{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/m-walk-learning-to-walk-over-graphs-using","title":"M-Walk: Learning to Walk over Graphs using Monte Carlo Tree Search","arxiv_id":"1802.04394","date":"2018-02-12","proceeding":"NeurIPS 2018 12","authors":["Yelong Shen","Jianshu Chen","Po-Sen Huang","Yuqing Guo","Jianfeng Gao"],"abstract":"Learning to walk over a graph towards a target node for a given query and a\nsource node is an important problem in applications such as knowledge base\ncompletion (KBC). It can be formulated as a reinforcement learning (RL) problem\nwith a known state transition model. To overcome the challenge of sparse\nrewards, we develop a graph-walking agent called M-Walk, which consists of a\ndeep recurrent neural network (RNN) and Monte Carlo Tree Search (MCTS). The RNN\nencodes the state (i.e., history of the walked path) and maps it separately to\na policy and Q-values. In order to effectively train the agent from sparse\nrewards, we combine MCTS with the neural policy to generate trajectories\nyielding more positive rewards. From these trajectories, the network is\nimproved in an off-policy manner using Q-learning, which modifies the RNN\npolicy via parameter sharing. Our proposed RL algorithm repeatedly applies this\npolicy-improvement step to learn the model. At test time, MCTS is combined with\nthe neural policy to predict the target node. Experimental results on several\ngraph-walking benchmarks show that M-Walk is able to learn better policies than\nother RL-based methods, which are mainly based on policy gradients. M-Walk also\noutperforms traditional KBC baselines.","url_abs":"http://arxiv.org/abs/1802.04394v5","url_pdf":"http://arxiv.org/pdf/1802.04394v5.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[],"tasks":[{"task_slug":"knowledge-base-completion","task_name":"Knowledge Base Completion"},{"task_slug":"link-prediction","task_name":"Link Prediction"},{"task_slug":"q-learning","task_name":"Q-Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/link-prediction-on-wn18rr","task":"Link Prediction","dataset":"WN18RR","model":"M-Walk","rank_in_archive_order":72,"of":75,"metrics":{"Hits@1":"0.414","Hits@3":"0.445","MRR":"0.437"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1802.04394","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}