Espace membre

Cet espace est dédié aux étudiants, aux enseignants et au personnel administratif de l'école


Mot de passe oublié?



Doctorate thesis defense of Armielle NOULAPEU NGAFFO

Doctorate thesis defense on March 13th 2021 at 10H00 ,in Sup’Com Amphitheater Ibn Khaldoun.

Entitled :Dynamic Service Discovery based on Big Data Analysis using ICN Networks

Presented by :Armielle NOULAPEU NGAFFO


President :


Professor at SUP'COM, University of Carthage




Reviewers :


Professor at ENSI, University of Manouba


Lamia Chaari Fourati

Professor at ISIMS, University of Sfax


Examiner :

Kaouthar SETHOM

Professor at ENICarthage, University of Carthage




Supervisor :


Professor at SUP'COM, University of Carthage


The ever-growing of contents (movies, books, games, etc.) on the Internet has enhanced complexities for telecommunication service providers (TSP) that compete to retain and satisfy their customers by proposing services and products matching their needs. Indeed, the growth of access technologies such as 2G, 3G, 4G, and 5G coupled to the Next Generation Networks (NGN) precisely the IP Multimedia Subsystem (IMS), have encouraged the emergence of numerous and various services that aim to fulfill customers' wishes. The multitude of existing Information Technology (IT) digital services is not effectively discovered by customers since it becomes a challenging task for them to easily and quickly retrieve products that they desire. This mismatching between wanted services and products, and their potential customers occurs a colossal shortfall for Services and Telecommunication Service Providers (STSP). In addition, the still-increasing mass of data namely Big Data on the Internet offers improved opportunities to refine the services or products proposal to customers. The major issue for competing TSPs remains to fully satisfy, and even to anticipate on users' expectations. In conjunction with services, Big data is defined through "4V" features to refer to the volume, the velocity, the variety, and the value of the colossal amount of data that needs to be processed. Big data offers tremendous opportunities since competing service providers explore an impressive amount of data and then services addressable by users, in order to adjust their proposal to users' expectations and therefore increase their incomes. For several years, e-commerce platforms and social networks have become a true granary of data that provides service providers with high-meaningful information related to users' interests and expectations. From this fact, the users' data analysis appears as a significant opportunity that enables the service discovery enhancement by fitting users' needs with relevant services that will, therefore, improve the user experience. The increased user demand regarding online various services has motivated researchers to develop high-performant service discovery approaches to meet users' expectations and enable the increase in profits of service providers. According to the literature, recommender systems are among the most influential tools that efficiently perform service discovery by meeting various constraints. They are used in particular by famous platforms such as YouTube, Facebook, Twitter, etc.

Recommender Systems (RS) are powerful filtering tools that have been developed to help users to quickly target services or products that meet their expectations. The idea lays on the usage of historical users' experiences to predict future ones. The additional and helpful sources of information are joined to the recommendation process to refine the prediction of users' interests. A service or product is therefore recommended when it presents a high predicted score regarding the end-user. RS are spreadly adopted to help fructify the product's proposal of STSP.

The concept of Information-Centric Networking (ICN) has been developed to fluidify contents distribution since the current Internet has shown some limitations in terms of availability, persistence, security, and reliability. We first study the challenges and opportunities of the service discovery fed by massive data and we present the limitations of existing service discovery solutions. Afterward, we introduce recommender systems since their effectiveness is widely recognized by researchers and involved in famous platforms such Facebook, YouTube, Twitter, etc. We deeply survey state-of-the-art recommendation models and we study possibilities of recommender systems in the ICN context. Several studies are proposed recommendation approaches to alleviate the information overload issue and therefore the item targeting problem. However, the current solutions show some limitations in terms of accuracy and reliability of results returned to the final user. In addition, the data sparseness and cold-start problems are issues that severely affect the performances of existing solutions. Indeed, in case of a lack of historical users' experiences or in case of poor users' profiles, the current solutions' performances are negatively impacted. In this work, we have focused on recommender systems since they are among the most influential tools adopted in the service discovery field. They ensure both precision of results, the relevance-in-time, the data sparsity robustness, and credibility of propositions made to the end-user and even in challenging conditions.

This work aims to effectively address the service targeting issue while meeting all pending challenges related to the accuracy of the users' interest prediction, the reliability of suggestions addressed to users, the effective handling of data sparsity problem, the relevance-in-time of results returned to users, and the system scalability in order to efficiently deal with the still-growing amount of users data. Our proposal is a recommender framework that tackles the information overload issues in order to enable Information and Communication Technologies (ICT) operators especially service providers to improve users' experiences by meeting their expectations but also to increase their profits. Our recommender framework is fed by various user data in order to meet users' requirements in terms of the righteousness and reliability of recommendations. For this purpose, we develop effective algorithms that efficiently perform a users' interest prediction and that thereafter proceed to trusted recommendations. To achieve this, we first survey state-of-the-art service discovery approaches and especially recommender systems that enable a relevant services discovery by users. Thereafter, we present our recommender framework and its components, and we detail the prediction process in this framework. Finally, we detail the trusted recommendation process that ensures the reliability of results returned to the final user. Our proposal outperforms other recommendations approaches thanks to its data-sparsity resilience, its prediction accuracy, and the trustworthiness of results.

Keywords :

Service Discovery, Big Data, Recommender Systems, Collaborative Filtering, Deep Neural Networks, Machine Learning, Bayesian Inference, Information-Centric Networking, Information Overload, Personalized Item targeting, Context-based Recommendation.