Search In this Thesis
   Search In this Thesis  
العنوان
Time Tagging for Enhancing Opinion Mining Prediction \
المؤلف
Diab, Ghada Hafez Hassan.
هيئة الاعداد
باحث / غادة حافظ حسن دياب
مشرف / عمر حسن كرم
مشرف / رشا اسماعيل
مناقش / عمر حسن كرم
تاريخ النشر
2019.
عدد الصفحات
98 p. :
اللغة
الإنجليزية
الدرجة
ماجستير
التخصص
Information Systems
تاريخ الإجازة
1/1/2019
مكان الإجازة
جامعة عين شمس - كلية الحاسبات والمعلومات - نظم المعلومات
الفهرس
Only 14 pages are availabe for public view

from 98

from 98

Abstract

Nowadays, opinion mining becomes one of the most important fields and it attracts the interest of many researchers. The ’electronic Word of Mouth’ (eWOM) statements that are expressed on the web, are important for business and service industry to enable customers share their point of view. In the last one and half decades, research communities, academia, public and service industries are working rigorously on opinion mining -which is also called, sentiment analysis to identify and categorize opinions from a piece of text. One key use of sentiment analysis is to extract and analyze public moods and views. Researchers used sentiment analysis in different ways. For example, to determine the market strategy that improve customer service.
One of the key challenges of sentiment analysis is how to extract temporal synsets from text. Temporal synsets may be events, dates, times, or even Explicit lyrics. Tempowordnet is one of the attempts to building a lexicon that may help in finding temporal synsets.
It is noticed that temporal word net contains 117598 word, 100512of them are classified atemporal, which means it doesn’t have any temporality in it. 12053of those atemporal words were found to be verbs, and this shows that they were wrongly classified to be atemporal. from this point we developed our system to work on those atemporal verbs and classify them.
This thesis presents a framework for enhancing opinion mining process. The framework presents two parts: The first part is for discovering temporal verb references (future, past and present) from opinions and using them to build accurate prediction models. The proposed framework improved the percentage of discovering (past, present and future verbs)over the tempowordnet.To enhance existing methods and make them more efficient, an algorithm that targets verbs is proposed. It extracts verbs based on tempowornet and classified them as past, present, and future. Experimental results showed that The accuracy for the proposed framework was 0.7%, 1.1%, 0.2%for the past, present and future respectively over that of the tempowornet.
The second part introduced in the proposed framework is enhancing the extraction of aspects and sentiment words by detecting shortcuts, emojis and sarcasm based on time tagging. The proposed framework are applied to a mobile reviews data set and compared to different systems based on theaccuracy, recall, precision and F1 measure. The results showed that our proposed work improved The accuracy of the proposed framework is 0.23%, 0.4% , 0.14 %and 0.24 %for total past, present , future and atemporal respectively over that of the compared systems, also improved the recall, and precision and f1 measure.