updates

ood_detector.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bcf2a6c78f52242f694de098c3a0619e5cf1f0d508c562db471dba2b66c8ce4
+size 698020

src/model/config.json

@@ -11,21 +11,31 @@
   "eos_token_id": null,
   "hidden_dim": 3072,
   "id2label": {
-    "0": "cmp-lg",
-    "1": "cs",
-    "2": "math",
-    "3": "physics",
-    "4": "q-bio",
-    "5": "stat"
+    "0": "math.AC",
+    "1": "cs.CV",
+    "2": "cs.AI",
+    "3": "cs.SY",
+    "4": "math.GR",
+    "5": "cs.CE",
+    "6": "cs.PL",
+    "7": "cs.IT",
+    "8": "cs.DS",
+    "9": "cs.NE",
+    "10": "math.ST"
   },
   "initializer_range": 0.02,
   "label2id": {
-    "cmp-lg": 0,
-    "cs": 1,
-    "math": 2,
-    "physics": 3,
-    "q-bio": 4,
-    "stat": 5
+    "cs.AI": 2,
+    "cs.CE": 5,
+    "cs.CV": 1,
+    "cs.DS": 8,
+    "cs.IT": 7,
+    "cs.NE": 9,
+    "cs.PL": 6,
+    "cs.SY": 3,
+    "math.AC": 0,
+    "math.GR": 4,
+    "math.ST": 10
   },
   "max_position_embeddings": 512,
   "model_type": "distilbert",

src/model/model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8af50efddce3c3be23f1a10a8e36a552bc103eee2e23f3f72932c28a2e408447
-size 267844872
+oid sha256:c04a94c7251860b13606a9f7f83c3fdac1ae53cbcc176d30850d475a470a83ae
+size 267860252

src/model/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a32c09b0f520aeb11a0e8a1125469cb8d0108f232ec0afb0098b0992077044b5
+oid sha256:32d46654a39fad3d312716bedc11d84a28e1cfd41badf3703ed0fc11c176fa7b
 size 5137

src/streamlit_app.py

@@ -3,6 +3,7 @@ import os
 import time
 
 import arxiv
+import joblib
 import streamlit as st
 from transformers import pipeline
 
@@ -20,7 +21,14 @@ def load_pipeline():
     return pipeline("text-classification", model=model_path, top_k=None)
 
 
+@st.cache_resource
+def load_gatekeeper():
+    base_dir = os.path.dirname(os.path.abspath(__file__))
+    return joblib.load(os.path.join(base_dir, "ood_detector.pkl"))
+
+
 classifier = load_pipeline()
+gatekeeper = load_gatekeeper()
 
 with st.sidebar:
     st.subheader("About the Model")
@@ -56,7 +64,11 @@ with st.expander("Load from link", expanded=True):
 st.text_input("Title", key="auto_title")
 st.text_area("Abstract", height=200, key="auto_abstract")
 
-if st.button("Classify"):
+col_btn, col_bypass = st.columns([3, 1])
+classify_clicked = col_btn.button("Classify", use_container_width=True)
+bypass_gatekeeper = col_bypass.toggle("⚡ Bypass Gatekeeper")
+
+if classify_clicked:
     title = st.session_state["auto_title"]
     abstract = st.session_state["auto_abstract"]
 
@@ -66,6 +78,14 @@ if st.button("Classify"):
 
     text = f"{title.strip()}. {abstract.strip()}" if title.strip() else abstract.strip()
 
+    if not bypass_gatekeeper:
+        is_science = gatekeeper.predict([text])[0]
+        if is_science == 0:
+            st.warning(
+                "This text is NOT a scientific paper. Please enter a valid scientific abstract."
+            )
+            st.stop()
+
     with st.spinner("Classifying paper"):
         start_time = time.time()
         predictions = classifier(text)[0]