Prediction of Trust Propagation in Social Commerce based on Ensemble Learning

Accurate prediction of trust propagation in social commerce is vital to recommendation and promotion of commodities. Existing prediction models have some shortages, such as simple process of influencing factors of trust and low prediction precision. To address these problems, a prediction model based on Soft-Voting ensemble learning was proposed. Firstly, features of influencing factors of information propagation in social commerce were constructed from user attributes, information text and user interaction. Secondly, XGBoost, LightGBM and Catboost models were trained according to the above constructed features to predict trust propagation in social commerce. Finally, results of XGBoost, LightGBM and Catboost models were integrated using Soft-Voting technique as the final prediction results. An experiment on a real dataset of Sina Weibo was carried out, which proved the higher precision of Soft-Voting ensemble learning compared to those of XGBoost, LightGBM and Catboost models.