Research On Personalized Recommendation Services Based On Item And Consumer Preference Perspectives

Recently,the problem of "information overload" becomes more and more serious with the development of the internet technology.Hence,the problem of achieving interesting and useful information from massive information has tended to be extremely important.The recommendation algorithm acts an essential role in the personalization recommendation service system.And it can alleviate "information overload" and improve the consumer experience through mining historical data generated by consumer behavior.But these data exhibit sparse,limiting its application.In addition,consumer interest and item popularity change over time,which can negatively affect recommendation results.Machine learning algorithms exhibit strong feature extraction and fitting ability.This paper combines it with the collaborative filtering algorithm to solve the above problems.Specifically:(1)A personalized recommendation model based on multiple similarity and Cat Boost is proposed.In order to solve the problem of low precision of recommendation system due to the sparse consumer-item scoring matrix,the modified cosine similarity function is first used to solve the similarity matrixes of item metadata and score data,and these similarity matrixes are fused.Then,the LINE is applied to perform multi-order similarity analysis on the fused similarity matrix to calculate a more accurate nearest neighbor set.Finally,the nearest neighbor set is sent into the Cat Boost to predict item scores and the items are recommended by top-N.Experiments are carried out on Movie Lens to verify the effectiveness of this algorithm.The results show that the proposed method enjoys higher recommendation accuracy and stronger stability,and can effectively solve the influence of historical data sparsity on recommendation systems.(2)A hybrid recommendation model based on category preference and item timeliness is presented.Aiming at the influence of category preference and item timeliness factors on the performance of recommendation algorithm,the Huffman coding is first applied to encode scoring data with consumer interest and item popularity.Then,we integrate user and item feature vectors extracted by the Deep Walk from consumer and item similarity matrices.Subsequently,the fused vectors are sent into the extreme learning machine to predict the item’s score.Finally,the Movie Lens dataset is taken as examples,and ICF,UCF,ELM,RBFCF and XGB-RF are regarded as comparison methods to examine this method under different proportional training sets.Extensive experimental results demonstrate that this method can effectively mitigate the influence of category preference and item timeliness factors on recommendation results.The above algorithms not only mitigate the impact of data sparsity,category preference,and item timeliness on personalized recommendations,but also help businesses to analyze consumer needs and provide more accurate personalized recommendations.