上文提到了贝叶斯算法的几种开源实现,本文说说如何将其中一种名为b8的开源实现集成进CakePHP。
下载b8及安装
-
去b8的站点下载最新版本,将其解压至vendors目录,文件位置如vendors/b8/b8.php;
-
用文本编辑器打开vendors/b8/etc/config_b8,修改databaseType为mysql;
-
用文本编辑器打开vendors/b8/etc/config_storage,修改tableName为你用来存储关键字的数据表的名字,修改createDB为TRUE,要注意的是,当你第一次运行b8后,它会建立上述数据表,然后你要重新把createDB改为FALSE;
-
用文本编辑器打开vendors/b8/lexer/shared_functions.php,将38行的代码(在echoError())注释掉,否则b8会直接把错误信息显示在你的Cake应用中,当然这在调试程序时还是有用的。
为b8写一个wrapper component
为了让你的Cake能够调用到b8,你需要写一个component。在controllers/components/新建一个spam_shield.php,加入如下代码:
class SpamShieldComponent extends Object {
/** * b8 instance */
var $b8;
/** * standard rating * * comments with ratings which are higher than this one will be considered as SPAM */
var $standardRating = 0.7;
/** * text to be classified*/
var $text;
/** * rating of the text */
var $rating;
/** * Constructor * * @date 2009-1-20 */
function startup(&$controller) {
//register a CommentModel to get the DBO resource link
$comment = ClassRegistry::init('Comment'); //import b8 and create an instance
App::import('Vendor', 'b8/b8');
$this->b8 = new b8($comment->getDBOResourceLink()); //set standard rating
$this->standardRating = Configure::read('LT.bayesRating') ? Configure::read('LT.bayesRating') : $this->standardRating;
}
/** * Set the text to be classified * * @param $text String the text to be classified * @date 2009-1-20 */
function set($text) {
$this->text = $text;
}
/** * Get Bayesian rating * * @date 2009-1-20 */
function rate() {
//get Bayes rating and return return
$this->rating = $this->b8->classify($this->text);
}
/** * Validate a message based on the rating, return true if it's NOT a SPAM * * @date 2009-1-20 */
function validate() {
return $this->rate() <$this->standardRating;
}
/** * Learn a SPAM or a HAM * * @date 2009-1-20 */
function learn($mode) {
$this->b8->learn($this->text, $mode);
}
/** * Unlearn a SPAM or a HAM * * @date 2009-1-20 */
function unlearn($mode) {
$this->b8->unlearn($this->text, $mode);
}
}
几点说明:
-
$standardRating是一个临界点。如果贝叶斯概率高于这个值,则此留言被认为是spam,否则是ham。我设置为0.7,你可以根据自己的情况修改;
-
Configure::read('LT.bayesRating')是从系统运行配置中动态地获取上述临界点的值,这是我的做法,你可能用不到,根据情况稍微修改甚至不修改都行;
-
Comment指的是评论的model;
-
由于b8需要获得数据库句柄以便能够操作数据表,所以在startup()中我写了$this->b8 = new b8($comment->getDBOResourceLink())一句,其中用到的getDBOResourceLink()马上会提及。
为b8传入数据库句柄
在models/comment.php中加入如下代码:
/** * get the resource link of MySQL connection */ public function getDBOResourceLink() { return $this->getDataSource()->connection; }
至此,准备工作全部做完,我们终于可以使用贝叶斯算法来分类留言。
使用b8分类留言
在controllers/comments_controller.php中,首先载入SpamShieldComponent:
var $compOnents= array('SpamShield');
然后在add()方法中,做如下操作:
//set data for Bayesian validation
$this->SpamShield->set($this->data['Comment']['body']); //validate the comment with Bayesian
if(!$this->SpamShield->validate()) { //set the status
$this->data['Comment']['status'] = 'spam'; //save
$this->Comment->save($this->data); //learn it $this->SpamShield->learn("spam"); //render
$this->renderView('unmoderated');
return;
}
//it's a normal post
$this->data['Comment']['status'] = 'published'; //save for publish
$this->Comment->save($this->data); //learn it
$this->SpamShield->learn("ham");
如此一来,b8就会在留言到来时自动的分类并学习,你基本上已经与spam绝缘了!
提醒一下:第一次运行后,别忘了把刚才提到的createDB改为FALSE。
http://dingyu.me/blog/spam-bayesian-chinese-4