Abstract: Forecasting the dissemination effect of cultural UGC in the short video platform environment is not only helpful to improve the operation efficiency of the platform for massive videos and achieve accurate recommendation service, but also important for content resource trading and high-quality UGC creation. Based on Laswell’s “5W” communication theory, we construct a cultural UGC communication effect prediction index system, and propose a combined GRA-PSO-WRF prediction model by integrating multiple sources of heterogeneous data such as operation statistics, third-party statistics and user comment data. The model uses comprehensive gray correlation analysis to extract the features of the fused data, achieves the prediction of UGC dissemination effect by particle swarm optimization weighted random forest algorithm, and finally conducts an empirical analysis based on the dissemination data of 90 cultural UGCs in four short video platforms. The study results show that the prediction index system of cultural UGC based on Laswell’s “5W” theory can systematically portray the key factors in the process of cultural content resource dissemination, and has good predictive applicability. The prediction accuracy of GRA-PSO-WRF model with multi-source heterogeneous data integration is significantly better than that of single RF model and BP model, and has certain interpretability. The influence of different features on the UGC spreading effect presents three levels: high-level, medium-level and low-level, and the optimal weights of the spread content factor and the spread channel factor are relatively higher, which is the key to influencing the spread effect of cultural UGC. The conclusion of this paper effectively complements the relevant techniques of combined prediction, multi-source heterogeneous data fusion modeling, and interpretable machine learning, and has practical insights for UGC creation and platform content operation.

Key words: cultural UGC, communication effectiveness prediction, comprehensive gray correlation analysis, particle swarm optimization, weighted random forest