التوصيات المعتمدة علي السياق في تطبيقات الويب الاجتماعية

الموضوعات

علوم تطبيقية وتكنولوجية

روابط وملفات

التعريف والنوع

رقم الوثيقة

fc8fbc2a-41ce-47dc-be32-4908d7d0f876

رقم العقد

0

نوع الوسائط

Crawler

نوع المحتوى

الرسائل العلمية

صيغة المصدر

رسائل دكتوراة

نوع الملف

pdf text

أسماء الملفات

1957393_1.pdf

بيانات النشر

ترجمة العنوان

Context-Aware Recommendations in Social Web Applications

ألقاب المؤلفين

[{"name_ar":"ايفا دياب حريقص","title_ar":"اعداد","title_en":"Preparation"},{"name_ar":"يسر السيد سليمان الأتاسي","title_ar":"اشراف","title_en":"Supervision"}]

اللغة

Arabic

المصدر والدورية

اسم المصدر

التوصيات المعتمدة علي السياق في تطبيقات الويب الاجتماعية

المحتوى والصفحات

عدد الصفحات

0

ترجمة الملخص

Recommender systems are one of the recent inventions to deal with information overload problem and provide users with personalized recommendations that may be of their interests. Collaborative filtering is the most popular and widely used technique to build recommender systems and has been successfully employed in many applications. The essential principle is that users with similar preference in the past are likely to prefer the same items in the future. However, it suffers from several inherent issues that affect the recommendations accuracy such as data sparsity and cold start problems caused by the lack of user ratings, which make the recommendation results often unsatisfactory .To solve these problems and model user preferences more accurately, additional information can be merged into the collaborative filtering method to compensate for inadequate rating information. Recent studies demonstrate that social context information can be exploited to improve performance of recommender systems. with the advent of online social networks, the social recommender systems has emerged, in which recommendations are made to the user based on the ratings of the users who have direct or indirect social relations with this user. The trust relation is one of the most important kinds of social relations, due to its positive and strong correlation with similarity between users, and several studies have shown the significant effectiveness in improving predictive accuracy for traditional recommendation techniques. The main objective of this thesis is to suggest effective recommendation methods that help in solving problems that the traditional collaborative filtering suffers from, in order to improve the recommendations accuracy by taking advantage of the user's social context. Specifically, we first propose a memory-based recommendation method that integrate sratings of trusted neighbors according to their importance in order to form a more complete rating record than the original record for the target user, then based on the new rating record, the rating for the target item is predicted using the conventional collaborative filtering method to find more reliable similar neighbors, but with suggesting slight difference from the traditional method so that if a similar neighbor had not rated the target item, we will predict that rating using the ratings of his directly trusted neighbors, to combine more similar users in generating a prediction for this target item. With this strategy, recommendations can be better generated with higher predictive accuracy and coverage, and is especially useful for the cold-start users as their preference is approximated by the trusted neighbors. Secondly, we propose a model-based recommendation method that uses matrix factorization technique to model user preferences, and exploit both global and local social context of trust relations simultaneously for recommendations. Local vin formation represents the preferences of the two different situations for user (as a trustor and a trustee) that are learned by modeling explicit and implicit interactions between users, while global information represents the reputation of the user in the social network as a whole. By taking both trustor's and trustee’s preferences at the same time in the learning process of the model, better recommendations can be generated with higher predictive accuracy, especially for cold-start users, that is due to expressing the users’ reciprocal influence on the opinions of each other more reasonably, rather than simply combine two types of data as most previous studies do, and the sparse trust data can be leveraged more effectively. Thirdly, we examine the possibility of using distrust information to additionally enhance the performance of trust-based recommender systems, and propose a model-based recommendation method that incorporates both trust and distrust relations simultaneously into recommender systems using matrix factorization technique, by finding the latent features for users such that each user is brought closer to the users he trusts and separated from the users that he distrusts and who have different interests, where the data analysis of dataset with both trust and distrust relations showed that the users are likely to be more similar with their trusted friends group than their distrusted foes group. With this strategy, better trust-based recommendations can be generated with higher predictive accuracy. To summarize, we have proposed three different methods that exploit the user's social context in handling the data sparsity and cold-start problems, and have been experimentally evaluated and compared against state-of-the-art methods on real life data sets. Experimental results showed that our proposed methods achieve substantial gains in recommendation performance compared to the existing methods

كلمات الباحثين

Collaborative Filtering, Distrust, Recommender Systems, Social Recommender Systems, Trust, التصفية التعاونية, النظم الناصحة الاجتماعية, النظم الناصحة, عدم الثقة

إشراف وإعداد

الإشراف

يسر السيد سليمان الأتاسي

الإعداد

ايفا دياب حريقص

الاقتباسات الببليوغرافية