TWC Game Dataset

该数据集来源于TWC游戏，游戏中玩家需要通过将物品放置在正确的位置来整理房子，这需要玩家具备关于物体关系的常识。数据集包括两个测试集：一个与训练集实体相同的IN集，以及一个实体不同的OUT集，这适合于评估常识的使用情况。规模上，该数据集包含了100个游戏生成的数据，分为50个训练集、40个测试集和10个验证集。任务的目的是测试代理在需要常识的文本游戏中的表现。

This dataset is derived from the TWC game, in which players are required to tidy a house by placing items in their correct locations, a task that demands commonsense knowledge regarding object relationships. The dataset contains two test sets: the IN set, which shares identical entities with the training set, and the OUT set, which features distinct entities, making it suitable for evaluating the utilization of commonsense knowledge. In terms of scale, the dataset consists of 100 game-generated samples, split into 50 training samples, 40 test samples, and 10 validation samples. The objective of this task is to assess the performance of agents in text-based games that require commonsense reasoning.