Tajuddeen / Distance2Pre

Code for my PAKDD-2019, Distance2Pre: Personalized Spatial Preference for Next Point-of-Interest Prediction

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Distance2Pre

Code for my PAKDD-2019. It is implemented by Python27 and Theano.

Distance2Pre: Personalized Spatial Preference for Next Point-of-Interest Prediction

BibTex:

@inproceedings{cui2019distance2pre,

title={Distance2Pre: Personalized Spatial Preference for Next Point-of-Interest Prediction},

author={Cui, Qiang and Tang, Yuyuan and Wu, Shu and Wang, Liang},

booktitle={Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD)},

pages={289--301},

year={2019},

organization={Springer} }

The annotations are in Chinese. Please note that we do not know any information during the test process.

If you have other questions or confusions, please send email to [cuiqiang1990@hotmail.com].

theano就大体看看吧,不用深究,我也不用这个了。

RNN的具体input什么。(1)t 时刻的poi embedding,20维。(2)t-1时刻到 t 时刻两个poi之间的距离间隔,转换为interval,并用20维的embedding表示。

输出预测。(1)训练结束,测试得结果之前,先再用用户训练序列重新走一下RNN,得到训练序列最后一个时刻的用户hidden state作为user vector。(2)用user vector * POI vector得到对所有poi的偏好得分。(3)用user vector得到对下次各个距离间隔不同偏好的得分st,就是一个interval预测出来一个得分。(4)每个训练序列的最后一个poi,计算它和所有poi的距离间隔并转换为interval。这一步在训模型之前做完,存起来,测试时直接取出来用。再根据刚计算出来的st把这些interval转换为距离偏好得分。(5)两个偏好分数都有了,再用模型里提到的线性、非线性融合方式得到总得分。

模型测试时,每个用户的test list就一个真值poi,并且预测时没有任何关于该真值poi的信息(比如空间位置等)。因此如果测试时已经有了真值poi的空间位置(有些文章这么做的),我的模型就不适用了。

About

Code for my PAKDD-2019, Distance2Pre: Personalized Spatial Preference for Next Point-of-Interest Prediction


Languages

Language:Python 46.3%Language:HTML 35.1%Language:MATLAB 13.0%Language:Jupyter Notebook 5.6%