Initial upload of MNB-Prohori (Bangla SMS smishing classifier)

README.md

@@ -0,0 +1,9 @@
+# MNB-Prohori (Bangla SMS smishing classifier)
+
+- **Model**: Multinomial Naive Bayes on TF-IDF (word 1–2g, char 3–5g)
+- **Labels**: ['normal', 'promo', 'smish']
+- **Files**: `multinomial_nb_model.joblib`, `tfidf_word.joblib`, `tfidf_char.joblib`,
+  `preprocess.py`, `id2label.json`, `label2id.json`
+
+## Quick Inference (Python)
+See code in this repo description.

id2label.json

@@ -0,0 +1,5 @@
+{
+  "0": "normal",
+  "1": "promo",
+  "2": "smish"
+}

label2id.json

@@ -0,0 +1,5 @@
+{
+  "normal": 0,
+  "promo": 1,
+  "smish": 2
+}

multinomial_nb_model.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:720f511b908ef6ddbef84d0fe8c20543b4ba05e71211d5042eec76319f30b6be
+size 1733503

preprocess.py

@@ -0,0 +1,11 @@
+
+import re
+cue_words_en = r"(check|click|visit|tap|verify|open|login|log\s*in|see|confirm|update|activate)"
+cue_words_bn = r"(চেক|ক্লিক|ভিজিট|ট্যাপ|যাচাই|লগইন|লগ\s*ইন|দেখুন|আপডেট|অ্যাকটিভেট|নিশ্চিত)"
+url_pat      = re.compile(r"(https?://\S+|www\.\S+|\b[A-Za-z0-9.-]+\.[A-Za-z]{2,}\S*)", re.IGNORECASE)
+cue_before_url_pat = re.compile(rf"(\b{cue_words_en}\b|\b{cue_words_bn}\b)\s*(?={url_pat.pattern})", re.IGNORECASE)
+
+def normalize_text(t: str) -> str:
+    s = re.sub(cue_before_url_pat, "<LINK_CUE> ", str(t))
+    s = re.sub(url_pat, "<URL>", s)
+    return s.lower().strip()

requirements.txt

@@ -0,0 +1,4 @@
+scikit-learn
+scipy
+numpy
+joblib

tfidf_char.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8411d5dbcd8531122704823430ddc8afa8041653dc74505a9cbf7f5f8822114
+size 1133805

tfidf_word.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:109c980c36cb5bb1c55f8d4ea1a9fd859fa850b1554566fd9b806e7e859098c9
+size 242572