You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
tal d81bfb3bbb add MaKE 1 year ago
data add MaKE 1 year ago
misc add MaKE 1 year ago
model add MaKE 1 year ago
processed_data add MaKE 1 year ago
test add MaKE 1 year ago
utils add MaKE 1 year ago
README.md add MaKE 1 year ago
dataset_dual.py add MaKE 1 year ago
preprocess_data.py add MaKE 1 year ago
requirements.txt add MaKE 1 year ago
train_no_plan.py add MaKE 1 year ago
train_no_sns.py add MaKE 1 year ago
train_no_symbolic.py add MaKE 1 year ago
train_res.py add MaKE 1 year ago
train_res_seed_16.py add MaKE 1 year ago
train_with_reg.py add MaKE 1 year ago

该算法提出一个端到端的神经网络模型,可以利用常识知识图谱和方程表达式生成丰富的数学应用题(MWP)。所提出的模型(1)从符号方程表达式和常识知识的 Levi 图中学习表征;(2)在生成MWP 时,通过自我规划模块自动融合方程式和常识知识信息。在教育场景数据集的实验表明,和其它模型相比,该方法在 MWP 生成任务上有明显优势,在自动评价指标和人类评价指标方面都优于 SOTA 模型。在教育数据集中,方程相关性分数达到 2.308,主题相关性分数达到 2.558,语言连贯性分数达到 2.505。

Text Python

Contributors (2)