Multi-user Multivariate Multi-order Markov Multi-modal Prediction And Its Efficient Computing

The globalization of big data has become the mainstream era,in view of the characteristics of large storage,categories,fast and low value of big data,the tensor model is introduced to effectively represent and efficiently analyze big data.However,with the rapid increase of tensor size,it leads to the problem of "curse of dimensionality".In order to reduce the complexity of curse of dimensionality calculation,tensors train can be used for distributed storage and parallel computation,so as to reduce the noise,reduce the storage and improve the computational efficiency.Therefore,this paper applies it to the trajectory prediction of multi-user multi-order Markov behavior.The study introduces the multi-user spatial and temporal distribution interdependence into the higher-order tensor model,increases the user correlation,and makes the model prediction more accurate.Then,combined with the advantages of tensor chain decomposition,the trajectory model is computed in parallel architecture to optimize the overall computation process.In view of the streaming characteristics of big data,the incremental feature tensor calculation method is studied,and the updated feature tensor is obtained by using the calculation results of historical data,so as to avoid the problem of repeated resource consumption in the reconstruction model.It is concluded that the model after joining user influence obviously improve the prediction accuracy,and parallel computing based on the tensor chain jumped to improve forecast accuracy,reduce memory overhead and improve calculation efficiency advantages,and based on the incremental feature tensor also reduce repeated computation time.