Research On Package Recommendation Algorithm For User Preferences

The traditional item recommendation system had not meet the needs of users in a number of specific field,such as travel plan and food catering recommendation.So package recommendation was born at the right moment.Due to the characteristics of package recommendation,the complexity of result analysis is higher,and recommended results are more difficult to select,it still faces many challenges to apply existing recommendation algorithm to the package recommendation scenario.For these purposes,user similarity calculation,package recommendation algorithm and other related research was discussed and summarized to study package recommendation algorithm for user preferences in this paper.The main work is as follows:(1)The project background,research significance,basic concepts and principles of package recommendation,recommended process,and related technologies in common use are introduced.We also made a review of the research and current applications about package recommendation,and analyze the existing problems in package recommendation system.(2)Aim to express the degree of similarity between users accurately,a similar calculation method based on grey correlation is proposed.Using a grey correlation to measure the overall level of user ratings and calculate user similarity;Also consider the importance of different projects in user similarity calculation,and determine the weight of the project based on the number of user ratings.our similarity calculation method can improve the accuracy of the rating prediction effectively.(3)To reduce the size of the package the recommended complexity,a package recommendation algorithm based on utility computing is proposed.Consider the factor of project rating and cost at the same time,a utility computing method is designed.According to the project utility on the basis of meet the constraint conditions,select the current highest utility item to fill the results package in turn.The results of the experiment show that compared with the PackRec algorithm and the GV selection algorithm,our method has a good performance in the evaluation indexes such as MAE,nDCG and Kendall correlation.(4)To improve the quality of recommendation and user experience.A package recommendation TOPSIS sorting method based on K-L divergence is proposed.Introducing a concept of goal ideal,adding user preferences to the TOPSIS method,making the sorting results more relevant to the user's needs;At the same time,according to the k-l divergence,a new method of calculating the method of proximity degree is proposed.Compared with the comparison algorithm,the results of this chapter method have higher utility and better quality of the recommended results.