SOCIAL NETWORK ANALYSIS USING MACHINE LEARNING APPROACHES

Abstract

Social network analysis is vital for uncovering meaningful patterns, structures, and dynamics within social networks, with applications spanning health, marketing, and finance. One prominent application is viral marketing, which harnesses social relationships and interactions on social media platforms to sway consumer behavior. Social media allow users to connect, share opinions about the product and services. In such a scenario Opinion leaders (OLs), individuals with expertise in specific subjects and a substantial number of followers, play a pivotal role in shaping others' opinions on social media provide a boost to viral marketing. Therefore the success of viral marketing campaigns is crucial to identifying opinion leaders and individuals with substantial influence within their networks. Machine learning techniques have significantly advanced the development of accurate algorithms for identifying and evaluating these influential users, thereby maximizing the impact of marketing efforts. This work explores the feasibility of leveraging deep learning to approximate user influence. DeepWalk-based Influence Maximization (DWIM) algorithm is proposed that employs graph embedding techniques to identify the most influential nodes within the network. The work introduces a seed selection framework for maximizing influence in pervasive healthcare, utilizing machine learning approaches to investigate the bidirectional effects of influence and trust. The proposed framework addresses challenges associated with a large number of patients, ultimately enhancing influence maximization through strategic seed selection. Specifically, the Fuzzy-VIKOR algorithm is proposed to identify target nodes that facilitate the rapid dissemination of information. By effectively tackling issues inherent in large patient populations, the framework proves beneficial for pervasive healthcare applications. Furthermore, the thesis presents a Multi-Neighbor seed selection approach to enhance influence maximization. This approach accounts for the network's memory effect or social reinforcement effect and employs the Neighbor Degree Value (NDV) to estimate Page | vii the influence strength of selected seeds. It addresses challenges related to seed selection, such as limited coverage area and inadequate discriminatory power. The proposed Judgment Leader Pick Weighting Grade method, incorporating Judgment Grade Value for leader selection, effectively resolves trust management issues in ubiquitous services, ultimately amplifying influence maximization. This thesis explores the potential of Ant Colony Optimization (ACO), a nature-inspired optimization algorithm based on ants' foraging behavior, for influence maximization in social networks. It presents the framework for maximize influence spread by identifying seeds in the network for specific situations and issues. Experiments are conducted to evaluate the proposed algorithms, and centrality measures are employed for result comparison. The experimental findings demonstrate that the proposed frameworks exhibit high precision, accuracy, F1-score, and recall compared to existing algorithms for influence maximization seed selection. Furthermore, the proposed methods efficiently identify a set of influential nodes within a computable timeframe, facilitating viral marketing efforts and enabling targeted recommendations for various products and services, In summary, this thesis underscores the importance of social network analysis and machine learning techniques for identifying influential individuals in social networks, enabling more effective viral marketing and recommendations. The research introduces the DeepWalk Based Influence Maximization (DWIM) algorithm, which surpasses traditional centrality measures by integrating topical and topological features. DWIM holds promise for applications in diverse fields such as online marketing and outlier detection. The thesis presents two algorithms for identifying influential opinion leaders that incorporates user interaction and trust management and can be potentially applied in viral marketing and outlier detection across various services. The effectiveness of the algorithms has been studied on pervasive healthcare, politics, product promotion, and service promotions. Notably, in healthcare, it enhances targeted service delivery, demonstrating superior performance compared to existing algorithms in terms of accuracy, recall, f1- score, and precision. Overall, this work contributes significantly to the advancement of Page | viii social network analysis by providing a comprehensive framework for influence maximization and trust management.