POISketch: Semantic Place Labeling over User Activity Streams

Authors: Dingqi Yang, Bin Li, Philippe Cudré-Mauroux

IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical evaluation on real-world datasets demonstrates the validity of our approach and shows that sketches can be efficiently and effectively used to infer place labels over user activity streams.
Researcher Affiliation Academia 1e Xascale Infolab, University of Fribourg, 1700 Fribourg, Switzerland 2Data61, CSIRO, Eveleigh NSW 2015, Australia
Pseudocode No The paper describes methods textually but does not include structured pseudocode or algorithm blocks.
Open Source Code No No concrete access (e.g., specific repository link, explicit statement of code release) to the source code for the methodology was found.
Open Datasets Yes We evaluate our approach on a check-in dataset collected by [Yang et al., 2015a; 2016] for about 18 months (from April 2012 to September 2013). Without loss of generality, we select check-in data from two big cities, New York City and Tokyo, for our experiments.
Dataset Splits No The paper describes a training and testing split (first 9 months for training, last 9 months for testing) but does not explicitly mention a separate validation split with specific details.
Hardware Specification Yes All experiments were conducted on a commodity PC (Intel Core i7-4770HQ@2.20GHz, 16GB RAM, Mac OS X) running MATLAB3 version 2014b.
Software Dependencies Yes All experiments were conducted on a commodity PC (Intel Core i7-4770HQ@2.20GHz, 16GB RAM, Mac OS X) running MATLAB3 version 2014b.
Experiment Setup Yes We empirically set KNN with the five nearest neighbors, and report the classification accuracy on both the 9 root categories (Lv1) and the 291 sub-categories (Lv2). The classification is triggered when 10 check-ins are observed for each POI tested. We set the parameters d = 10, w = 50 to guarantee an error within 4% with probability 0.999. In this experiment, we compare our method to the baseline approaches by fixing the sketch length to 50 for all sketching methods.