Research On Many-objective Hybrid Recommendation Model

Recommendation system are designed to recommendation personalized items or information for users.It is now widely used in many network applications to deal with information overload caused by massive amounts of data and information to enhance the user experience.With the development of Internet,the proportion of online life in people’s daily life has increased,and uses’ demand for recommendation systems has also gradually increased.However,most of existing recommendation models only focus on the accuracy and diversity of recommendations,which are not enough to meet the diverse needs of users.To address this problem,this paper constructs a hybrid recommendation model based on a many-objective optimization algorithm that can optimize multiple recommendation performance simultaneously,and the specific work is as follows:To address the problem that most existing recommendation models only focus on the accuracy and diversity of recommendations and the limitations of the applicable scenarios of a single recommendation technique,this paper proposes a hybrid recommendation model based on many-objective optimization.First,the relationship between multiple optimization objectives is analyzed to simulate the user’s needs from multiple perspectives.Then,by linearly mixing three basic recommendation techniques and using a rating-based recommendation mechanism,the accuracy,recall,diversity,novelty and coverage of recommendations can be optimized at the same time.The experimental results demonstrate that solving this model using multiple existing many-objective optimization algorithms can provide users with a recommendation list that contains a variety of excellent performances,and the superiority of this model is verified by comparing the performance with the rest of the models.In order to improve the performance of the above proposed hybrid recommendation model based on many-objective optimization,two different mechanisms of many-objective optimization recommendation algorithms are designed.In view of the fact that most of the individual evaluation strategies in many-objective optimization algorithms only focus on the convergence and diversity of individuals,and ignore the impact of algorithm running algebra on individual evaluation.According to the core idea of evolutionary algorithms "survival of the fittest",generation based individual evaluation strategy and partition-based knowledge mining strategy are used to construct a many-objective recommendation algorithm based on knowledge mining.Among them,the generation based individual evaluation strategy gives the individual convergence a larger proportion of the individual evaluation in the early stage of the algorithm operation.On the contrary,the diversity accounts for a larger proportion in the later stage of the algorithm,thereby increasing the selection pressure of the algorithm.The partition-based knowledge mining strategy extracts the knowledge generated during the operation of the algorithm to guide the constraint processing process.Since the above model contains constraints,in order to ensure the individuality of the solution within the population during the operation of the algorithm,the individuals that do not satisfy the constraints after the population update can be retraced to the individual’s historical optimal position to speed up the optimization process.Based on the core idea of "experience sharing" of swarm intelligence algorithm,we propose a partition-based individual updating strategy and a many-objective bacterial foraging strategy,and use the idea of "complementary advantage" to build a many-objective particle swarm optimization algorithm based on multiple criteria.Firstly,we select the best solution by multiple evaluation criteria,and secondly,we use partition-based individual update strategy to add perturbation in the process of individual update to improve the convergence and diversity of the algorithm.A many-objective bacterial foraging strategy is used in the later stage of the algorithm to speed up the local search of the algorithm.The experimental results demonstrate that both of the above algorithms can produce excellent results in solving the benchmark problems of the optimization algorithm and the recommended model.