Research On Personalized Recommendation Algorithm Based On Multi-objective Immune Optimization

With the development of Internet technology and the increasing of network users,the problem of information overload has been more and more serious.Personalized recommendation has become one of the most potential technologies to deal with the problem of information overload.When making news recommendations,the existing recommendation algorithms have problems such as lack of time utility,limitation of similar user groups,and conflicting of multiple targets.For solving the above problems,this thesis proposes a personalized recommendation algorithm based on multi-objective immune optimization by combining recommendation algorithm with multi-objective evolutionary algorithm to solve the conflict between recommendation accuracy and diversity.Firstly,in view of the strong timeliness of news recommendation,the time influence factor is introduced to improve the content-based recommendation algorithm,and the user's own preference model is constructed.Next considering the relevance of time utility and popularity of news towards the users' interests,a new hybrid similarity calculation method which can find the nearest neighbor discover the users potential preference,and construct the user's potential preference model is proposed.Then the fusion interest model is generated by integrating the user's own preference model with the potential preference model.And the candidate news set is obtained through using the fusion interest model to eliminate the news that have nothing to do with the users' interests.Lastly,the news recommendation problem is modeled as a multi-objective optimization problem for recommendation accuracy and diversity and the candidate news sets are further optimized by using NNIA to obtain the final recommendation results.For verifing the effectiveness of the recommendation algorithm based on fusion preference model and the personalized recommendation algorithm based on the multi-objective immune optimization in this thesis,the Caixin data sets are used to make experiments.The experimental results show that the accuracy,recall rate and diversity of the recommendation algorithm based on the fusion preference model proposed in this thesis are all superior to the comparison algorithm,and it can get an effective news candidate set.The personalized recommendation algorithm based on the multi-objective immune optimization can effectively obtain the optimal solution of personalized remmendations.The accuracy and diversity of this algorithm are better than existing algorithms,The maximum accuracy increased by 11.07% and the maximum increase in diversity is 2.52%.So it has the ability of providing users with the diversification news content for conforming to the user's own interests.