公共安全事件风险知识图谱推理实验数据集

本数据集本数据集主要用于研究训练和验证风险推理技术所涉及的基于强化学习的时间区间时态知识图推理方法的相关性能指标，包括平均倒数排名（Mean Reciprocal Ranking）、候选三选组中，正确答案排在第一的数量（Hits@1）、正确答案排在第三的数量（Hits@3）、正确答案排在第十的数量（Hits@10）。数据集包括4个文件夹：ICEWS14数据集文件、ICEWS18数据集文件、WIKI数据集文件以及YAGO数据集文件，存放了用于测试验证风险推理模型所涉及的知识图谱多跳推理任务相关性能指标所需要的实验数据。 数据量共10.16Gb，其中ICEWS14数据集：623MB; ICEWS18数据集：523MB；WIKI数据集：5.00GB；YAGO数据集：4.02GB。This dataset is primarily used to investigate the performance metrics of reinforcement learning-based reasoning methods for temporal interval knowledge graphs, which are involved in training and validating risk reasoning technologies. The included metrics are Mean Reciprocal Ranking (MRR), the number of instances where the correct answer ranks first in the candidate triplet sets (Hits@1), the number of instances where the correct answer ranks third (Hits@3), and the number of instances where the correct answer ranks tenth (Hits@10). The dataset comprises four folders: ICEWS14 dataset files, ICEWS18 dataset files, WIKI dataset files, and YAGO dataset files. These folders hold the experimental data necessary for testing and verifying the performance metrics of knowledge graph multi-hop reasoning tasks associated with risk reasoning models. The overall data size of the dataset is 10.16 GB, with detailed sizes as follows: ICEWS14 dataset: 623 MB; ICEWS18 dataset: 523 MB; WIKI dataset: 5.00 GB; YAGO dataset: 4.02 GB.

搜集汇总

背景与挑战

背景概述

该数据集旨在支持公共安全事件风险知识图谱推理的研究，主要用于训练和验证基于强化学习的时间区间时态知识图推理方法，评估性能指标包括平均倒数排名和Hits@1等。它包含ICEWS14、ICEWS18、WIKI和YAGO四个子数据集，总数据量为10.16GB，为知识图谱多跳推理任务提供实验数据。