添加svm模型

Author: libowei1213

online/svm.py

@@ -0,0 +1,63 @@
+# coding=utf-8
+
+# 输入一个string 首先分词，然后读取本地word2vec模型，计算向量,
+# 读取本地的svm模型，做预测
+# 返回预测结果
+import warnings
+
+warnings.filterwarnings("ignore")
+import jieba
+import numpy as np
+from numpy import *
+
+try:
+    import cPickle as pickle
+except:
+    import pickle
+
+
+def kernel(x, y, sigma):
+    x = mat(x)
+    y = mat(y)
+    temp = x - y
+    return math.exp(temp * temp.T / (-2) * sigma * sigma)
+
+
+def label(x):
+    alphs_result = pickle.loads(open("data/svm_model_1", 'rb').read(), encoding='iso-8859-1')
+    x_result = pickle.loads(open("data/svm_model_2", 'rb').read(), encoding='iso-8859-1')
+    y_result = pickle.loads(open("data/svm_model_3", 'rb').read(), encoding='iso-8859-1')
+    b = pickle.loads(open("data/svm_model_4", 'rb').read(), encoding='iso-8859-1')
+
+    num = len(alphs_result)
+    re = 0.0
+    for i in range(num):
+        re += alphs_result[i] * y_result[i] * kernel(x_result[i], x, 1)
+    re += b
+    if (re < 0):
+        return 0
+    else:
+        return 1
+
+
+def feature(word_model, sentence):
+    seg_list = jieba.cut(sentence, cut_all=False)
+    count = 0
+    for j in seg_list:
+        if j not in word_model:
+            continue
+        if count == 0:
+            old = word_model[j]
+            new = np.zeros(shape=old.shape)
+        else:
+            new = word_model[j]
+        old = old + new
+        count += 1
+    if (count != 0):
+        old = old * (1.0 / count)
+        return list(old)
+
+
+def svm_predict(word_model, sentence):
+    x = feature(word_model, sentence)
+    return label(x)

online/templates/index.html

@@ -13,6 +13,10 @@
 
 
 
+
+
+
+
 </style>
 {% endblock %}
 
@@ -21,17 +25,17 @@
 function check(){
 
     $("#alert").hide();
-
     msg=$("#msg").val();
-
     if(msg==""){
         return false;
     }
 
+    radio= $('input[name="option"]:checked ').val();
+
     $.ajax({
         url: "check",
         method: "POST",
-        data: "msg="+msg,
+        data: "msg="+msg+"&method="+radio,
         success: function(data){
             if(data=="1"){
                 $("#alert").text("这条短信是垃圾短信！");
@@ -63,6 +67,10 @@
 
 
 
+
+
+
+
 </script>
 {% endblock %}
 
@@ -72,6 +80,17 @@
     <div class="row">
         <div class="page-header">
             <h2>输入短信文本</h2></div>
+        <div class="input-group">
+            <label>
+                <input type="radio" name="option" id="option1" value="1" checked>
+                朴素贝叶斯
+            </label>
+            <label>
+                <input type="radio" name="option" id="option2" value="2" checked>
+                SVM
+            </label>
+        </div>
+
         <div class="input-group">
             <input id="msg" class="form-control input-lg" type="text" placeholder="输入文本"><span class="input-group-btn">
         <button id="check" onclick="check();" class="btn btn-lg btn-default" type="button">检测</button>

online/view.py

@@ -2,11 +2,14 @@
 from flask_bootstrap import Bootstrap
 import bayes
 import random
+import online.svm
+import gensim
 
 app = Flask(__name__)
 Bootstrap(app)
 
 prior, condprob = bayes.runBayes()
+word_model = gensim.models.Word2Vec.load_word2vec_format('data/word2vec_model', binary=True)
 
 
 @app.route("/")
@@ -18,7 +21,13 @@ def index():
 @app.route("/check", methods=['POST', 'GET'])
 def checkmsg():
     msg = request.form['msg']
-    result = bayes.predict(msg, prior, condprob)
+    method = request.form['method']
+    if msg == "" or method == "":
+        return ""
+    if method == "1":
+        result = bayes.predict(msg, prior, condprob)
+    elif method == "2":
+        result = online.svm.svm_predict(word_model, msg)
     return str(result)