Arabic SMS spam detection based on semantic classification

العناوين الأخرى

كشف الرسائل العربية القصيرة المزعجة اعتمادا على التصنيف الدلالي

مقدم أطروحة جامعية

Abu Awdah, Ayman Muhammad Najib Ismail

مشرف أطروحة جامعية

Barakah, Ribhi Sulayman

الجامعة

الجامعة الإسلامية

الكلية

كلية تكنولوجيا المعلومات

القسم الأكاديمي

تكنولوجيا المعلومات

دولة الجامعة

فلسطين (قطاع غزة)

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2017

الملخص الإنجليزي

Short Messaging Service (SMS) Spam is unwanted messages sent over the web or mobile system to mobile phone devices.

SMS is attractive for spammers due to its cheap services, easily deciding the destination country and its higher response rates than email.

Existing solutions to this issue are no longer adequate as they are either costly in terms of resources, inefficient, most of the existing detection techniques for SMS spam have been adapted from other contexts such as email spam detection methods.

Spammers are constantly developing more sophisticated tactics causing previous methods for spam detection as ineffective.

Additionally, when it comes to Arabic SMS messages, most SMS spam filtering system based on English language.

This research presents an Arabic SMS spam detection and classification approach using ontology with semantic rules.

An Arabic SMS spam ontology with a support of Arabic WordNet is built by defining spam classes and hierarchy and adding a collection of various spam messages as instances creating a knowledge base reflecting the domain.

To enable the detection and classification of messages based on the knowledge base, a set of SWRL rules were written.

These rules are used by the reasoner to filter out messages as spam or legitimate.

Based on the enriched knowledge base, an SMS spam detection system is built.

It consists of several modules such as query module, reasoning module, synonym module, SMS module and finally classifier module.

The approach is evaluated based on its ability to classify and detect SMS messages as spam or legitimate.

A number of performance measures are used for this purpose.

The evaluation resulted in an accuracy of 96.5% and in a f-measure of 90.5% which are better than those achieved using a traditional classifier such as Naïve Bayes.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

عدد الصفحات

92

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-905275