Initial commit - PICO extractor with GLiNER

.gitignore

@@ -0,0 +1,4 @@
+venv/
+__pycache__/
+*.pyc
+.env

README.md

@@ -1,12 +1,32 @@
----
-title: Testing Pico
-emoji: ⚡
-colorFrom: indigo
-colorTo: red
-sdk: gradio
-sdk_version: 5.34.1
-app_file: app.py
-pinned: false
+# GLiNER-BioMed PICO Extractor
+
+This Hugging Face Space extracts PICO elements (Population, Intervention, Comparison, Outcome) from:
+
+- Raw biomedical abstracts
+- `.nbib` reference files
+
+### Model
+
+Powered by `Ihor/gliner-biomed-bi-small-v1.0` — a compact BERT-like NER model trained for biomedical text using synthetic annotations.
+
+### Features
+
+- ✅ Zero-shot extraction using natural language entity descriptions
+- 📄 NBIB parser for PubMed export files
+- ⚡ Lightweight: deploys on CPU-only Spaces
+
+### How to Use
+
+1. **Paste a biomedical abstract** in the textbox → Get labeled PICO entities.
+2. **Upload a `.nbib` file** → Get per-abstract PICO extractions.
+
+### Dependencies
+
+- `gradio`
+- `gliner`
+- `torch`
+- `transformers`
+
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Inspired by the needs of evidence-based medicine and large-scale systematic reviews.

app.py

@@ -0,0 +1,57 @@
+# app.py
+import gradio as gr
+from gliner import GLiNER
+import re
+
+# Load GLiNER-BioMed Small model
+model = GLiNER.from_pretrained("Ihor/gliner-biomed-bi-small-v1.0")
+labels = ["Population", "Intervention", "Comparison", "Outcome"]
+
+def parse_nbib(file):
+    content = file.read().decode("utf-8")
+    entries = re.findall(r'(?=PMID- .+?)(.*?)(?=(?:PMID- |\Z))', content, re.DOTALL)
+    refs = []
+    for entry in entries:
+        title = ""
+        abstract = ""
+        for line in entry.splitlines():
+            if line.startswith("TI  - "):
+                title = line.replace("TI  - ", "").strip()
+            elif line.startswith("AB  - "):
+                abstract += " " + line.replace("AB  - ", "").strip()
+        if title or abstract:
+            refs.append(f"{title}. {abstract.strip()}")
+    return refs
+
+def extract_pico_from_nbib(file):
+    refs = parse_nbib(file)
+    results = []
+    for ref in refs:
+        entities = model.predict_entities(ref, labels, threshold=0.5)
+        results.append({"text": ref, "pico": entities})
+    return results
+
+def extract_from_text(text):
+    entities = model.predict_entities(text, labels, threshold=0.5)
+    return {ent['label']: ent['text'] for ent in entities}
+
+demo = gr.Interface(
+    fn=extract_from_text,
+    inputs=gr.Textbox(lines=10, placeholder="Paste biomedical abstract here..."),
+    outputs="json",
+    title="PICO Extractor (GLiNER-BioMed Small)",
+    description="Extract PICO (Population, Intervention, Comparison, Outcome) elements from biomedical text using the GLiNER-BioMed Small model."
+)
+
+nbib_demo = gr.Interface(
+    fn=extract_pico_from_nbib,
+    inputs=gr.File(file_types=[".nbib"]),
+    outputs="json",
+    title="PICO Extractor from NBIB",
+    description="Upload .nbib files to extract PICO elements using GLiNER-BioMed Small."
+)
+
+tabs = gr.TabbedInterface([demo, nbib_demo], ["From Text", "From NBIB File"])
+
+if __name__ == "__main__":
+    tabs.launch()

requirements.txt

@@ -0,0 +1,4 @@
+gradio>=4.0.0
+gliner==0.2.0
+torch>=1.10
+transformers