UCGGAT: User Purchase Intention Prediction Based on User Fine-grained Module Click Stream of Product Detail Page
Here we provide the implementation of a User Click Graph-Graph Attention Network (UCGGAT) layer in TensorFlow. The repository is organised as follows:
data/put you data here (The data form is a diagram composed of click module flow);models/contains the implementation of the UCGGAT network (ucggat.py);pre_trained/contains a pre-trained UCGGAT model;
Finally, execute_ucg.py puts all of the above together and may be used to execute a full training run on you data by executing python execute_ucg.py.
The script has been tested running under Python 3.5.2, with the following packages installed (along with their dependencies):
numpy==1.14.1scipy==1.0.0networkx==2.1tensorflow-gpu==1.6.0
In addition, CUDA 9.0 and cuDNN 7 have been used.
This work was supported by the National Key R&D Program of China under Grant No. 2020AAA0103804 (Sponsor: Hefu Liu) and partially supported by grants from the National Natural Science Foundation of China (No.72004021). This work belongs to the University of science and technology of China.