| The Internet wave that has swept the world has not only accelerated the development of various information industries,but has also further expanded the ways in which users can access and disseminate information.In a sense,the more information one has,the more certain one can be about certain things,but we must understand that when there is too much information,it becomes an "information flood" and finding important information can be as painful as finding a needle in a haystack.The influx of information has left decision makers in a 'sea of information' and in desperate need of a technology to help them make quick decisions and realise their individual needs.Personalised recommendation technology has emerged to fill this gap,as one of the key applications of artificial intelligence,by analysing the user's historical behaviour in various aspects and improving the efficiency of the user's access to effective information,freeing people from the clutter of big data.Although personalised recommendation technology has developed rapidly and achieved a series of important results,along with the increasing scale of Internet users and the continuous growth of various types of information such as market products,problems such as sparse data,uneven data quality and the difficulty of accurately portraying users' multiple interests are still prominent.The article focuses on the above problems and investigates how to further improve the accuracy of recommendation models,combining theory with practice and exploring its practical application value.In this paper,we focus on improving the accuracy of similarity calculation,and conduct an in-depth study on data quality,user rating behaviour and user multi-interest portrayal.The article proposes a multi-interest preference model based on a user rating penalty factor by taking the user's multi-interest preference as a guide to portray the user's multiple interest tendencies in a holistic manner and conducting research on the data quality problem and,to a certain extent,penalising low-quality users.The main research components are as follows.1)Digging deeper into the valid information in the user-item behaviour matrix,extracting items for which users have had positive feedback tendencies,completing the construction of the user multi-interest preference behaviour matrix,and further calculating the item attribute preference similarity,so as to achieve a more detailed portrayal of the consistent behaviour of users with multi-attribute tendencies.2)Through a comprehensive analysis of the shortcomings of the Jaccard similarity,certain improvements are made to the Jaccard similarity based on its shortcomings by fusing user interest propensity consistent behavioural characteristics and user rating penalty factors.The improved algorithm takes into account the impact of the complexity of user rating behaviour on recommendation accuracy and the inaccuracy of user rating preference measurement.3)Considering the impact of inconsistency in the degree of interest overlap between users and the degree of influence of certain attributes of items on the differentiation of users' interest points on the accuracy of multi-attribute preference similarity when portraying users' multi-attribute preferences,this paper proposes to improve the multi-attribute preference similarity algorithm by incorporating preference correction coefficients of attribute influence factors.This article focuses on validating the recommendation model proposed in this paper using the MAE review metric as a guideline.The experimental object chosen for the article is the Movie-Lens 100 k public dataset,which contains 100,000 reviews of 1682 movies by 943 users.By designing different experiments,the validity and stability of the recommendation model proposed in this paper are verified,and the problems of blurring user interest tendencies and imprecise portrayal of users' multiple interest preferences caused by the poor quality of user ratings are effectively solved. |
PDF Full Text Request