Trajectory k nearest neighbor query method based on sparse multi-head attention
The application of location information generates a large volume of spatio-temporal data with multidimensional attributes such as latitude,longitude and time.The index-based trajectory query method cannot obtain the spatio-temporal semantic features of the trajectory,and its trajectory representation ignores the spatio-temporal correlation of the trajectory and is inefficient for large-scale trajectory data query.To solve the above problems,a kind of trajectory encoder based on sparse multi-head attention is proposed,through the trajectory of the encoder can extract high-level semantic features,trajectories are represented as encoding vectors.On the basis of trajectory coding vector,a trajectory query method based on locally sensitive hash function is developed,which can query large-scale trajectory data quickly.The experiment results show that the proposed trajectory query method can effectively deal with kNNT query problems and it is superior to the current trajectory query method in terms of accuracy and efficiency.
computer applicationspatio-temporal datatrajectory data miningdeep learningtrajectory representation learningmulti-head self-attention
万方数据
