| With the advent of the era of big data and the continuous expansion of the Internet,users need to extract useful information for decision-making.In order to meet the information needs of users in suitable scenarios,recommender systems have emerged as the times require.Usually,the recommendation system will analyze the preferences of the current user,find out the information objects that the user may be interested in from a large amount of information according to the user’s historical interaction data,sort them,and finally present them to the user in the form of a list.At present,most recommendation models regard the recommendation process as a static process,and follow a fixed recommendation strategy when recommending.When the user’s preferences change dynamically,the recommendation model cannot effectively capture these changes,resulting in unsatisfactory recommendation results.According to the limitations of the above recommendation model,this paper proposes a method based on deep reinforcement learning for recommendation.The specific research contents are as follows.First,the method of deep reinforcement learning is used to introduce the Markov decision process to define the Top-N recommendation task and the Top-N recommendation process.The idea of knowledge distillation is introduced,and a Top-N recommendation model based on value improvement incorporating knowledge distillation is proposed.The model is divided into two parts: teacher model and student model.The knowledge in the teacher model is used to further guide the learning of the student model,thereby improving the performance of the student model on the Top-N recommendation task.Secondly,considering the limitations of policy degradation and high-dimensional action space when using value-based deep reinforcement learning methods for recommendation,a Top-N recommendation model based on policy improvement and incorporating knowledge distillation is proposed.Using a policy-based deep reinforcement learning approach,the student model is improved to enable it to learn a more effective recommendation policy.Finally,this paper conducts relevant experiments on three real datasets,compares the proposed model with the comparative models and conducts a comprehensive analysis of the proposed model to verify the effectiveness and feasibility of the proposed model. |
