Automatic clustering of Arabic documents

العناوين الأخرى

التنجميع التلقائي للوثائق العربية

مقدم أطروحة جامعية

al-Saadi, Sami Abd al-Karim

مشرف أطروحة جامعية

Awajan, Arafat

أعضاء اللجنة

Atum, Jalal Yusuf
Abu al-Suud, Salih Mustafa
Abu Shawir, Bayan

الجامعة

جامعة الأميرة سمية للتكنولوجيا

الكلية

كلية الملك الحسين لعلوم الحوسبة

القسم الأكاديمي

قسم علم الحاسوب

دولة الجامعة

الأردن

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2012

الملخص الإنجليزي

Arabic Documents Clustering is an important task for obtaining good results in traditional Information Retrieval systems especially with the rapid growth of the number of online documents existed in Arabic language.

Document clustering aims to automatically group similar documents into clusters so that each document in a set are similar to other documents in the same set and dissimilar to documents in other sets.

In this study we will discuss Natural languages processing and their applications then we will study current clustering algorithms and clustering of Arabic documents and their problems which are : stemming vs.

rooting, synonyms and homonyms, antonyms when removing stop words, words order and weighting, strange words and names and sentence level meaning.

Then we proposed solutions and new techniques to improve the clustering quality and solve current problems, the proposed solutions are : using a combination of all stems and all roots to represent the document on all level of analysis, implement a dictionary to find all the synonyms and possible meanings for all the words in the documents set and to find the antonyms of the word if it is preceded by negation, using N-Grams (unigram, bigram and trigram) instead of tokenization, and to use weights for words based on term frequency inverse document frequency instead of counting the frequency of words only, the improvements applied on a non-hierarchal approach.

The study concludes that these problems can be solved totally or partially by using the suggested solutions and that the clustering quality can be improved by up to 55 % than the traditional clustering (which is based on five stages preprocessing only) when using the suggested improvements and techniques.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

الموضوعات

• الخطوط العربية
• معالجة البيانات
• المخطوطات

عدد الصفحات

129

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-305146