Update README.md

README.md

@@ -6,125 +6,156 @@ tags:
 - feature-extraction
 - sentence-similarity
 - transformers
+language:
+- en
+base_model:
+- cambridgeltl/SapBERT-from-PubMedBERT-fulltext-mean-token
+---
+
+# NA-SapBERT: Noise-Augmented SapBERT for Clinical Concept Normalization
+
+NA-SapBERT is a dense retrieval model designed for clinical concept normalization over large ontologies such as SNOMED CT. It extends SapBERT by incorporating noise-aware training, enabling robust retrieval for real-world clinical mentions.
+
+Unlike standard SapBERT, this model is trained to handle:
+- abbreviations (e.g., "NAD", "DM")
+- misspellings
+- shorthand and telegraphic clinical text
+- surface form variation across notes
 
 ---
 
-# {MODEL_NAME}
+## Overview
 
-This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
+Clinical concept normalization maps noisy text mentions to standardized ontology concepts. While modern NER systems perform well, entity linking remains challenging due to:
 
-<!--- Describe your model here -->
+- large ontology size
+- noisy clinical text
+- ambiguous abbreviations
+- mismatch between ontology terms and real-world mentions
 
-## Usage (Sentence-Transformers)
+NA-SapBERT addresses this by learning invariant embeddings across noisy and canonical forms.
 
-Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:
+---
 
-```
-pip install -U sentence-transformers
-```
+## Key Idea
 
-Then you can use the model like this:
+During training, the model learns to align:
+- noisy mentions (LLM-generated variants, abbreviations)
+- clean ontology terms (concept names and synonyms)
 
-```python
-from sentence_transformers import SentenceTransformer
-sentences = ["This is an example sentence", "Each sentence is converted"]
+This is achieved using contrastive learning:
+- clean–clean pairs preserve structure
+- noisy–clean pairs improve robustness
 
-model = SentenceTransformer('{MODEL_NAME}')
-embeddings = model.encode(sentences)
-print(embeddings)
-```
+---
+
+## Model Architecture
+
+SentenceTransformer:
+- Transformer (PubMedBERT backbone)
+- Mean Pooling
+
+Embedding dimension: 768  
+Max sequence length: 64
 
+---
+
+## Training Details
+
+### Data
+- SNOMED CT concepts (subset of key semantic types)
+- Synthetic variants:
+  - LLM-generated (MedGemma) noise
+  - abbreviation mappings
 
+### Objective
+MultipleNegativesRankingLoss (InfoNCE-style)
 
-## Usage (HuggingFace Transformers)
-Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
+### Training Configuration
+- epochs: 1
+- batch_size: 256
+- learning_rate: 1e-5
+- warmup_steps: 85
+
+---
 
+## Usage
+
+### Install
+pip install -U sentence-transformers
+
+### Encode Mentions
 ```python
-from transformers import AutoTokenizer, AutoModel
-import torch
+from sentence_transformers import SentenceTransformer
+
+model = SentenceTransformer("YOUR_MODEL_NAME")
 
+mentions = ["NAD", "hx of diabetes", "left axillary lymph node"]
+embeddings = model.encode(mentions, normalize_embeddings=True)
+```
 
-#Mean Pooling - Take attention mask into account for correct averaging
-def mean_pooling(model_output, attention_mask):
-    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
-    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
-    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
+---
 
+## Retrieval Example (FAISS)
 
-# Sentences we want sentence embeddings for
-sentences = ['This is an example sentence', 'Each sentence is converted']
+```python
+import faiss
+import numpy as np
+from sentence_transformers import SentenceTransformer
 
-# Load model from HuggingFace Hub
-tokenizer = AutoTokenizer.from_pretrained('{MODEL_NAME}')
-model = AutoModel.from_pretrained('{MODEL_NAME}')
+model = SentenceTransformer("YOUR_MODEL_NAME")
 
-# Tokenize sentences
-encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
+concept_embeddings = np.load("concept_embeddings.npy").astype("float32")
 
-# Compute token embeddings
-with torch.no_grad():
-    model_output = model(**encoded_input)
+index = faiss.IndexFlatIP(768)
+index.add(concept_embeddings)
 
-# Perform pooling. In this case, mean pooling.
-sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
+query = "NAD"
+q_emb = model.encode([query], normalize_embeddings=True)
 
-print("Sentence embeddings:")
-print(sentence_embeddings)
+scores, indices = index.search(q_emb, k=10)
 ```
 
+---
 
+## Pipeline Integration
 
-## Evaluation Results
+Typical pipeline:
+1. Exact match
+2. Dense retrieval (NA-SapBERT)
+3. Optional rewrite / multi-query
+4. Optional reranking
 
-<!--- Describe how your model was evaluated -->
+---
 
-For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: [https://seb.sbert.net](https://seb.sbert.net?model_name={MODEL_NAME})
+## Performance Summary
 
+- SapBERT: XX recall@1
+- NA-SapBERT: XX recall@1
 
-## Training
-The model was trained with the parameters:
+Improvements:
+- Better handling of noisy mentions
+- Strong generalization to full SNOMED CT
 
-**DataLoader**:
+---
 
-`torch.utils.data.dataloader.DataLoader` of length 853 with parameters:
-```
-{'batch_size': 256, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
-```
+## Limitations
 
-**Loss**:
+- No explicit modeling of negation or temporality
+- Abbreviations remain ambiguous without context
+- Depends on ontology synonym quality
 
-`sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss` with parameters:
-  ```
-  {'scale': 20.0, 'similarity_fct': 'cos_sim'}
-  ```
+---
 
-Parameters of the fit()-Method:
-```
-{
-    "epochs": 1,
-    "evaluation_steps": 0,
-    "evaluator": "NoneType",
-    "max_grad_norm": 1,
-    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
-    "optimizer_params": {
-        "lr": 1e-05
-    },
-    "scheduler": "WarmupLinear",
-    "steps_per_epoch": null,
-    "warmup_steps": 85,
-    "weight_decay": 0.01
-}
-```
+## Use Cases
 
+Use for:
+- clinical NLP
+- concept normalization
+- ontology retrieval
 
-## Full Model Architecture
-```
-SentenceTransformer(
-  (0): Transformer({'max_seq_length': 64, 'do_lower_case': False}) with Transformer model: BertModel 
-  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
-)
-```
+Not intended for:
+- general semantic similarity
+- non-biomedical tasks
 
-## Citing & Authors
 
-<!--- Describe where people can find more information -->