi-Condition Product Retrieval in Chinese E-Commerce"*.
+
+## Overview
+
+Product search increasingly involves queries that combine multiple requirements — product attributes, brands, prices, exclusions, and visual descriptions. Existing retrieval benchmarks provide limited support for diagnosing which conditions cause failures and how performance changes when conditions are combined.
+
+**eMCR** is a diagnostic benchmark for multi-condition product retrieval in Chinese e-commerce. Built from a large-scale real-world Taobao product catalog, eMCR adopts a search-log-informed taxonomy of **15 atomic conditions** spanning expression variation, intent type, and constraint signals. Controlled LLM-assisted generation composes these conditions into **48 task types** with one to three condition signals.
+
+### Key Statistics
+
+| Item | Count |
+|------|-------|
+| Queries | 1,399 |
+| Products | 36,722 |
+| Relevance Judgments | 42,022 |
+| Relevance Levels | 4 (0–3) |
+| Atomic Conditions | 15 |
+| Task Types | 48 |
+
+![eMCR Benchmark Construction Pipeline](assets/pipeline.png)
+*Figure 1: Overview of the eMCR benchmark construction pipeline, including product sampling from a real-world e-commerce catalog, query generation with multi-condition constraints, candidate pool construction, and graded relevance annotation.*
+
+## Evaluated Systems
+
+We evaluate **16 systems** across four paradigms:
+- **Sparse retrieval**: BM25
+- **Dense retrieval**: BGE-M3, E5-Mistral, GritLM, NV-Embed-v2, Qwen3-Embedding (0.6B/4B/8B)
+- **Multimodal**: Qwen3-VL (2B/8B), VLM2Vec, MM-Embed
+- **Reranking**: Cross-Encoder, Qwen3-Reranker, Qwen3-VL-Reranker
+- **Agentic**: LLM-driven retrieval agent with tool use
+
+![Cross-system P@1 by Atomic Condition](assets/figure3.png)
+*Figure 3: Cross-system P@1 performance breakdown by atomic condition.*
+
+## Getting Started
+
+### Installation
+
+```bash
+pip install -r requirements.txt
+```
+
+### Data Structure
+
+```
+data/
+├── emcr_benchmark.jsonl    # Benchmark queries and relevance labels
+└── images/                 # Product images
+    ├── part_0/
+    ├── part_1/
+    ├── part_2/
+    └── part_3/
+```
+
+### Run Evaluation
+
+**Dense Retrieval (e.g., BGE-M3):**
+
+```bash
+python -m mcr.cli \
+  --benchmark configs/benchmarks/emcr.yaml \
+  --model configs/models/bge_m3.yaml
+```
+
+**Reranking (e.g., Qwen3-Reranker):**
+
+```bash
+python -m mcr.cli \
+  --benchmark configs/benchmarks/emcr.yaml \
+  --model configs/models/qwen3_reranker_8b.yaml
+```
+
+**Agentic Retrieval:**
+
+```bash
+export DASHSCOPE_API_KEY=your_key_here
+python run_agentic_v2.py \
+  --benchmark configs/benchmarks/emcr.yaml \
+  --model configs/models/agentic_qwen3_7_max.yaml
+```
+
+### Configuration
+
+- Benchmark configs: `configs/benchmarks/`
+- Model configs: `configs/models/`
+
+Each model config specifies the encoder/reranker class and parameters. Set API keys via environment variables where needed.
+
+## Citation
+
+If you find this benchmark useful, please cite:
+
+```bibtex
+@article{lu2025emcr,
+  title={eMCR: A Benchmark for Multi-Condition Product Retrieval in Chinese E-Commerce},
+  author={Lu, Xuan and Li, Qiang and Du, Xiaoxiong and Huang, Haohang and Zheng, Junjun and Kong, Xiangheng and Ou, Dan and Shen, Xiaoyu},
+  year={2025}
+}
+```
+
+## License
+
+This dataset is released under the [Apache 2.0 License](LICENSE).

configs/benchmarks/emcr.yaml

@@ -0,0 +1,11 @@
+name: emcr
+data_root: ./data
+filename: emcr_benchmark.jsonl
+image_dir: ./data/images
+modality: text
+k_values: [1, 5, 10, 20, 50, 100]
+retriever:
+  index_type: flat
+  normalize: true
+# binary relevance threshold: rel = 2^score - 1; score==3 -> rel 7 == positive
+pos_threshold: 7

configs/benchmarks/mcmr.yaml

@@ -0,0 +1,7 @@
+name: mcmr
+# data_root: ~/.cache/mcr/datasets/MCMR  # override if data is elsewhere
+modality: image_text
+k_values: [1, 5, 10, 50, 100]
+retriever:
+  index_type: flat
+  normalize: true

configs/benchmarks/merit.yaml

@@ -0,0 +1,8 @@
+name: merit
+# data_root: ~/.cache/mcr/datasets/merit  # override if data is elsewhere
+modality: interleaved
+split: test
+k_values: [1, 5, 10, 50, 100]
+retriever:
+  index_type: flat
+  normalize: true

configs/benchmarks/multiconir.yaml

@@ -0,0 +1,7 @@
+name: multiconir
+# data_root: ~/.cache/mcr/datasets/MultiConIR_v2  # override if data is elsewhere
+modality: text
+k_values: [1, 5, 10, 20, 50, 100]
+retriever:
+  index_type: flat
+  normalize: true

configs/benchmarks/ssrb.yaml

@@ -0,0 +1,7 @@
+name: ssrb
+# data_root: ~/.cache/mcr/datasets/struct-ir  # override if data is elsewhere
+modality: text
+k_values: [1, 5, 10, 20, 50, 100]
+retriever:
+  index_type: flat
+  normalize: true

configs/benchmarks/ssrb_subset.yaml

@@ -0,0 +1,6 @@
+name: ssrb
+# data_root: ~/.cache/mcr/datasets/struct-ir  # override if data is elsewhere
+modality: text
+schemas_per_domain: 2
+k_values: [1, 5, 10, 20, 50, 100]
+retriever: {index_type: flat, normalize: true}

configs/models/agentic_qwen3.yaml

@@ -0,0 +1,15 @@
+tag: agentic-qwen3-8b
+
+llm:
+  model: Qwen3-8B
+  api_base: http://localhost:8000/v1
+  api_key: EMPTY
+  temperature: 0.0
+  max_tokens: 2048
+
+encoder: qwen3_emb_0.6B
+reranker: cross_encoder
+
+max_rounds: 3
+search_top_k: 20
+rerank_top_k: 30

configs/models/agentic_qwen3_5.yaml

@@ -0,0 +1,17 @@
+tag: agentic-qwen3.5-9b-vllm
+
+llm:
+  backend: api
+  model: Qwen/Qwen3.5-9B
+  api_base: http://localhost:8100/v1
+  api_key: EMPTY
+  temperature: 0.0
+  max_tokens: 8192
+
+encoder: qwen3_vl_2B
+reranker: qwen3_vl_reranker_2b
+
+max_rounds: 3
+search_top_k: 10
+rerank_top_k: 50
+rerank_return_k: 10

configs/models/agentic_qwen3_7_max.yaml

@@ -0,0 +1,17 @@
+tag: agentic-qwen3.7-max
+
+llm:
+  backend: api
+  model: qwen3.7-plus
+  api_base: https://dashscope.aliyuncs.com/compatible-mode/v1
+  api_key: ${DASHSCOPE_API_KEY}
+  temperature: 0.0
+  max_tokens: 8192
+
+encoder: qwen3_vl_8B
+reranker: qwen3_vl_reranker_8b
+
+max_rounds: 3
+search_top_k: 10
+rerank_top_k: 30
+rerank_return_k: 10

configs/models/bge_m3.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: bge-m3
+model_name_or_path: ./models/bge-m3
+query_prompt: ""
+doc_prompt: ""
+normalize: true
+batch_size: 64
+trust_remote_code: false
+multimodal: false

configs/models/bm25.yaml

@@ -0,0 +1,4 @@
+type: bm25
+tag: bm25
+normalize: false
+multimodal: false

configs/models/cross_encoder.yaml

@@ -0,0 +1,6 @@
+type: cross_encoder
+tag: bge-reranker-v2-m3
+model_name_or_path: ./models/bge-reranker-v2-m3
+batch_size: 64
+max_length: 512
+trust_remote_code: false

configs/models/e5_mistral.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: e5-mistral-7b
+model_name_or_path: ./models/E5-mistral-7B
+query_prompt: "query: "
+doc_prompt: "passage: "
+normalize: true
+batch_size: 32
+trust_remote_code: false
+multimodal: false

configs/models/gritlm.yaml

@@ -0,0 +1,7 @@
+type: gritlm
+tag: gritlm-7b
+model_name_or_path: ./models/GritLM-7B
+query_prompt: "Given a web search query, retrieve passages that answer the query."
+device: cuda
+batch_size: 16
+multimodal: false

configs/models/jina_reranker_m0.yaml

@@ -0,0 +1,7 @@
+type: jina_reranker
+tag: jina-reranker-m0
+model_name_or_path: ./models/jina-reranker-m0
+dtype: bfloat16
+max_length: 8192
+batch_size: 4
+device: cuda

configs/models/mm_embed.yaml

@@ -0,0 +1,6 @@
+type: mm_embed
+tag: mm-embed
+model_name_or_path: ./models/MM-Embed
+device: cuda
+batch_size: 4
+multimodal: true

configs/models/nv_embed_v2.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: nv-embed-v2
+model_name_or_path: ./models/NV-Embed-v2
+query_prompt: "query: "
+doc_prompt: "passage: "
+normalize: true
+batch_size: 16
+trust_remote_code: true
+multimodal: false

configs/models/qwen3_5_reranker.yaml

@@ -0,0 +1,6 @@
+type: generative_pointwise
+tag: qwen3.5-9b-reranker
+model_name_or_path: Qwen/Qwen3.5-9B
+dtype: bfloat16
+max_text_chars: 1500
+device: "cuda:1"

configs/models/qwen3_emb_0.6B.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: qwen3-emb-0.6B
+model_name_or_path: Qwen/Qwen3-Embedding-0.6B
+query_prompt: "Instruct: Given a multi-condition retrieval query, retrieve documents that satisfy all the conditions.\nQuery: "
+doc_prompt: ""
+normalize: true
+batch_size: 64
+trust_remote_code: true
+multimodal: false

configs/models/qwen3_emb_4B.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: qwen3-emb-4B
+model_name_or_path: ./models/Qwen3-Embedding-4B
+query_prompt: "Instruct: Given a multi-condition retrieval query, retrieve documents that satisfy all the conditions.\nQuery: "
+doc_prompt: ""
+normalize: true
+batch_size: 16
+trust_remote_code: true
+multimodal: false

configs/models/qwen3_emb_8B.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: qwen3-emb-8B
+model_name_or_path: ./models/Qwen3-Embedding-8B
+query_prompt: "Instruct: Given a multi-condition retrieval query, retrieve documents that satisfy all the conditions.\nQuery: "
+doc_prompt: ""
+normalize: true
+batch_size: 8
+trust_remote_code: true
+multimodal: false

configs/models/qwen3_reranker_4b.yaml

@@ -0,0 +1,8 @@
+type: qwen3_text_reranker
+tag: qwen3-reranker-4b
+model_name_or_path: ./models/Qwen3-Reranker-4B
+dtype: bfloat16
+batch_size: 16
+max_length: 8192
+max_text_chars: 4000
+instruction: "Given a multi-condition product search query, retrieve relevant products that satisfy all conditions in the query."

configs/models/qwen3_reranker_8b.yaml

@@ -0,0 +1,8 @@
+type: qwen3_text_reranker
+tag: qwen3-reranker-8b
+model_name_or_path: ./models/Qwen3-Reranker-8B
+dtype: bfloat16
+batch_size: 16
+max_length: 8192
+max_text_chars: 4000
+instruction: "Given a multi-condition product search query, retrieve relevant products that satisfy all conditions in the query."

configs/models/qwen3_vl_2B.yaml

@@ -0,0 +1,8 @@
+type: qwen3_vl_emb
+tag: qwen3-vl-2B
+model_name_or_path: ./models/Qwen3-VL-Embedding-2B
+query_prompt: "Represent this query to retrieve the item that satisfies all conditions."
+doc_prompt: "Represent the user's input."
+device: cuda
+batch_size: 8
+multimodal: true

configs/models/qwen3_vl_2B_merit.yaml

@@ -0,0 +1,10 @@
+type: st
+tag: qwen3-vl-2B
+model_name_or_path: Qwen/Qwen3-VL-Embedding-2B
+query_prompt: "Represent this query to retrieve the item that satisfies all conditions."
+doc_prompt: "Represent the user's input."
+normalize: true
+batch_size: 8
+trust_remote_code: true
+multimodal: true
+max_pixels: 602112

configs/models/qwen3_vl_8B.yaml

@@ -0,0 +1,8 @@
+type: qwen3_vl_emb
+tag: qwen3-vl-8B
+model_name_or_path: ./models/Qwen3-VL-Embedding-8B
+query_prompt: "Represent this query to retrieve the item that satisfies all conditions."
+doc_prompt: "Represent the user's input."
+device: cuda
+batch_size: 4
+multimodal: true

configs/models/qwen3_vl_8B_merit.yaml

@@ -0,0 +1,10 @@
+type: st
+tag: qwen3-vl-8B
+model_name_or_path: Qwen/Qwen3-VL-Embedding-8B
+query_prompt: "Represent this query to retrieve the item that satisfies all conditions."
+doc_prompt: "Represent the user's input."
+normalize: true
+batch_size: 8
+trust_remote_code: true
+multimodal: true
+max_pixels: 602112

configs/models/qwen3_vl_emb.yaml

@@ -0,0 +1,9 @@
+type: st
+tag: qwen3-vl-embedding
+model_name_or_path: Qwen/Qwen3-VL-Embedding-2B
+query_prompt: ""
+doc_prompt: ""
+normalize: true
+batch_size: 32
+trust_remote_code: true
+multimodal: true

configs/models/qwen3_vl_reranker_2b.yaml

@@ -0,0 +1,7 @@
+type: qwen3_vl_reranker
+tag: qwen3-vl-reranker-2b
+model_name_or_path: ./models/Qwen3-VL-Reranker-2B
+dtype: bfloat16
+max_text_chars: 2000
+device: cuda
+instruction: "Given a multi-condition product search query, retrieve candidates that satisfy all conditions in the query."

configs/models/qwen3_vl_reranker_8b.yaml

@@ -0,0 +1,7 @@
+type: qwen3_vl_reranker
+tag: qwen3-vl-reranker-8b
+model_name_or_path: ./models/Qwen3-VL-Reranker-8B
+dtype: bfloat16
+max_text_chars: 2000
+device: cuda
+instruction: "Given a multi-condition product search query, retrieve candidates that satisfy all conditions in the query."

configs/models/st_text.yaml

@@ -0,0 +1,10 @@
+type: st
+tag: bge-small-en
+model_name_or_path: BAAI/bge-small-en-v1.5
+# instruction prompts (leave empty for models that don't use them)
+query_prompt: "Represent this sentence for searching relevant passages: "
+doc_prompt: ""
+normalize: true
+batch_size: 256
+trust_remote_code: false
+multimodal: false

configs/models/vlm2vec_v2.yaml

@@ -0,0 +1,8 @@
+type: vlm2vec_v2
+tag: vlm2vec-v2
+model_name_or_path: ./models/VLM2Vec-V2.0
+base_model_path: ./models/Qwen2-VL-2B-Instruct
+vlm2vec_src: ./third_party/VLM2Vec
+device: cuda
+batch_size: 4
+multimodal: true

configs/models/vlm_reranker.yaml

@@ -0,0 +1,5 @@
+type: vlm_pointwise
+tag: qwen2.5-vl-7b-reranker
+model_name_or_path: Qwen/Qwen2.5-VL-7B-Instruct
+dtype: bfloat16
+max_text_chars: 1500

data/emcr_benchmark.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c33001d33a5536593bb4d5361437fed0e87d18b9dcbf29cf404d65d74c0fa622
+size 119407487