Research On Cold-Start Recommendation Model Based On Model-Agnostic Meta-Learning

The recommendation system can combine user and item information for information filtering,efficiently and accurately locate the information needed by the user.However,for the increasing number of new users and new items,there is insufficient interactive information and description information in the recommendation system,these new users and new items cannot be recommended effectively,which we call the cold-start problem.In recent years,deep learning has made remarkable achievements in various fields.Many scholars try to combine deep learning methods to solve the cold-start problem.A recommendation model based on deep learning requires sufficient data to support the construction of the model.However,the interaction information between users and items in actual application scenarios is sparse,and users or items with less interaction information are easily ignored by the deep recommendation model,which affects its recommendation performance in a cold-start environment.To solve the above problems,we propose a Model-Agnostic Meta-Learning based coldstart recommendation model RCSM(Reptile-based Cold-start Recommendation Model)based on the meta-learning user preference estimation model MELU(Meta-learning User Preference Estimation).RCSM regards the preference estimation of each user as a new task,models the preference estimation of users with sparse interactive information as a few-shot learning problem.After that,the user preference prediction model is trained with the help of the Reptile meta-learning training framework,which forces the prediction model to quickly grasp the user's preferences based on experience and knowledge through a small number of training samples,to better solving the cold-start recommendation problem.To verify the effectiveness of the model,we have carried out a lot of comparative experiments on two real data sets.The results of the comparative experiments show that RCSM is better than the existing meta-learning recommendation model and user preference prediction model.we designed a control experiment of the RCSM and MELU model with the help of the controlled variable method.The effect of RCSM is better than the MELU model in different variable environments,which proves the effectiveness of Reptile application in solving the cold-start recommendation problem.