This project develops reinforcement learning methods for combinatorial optimization, using influence maximization in networked systems as a representative case study. Influence maximization seeks to select a limited number of seed nodes that maximize the expected spread of information, behaviors, or interventions over a graph under a given diffusion model. The problem is computationally challenging because evaluating a candidate solution typically requires repeated stochastic diffusion simulations, and the combinatorial search space grows rapidly with network size. The proposed work will formulate influence maximization as a sequential decision-making problem, where an agent learns a node-selection policy conditioned on graph structure, diffusion dynamics, and previously selected seeds. The learned policy will be trained and evaluated across different network topologies and diffusion models, with the goal of improving scalability and generalization relative to simulation-heavy classical approaches. The method will be benchmarked against greedy approximation, degree centrality, PageRank, and community-based heuristics. The expected outcome is a scalable learning-based framework for graph-based combinatorial optimization, with applications in mobility systems, infrastructure planning, public information campaigns, and the adoption of sustainable technologies.