In these days, personalized news recommender systems (RSs) have become an important and promising research field in the internet. RSs provide fast access to real-time news items from millions ofsources around the world. A key challenge of news sources is to help humans get the news that are interesting to read. Personal assistant agents (PAAs) can assist users to deal with the task of selecting news items and making decisions.
PAAs consider their mental states such as emotions at time recommend news, Affective recommender system (ARS) which represents an important and newly trending era of research. ARS can consider associating with human behavior, emotions, mood, and physiology with human computer interaction. Anyway in news field, user emotion plays significant roles in the decision making process.Taking advantage of any language is a means to expression of user emotion, emotion extractionone of the applications of natural language processing.The objective of this thesis is offering news to users accordingly to their preferences and to making positive news items that can have positive impact on users’ (mind and soul). Arabic news recommender system relies on users’emotions statein this thesis, and emotional Arabic news recommender system(E-ANRS)has focused on learning what users desire to read.We investigate the feasibility of a hybrid RS which integrates two algorithms as follows collaborative filteringand content based approaches. Our thesis considers the following aspects, implementation of (E-ANRS) application to display news articles for our users using android platform. It depends on RSSfor gathering news from multiple popular news sources involving four types of news such as Politics, Sport, Social, and Economy. Gathered Arabic news is suitable to user’s preferences and user’s needs on the basis of their profile, context, and data quality concerning the personal path. It stores users’ feedbacks including two categories as like/dislike in our database about recommend Arabic news to use in the future to perfection of RS operation. In order to predict similarities two approaches have been implementedTF/IDFand vector space model, as well as a validation factor.In this thesis we used Arabic news for the first time, and stemming algorithm for Arabic root extraction through morphological analysis from news title. In order to support the efficacy of our manner, we have developed this approach by using some techniques. Watson is an artificial intelligent cognitive system. The IBM Watson Developer cloud provides a library of cognitive services as REST APIs which are available on IBM Bluemix. We introduce an E-ANRS as a solution to problem of cold-start by using IBM Bluemix serverfor first time, which provides two servicesare (language translator and tone analyzer). The experimental results obtained from our research are, evaluation of RS which givesthree parameters being precision (86%), recall (87%), and F1-score (86%). In another side we have two ways to measure accuracy of emotion for our model by using EEG and Self-Assessment-Manikin techniques. By the use of EEG signals as (attention and meditation), electrical activity of neurons within the brain EEG is used. The obtained result of EEG is 90%, and from SAM technique it measures 3 states being pleasure, arousal, and dominance related to users’ emotional reaction. Reading Arabic news was at the beginning of the test is “5.928571”, and then becomes “6.585714” in the end of test. The results of IBM service Language translation and tone analyzer accuracy for 40 testsare 42%. We compared performance of our proposed model with other studies and the results proved our model offers a better and perfect recommendation process and emotion extract performance greatly improved with the use of news texts in Arabic languages.