An Improved Multi-Context Trajectory Embedding Model using Parameter Tuning Optimization for Human Trajectory Data Analysis

摘要

As the explosion of location-based social networks such as Facebook, etc., a number of ways have been provided for tracing human mobility, including user-generated Geo-tagged contents, check-in-services and mobile applications. For mining human trajectory data, different techniques were proposed over the past decades. However, the issue in many applications was analyzing and mining trajectory data due to the complex characteristics reflected in human mobility which is affected by multiple contextual information. As a result, Multi-Context Trajectory Embedding Model using Convolutional Neural Network (MC-TEM-CNN) was proposed that reduces the computation time during the learning process of contextual features. However, it requires an optimization algorithm to enhance the tuning of parameters which are needed to model the contextual information. Hence in this article, an Improved Multi-Context Trajectory Embedding Model (IMC-TEM) is proposed based on the frog-leaping optimization algorithm. In this algorithm, the parameters are tuned according to the frog characteristics. In each iteration, the global best fitness is chosen to adjust the position of worst fitness frogs. Thus, the proposed IMC-TEM tunes parameters in a better manner. Finally, the experimental results are conducted based on three real-world datasets to observe the performance efficiency of the IMC-TEM than MC-TEM-CNN.