The semantic similarity measures using Arabic ontology

العناوين الأخرى

مقاييس التشابه الدلالي باستخدام الانتولوجي العربي

مقدم أطروحة جامعية

al-Dayri, Muhammad Ghandi

مشرف أطروحة جامعية

Kayid, Ahmad K. A.

أعضاء اللجنة

al-Sadi, Jihad
Abu Shurayhah, Ahmad

الجامعة

جامعة الشرق الأوسط

الكلية

كلية تكنولوجيا المعلومات

القسم الأكاديمي

قسم علم الحاسوب

دولة الجامعة

الأردن

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2017

الملخص الإنجليزي

The semantic similarity measures have been used in many applications including information retrieval and natural language processing.

There are many measures that use a lexical database such as WordNet to calculate the similarity between English concepts.

However, few researches have been studied semantic similarity measures using Arabic WordNet.

The traditional semantic similarity measures were classified into four categories: path-based measures, information content-based, feature-based measures, and hybrid measures.

Several measures from different categories have been applied on Arabic WordNet to which measure has the best performance using Arabic WordNet.

Human benchmark has been used to evaluate the performance of these measures over Arabic WordNet.

Experimental results show that the WuP measure has achieved the minimum mean square error (MSE) with value of (1.64%), and highest value of correlation coefficient with human ratings (0.92).

These results indicate that WuP measure has the best performance on Arabic WordNet compared to other measures.

Also, the results show that PATH measure has the worst performance.

This thesis proposed a new semantic similarity measure using the taxonomy of Arabic WordNet.

The new measure takes three factors into account: depth of concepts in Arabic WordNet tree, distance between two compared concepts and information content of the least common concept that subsumed two compared concepts.

The weight of these factors can be adapted manually.

However, several experiments have been conducted to find the best weight that achieves the minimum MSE.

In order to evaluate the new measure, the Arabic dataset that used previously to evaluate the measures has been used to test the new measure.

Then, the results of applying new measure over Arabic WordNet have been compared with the results of the other measures.

However, the results showed that the new measure has achieved the highest correlation coefficient with human ratings (0.96), furthermore, the new measure has obtained a very good value of MSE (1.89%) compared with the other measures

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

الموضوعات

• الويب الدلالي

عدد الصفحات

87

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-762685