CatDM

A Category-Aware Deep Model for Successive POI Recommendation on Sparse Check-in Data
Fuqiang Yu, Lizhen Cui, Wei Guo, Xudong Lu, Qingzhong Li, Hua Lu

Installation

Install Python 3.5.

Install tensorflow 1.12.2.

Data

Datasets for training/evaluation.

NYC: the Foursquare check-ins in New York
TKY: the Foursquare check-ins in Tokyo

Usage

1. Generate Candidates

To filter POIs and reduce the search space.

$ python train.py

To train and evaluate Encoder 1 and Filter, we split each dataset into a training set, a validation set and a test set, here. Encoder 1 and filtering layers form a reasonable filter capable of reducing search space, i.e., reducing the number of candidates from which recommended POIs are selected finally.

Note that the value of variable 'tf.flags.DEFINE_string' can be selected by train or test.

2. Rank POI

To sort the POIs in the candidate set.

$ python train_rankpoi.py