Join order selection is a sub-field of query optimization that aims to find\nthe optimal join order for an SQL query with the minimum cost. The challenge\nlies in the exponentially growing search space as the number of tables\nincreases, making exhaustive enumeration impractical. Traditional optimizers\nuse static heuristics to prune the search space, but they often fail to adapt\nto changes or improve based on feedback from the DBMS. Recent research\naddresses these limitations with Deep Reinforcement Learning (DRL), allowing\nmodels to use feedback to dynamically search for better join orders and enhance\nperformance over time. Existing research primarily focuses on capturing join\norder sequences and their representations at various levels, with limited\ncomparative analysis of reinforcement learning methods. In this paper, we\npropose GTDD, a novel framework that integrates Graph Neural Networks (GNN),\nTreestructured Long Short-Term Memory (Tree LSTM), and DuelingDQN. We conduct a\nseries of experiments that demonstrate a clear advantage of GTDD over state-of\nthe-art techniques.\n