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

README.md

@@ -0,0 +1,4 @@
+# Ridge-Prohori (Bangla Smishing Classifier)
+
+**RidgeClassifier (L2-regularized linear model)** on TF-IDF **word(1–2)** + **char_wb(3–5)**.
+Preprocessing: cue-before-URL → `<LINK_CUE>`, URLs → `<URL>`, lowercase.

id2label.json

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

label2id.json

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

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>=1.2
+scipy
+numpy
+joblib

ridge_model.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff91406f0141931d3b0910dab95e530584b5261ff3cc54202b606084a60fe66f
+size 1135656

tfidf_char.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aca3bc916fcd9cc4b93cb85f7900bc59434805664a4f9afff89b71ec22cf375
+size 902539

tfidf_word.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b0abae11775f89f228755d7cc1772b920a099950fbb6cee225aad68b890948
+size 144671