Research On Item Collaborative Filtering Algorithm Based On Long Tail Data

The change from information age to data age has brought about information overload and information asymmetry.Redundant information increases the difficulty for users to find requirements and obtain interested items,and the difficulty of obtaining effective information directly affects users' decision-making.The recommendation system came into being to satisfy the individual needs of users by means of prediction and recommendation.However,the existing recommendation algorithms presented similar recommendation results,tending to popular items and ignoring unpopular items.The effective recommendation of unpopular items has important research significance for avoiding the homogenization of user interests,tapping the potential needs of users,and opening up new markets for retailers.Regarding the long tail phenomenon of the recommendation system,this paper proposes improvement measures based on the research of the traditional Item-based CF algorithm.In order to solve the problem that the recommendation algorithm is not capable of recommending long-tail articles,three levels of similarity calculation,recommendation methods and evaluation indicators are improved.It aims to improve the prediction accuracy of the recommendation system and the ability to mine unpopular items.Then,the final recommendation list has a certain coverage of long-tail items,the recommendation algorithm improves the recommendation rate of unpopular items under the long-tail theory.In this paper,firstly,the calculation of item similarity is corrected by integrating the user's interest degree and the item's own characteristic attributes.The time weight function is used to process the user's historical score to alleviate the user interest transfer error caused by time and improve the accuracy of similarity calculation.Secondly,the sliding recommendation algorithm eliminates popular items with higher prediction scores,so that items with large differences appear in the user's personalized list.In this way,the recommended coverage of unpopular items is expanded,and the recommendation algorithm's ability to mine unpopular items is improved.Finally,in the calculation of coverage,similar item sets are added to optimize the calculation method of recommended coverage.When mining or recommending the same number of items,the recommendation ability of the algorithm can be accurately distinguished.In this paper,prediction accuracy and recommendation coverage are used as indicators to test the performance of the optimized recommendation algorithm.Experiments prove the SSFItem-CF recommendation algorithm which integrates multiple optimization methods is better than the traditional Item-CF recommendation algorithm in terms of prediction accuracy and recommendation coverage.In addition,the improved coverage rate can also achieve the evaluation effect of the original coverage rate,making a more effective distinction between algorithm performance.