基于用戶聚類的協(xié)同過濾推薦算法研究
發(fā)布時間：2019-03-07 22:18
【摘要】：隨著互聯(lián)網(wǎng)的普及,大量無意義的數(shù)據(jù)給人們篩選有效信息帶來巨大的困難。為了幫助人們快速有效的篩選信息,個性化推薦系統(tǒng)應運而生。推薦算法作為推薦系統(tǒng)的核心,一直是研究的重點。在眾多的推薦算法中,協(xié)同過濾算法是應用最廣泛的。協(xié)同過濾算法通過對用戶歷史行為數(shù)據(jù)的挖掘發(fā)現(xiàn)用戶的偏好,從而基于不同的偏好對用戶進行群組劃分并推薦品味相似的物品。然而,隨著電子商務系統(tǒng)中用戶數(shù)和項目數(shù)的不斷增大,數(shù)據(jù)的稀疏性和推薦效率逐漸成為制約協(xié)同過濾算法發(fā)展的瓶頸。為了提高協(xié)同過濾算法的推薦質(zhì)量和推薦效率,本文提出一種基于改進的用戶聚類協(xié)同過濾推薦算法,并基于改進算法設計和實現(xiàn)了一個B/S架構(gòu)的電影推薦系統(tǒng)。本文介紹了個性化推薦系統(tǒng)的發(fā)展背景和架構(gòu)設計,給出了傳統(tǒng)協(xié)同過濾算法的基本思想和面臨的主要問題,從而從離線用戶聚類和用戶相似度計算兩個方面改進了傳統(tǒng)算法。對算法的改進主要體現(xiàn)在兩個方面:一是綜合考慮了用戶評分信息和項目類別偏好信息對用戶聚類的影響,提出一種聯(lián)合用戶聚類算法。該算法分別基于用戶評分信息和項目類別偏好信息對基本用戶聚類,產(chǎn)生兩個聚類中心和兩個用戶類別所屬矩陣,計算目標用戶與兩個聚類中心的相似度以及目標用戶在不同聚類中所屬的類簇,對結(jié)果合并去重后得到目標用戶的最近鄰居搜索空間。二是針對傳統(tǒng)Pearson相關系數(shù)計算相似度時對絕對數(shù)值不敏感等問題,提出一種基于差異因子的加權(quán)Pearson相關系數(shù)計算方法,將評分差異因子作為權(quán)重來修正傳統(tǒng)的Pearson相關系數(shù)。采用MovieLens數(shù)據(jù)集,以MAE值、準確率、召回率和F1值為度量標準,通過多組實驗對改進算法、傳統(tǒng)基于用戶的協(xié)同過濾算法(CF)、傳統(tǒng)基于用戶聚類的協(xié)同過濾算法(UCCF)進行評估,實驗結(jié)果表明改進算法能夠有效提高推薦系統(tǒng)的推薦效率和推薦精度。本文基于改進算法設計并實現(xiàn)了電影推薦系統(tǒng),系統(tǒng)采用豆瓣Top250電影信息作為數(shù)據(jù)集,使用PHP和Matlab混合編程實現(xiàn),能夠根據(jù)用戶的偏好信息為用戶提供個性化的推薦服務。
[Abstract]:With the popularity of the Internet, a large number of meaningless data for people to screen effective information has brought great difficulties. In order to help people to screen information quickly and effectively, personalized recommendation system emerges as the times require. As the core of recommendation system, recommendation algorithm has always been the focus of research. Among the many recommendation algorithms, collaborative filtering algorithm is the most widely used. The collaborative filtering algorithm discovers the user's preference by mining the user's historical behavior data, and then groups the users based on different preferences and recommends the items with similar taste. However, with the increasing number of users and items in e-commerce system, the sparsity of data and the efficiency of recommendation gradually become the bottleneck to restrict the development of collaborative filtering algorithm. In order to improve the recommendation quality and efficiency of collaborative filtering algorithm, this paper proposes an improved collaborative filtering recommendation algorithm based on user clustering, and designs and implements a movie recommendation system based on the improved algorithm. This paper introduces the development background and architecture design of personalized recommendation system, gives the basic idea and main problems of traditional collaborative filtering algorithm, and improves the traditional algorithm from two aspects: off-line user clustering and user similarity calculation. The improvement of the algorithm is mainly reflected in two aspects: first, considering the influence of user rating information and item class preference information on user clustering, a joint user clustering algorithm is proposed. Based on the user rating information and item class preference information, the algorithm clusters the basic users respectively, and generates two clustering centers and two user categories belong to the matrix. The similarity between the target user and the two clustering centers and the cluster belonging to the target user in different clusters are calculated. The nearest neighbor search space of the target user is obtained after the result is merged and deduplicated. Secondly, aiming at the problem that the traditional Pearson correlation coefficient is insensitive to absolute value when calculating the similarity degree, a weighted Pearson correlation coefficient calculation method based on the difference factor is proposed, in which the score difference factor is used as the weight to correct the traditional Pearson correlation coefficient. Using MovieLens data set and mae value, accuracy rate, recall rate and F1 value as metrics, the improved algorithm is improved by multi-group experiments, and the traditional user-based collaborative filtering algorithm (CF), is used. The traditional collaborative filtering algorithm based on user clustering (UCCF) is evaluated. The experimental results show that the improved algorithm can effectively improve the recommendation efficiency and accuracy of the recommendation system. In this paper, a movie recommendation system is designed and implemented based on the improved algorithm. The system uses Douban Top250 movie information as data set and mixed programming with PHP and Matlab, which can provide personalized recommendation service to users according to their preference information.
【學位授予單位】：北京交通大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.3
本文編號：2436505
[Abstract]:With the popularity of the Internet, a large number of meaningless data for people to screen effective information has brought great difficulties. In order to help people to screen information quickly and effectively, personalized recommendation system emerges as the times require. As the core of recommendation system, recommendation algorithm has always been the focus of research. Among the many recommendation algorithms, collaborative filtering algorithm is the most widely used. The collaborative filtering algorithm discovers the user's preference by mining the user's historical behavior data, and then groups the users based on different preferences and recommends the items with similar taste. However, with the increasing number of users and items in e-commerce system, the sparsity of data and the efficiency of recommendation gradually become the bottleneck to restrict the development of collaborative filtering algorithm. In order to improve the recommendation quality and efficiency of collaborative filtering algorithm, this paper proposes an improved collaborative filtering recommendation algorithm based on user clustering, and designs and implements a movie recommendation system based on the improved algorithm. This paper introduces the development background and architecture design of personalized recommendation system, gives the basic idea and main problems of traditional collaborative filtering algorithm, and improves the traditional algorithm from two aspects: off-line user clustering and user similarity calculation. The improvement of the algorithm is mainly reflected in two aspects: first, considering the influence of user rating information and item class preference information on user clustering, a joint user clustering algorithm is proposed. Based on the user rating information and item class preference information, the algorithm clusters the basic users respectively, and generates two clustering centers and two user categories belong to the matrix. The similarity between the target user and the two clustering centers and the cluster belonging to the target user in different clusters are calculated. The nearest neighbor search space of the target user is obtained after the result is merged and deduplicated. Secondly, aiming at the problem that the traditional Pearson correlation coefficient is insensitive to absolute value when calculating the similarity degree, a weighted Pearson correlation coefficient calculation method based on the difference factor is proposed, in which the score difference factor is used as the weight to correct the traditional Pearson correlation coefficient. Using MovieLens data set and mae value, accuracy rate, recall rate and F1 value as metrics, the improved algorithm is improved by multi-group experiments, and the traditional user-based collaborative filtering algorithm (CF), is used. The traditional collaborative filtering algorithm based on user clustering (UCCF) is evaluated. The experimental results show that the improved algorithm can effectively improve the recommendation efficiency and accuracy of the recommendation system. In this paper, a movie recommendation system is designed and implemented based on the improved algorithm. The system uses Douban Top250 movie information as data set and mixed programming with PHP and Matlab, which can provide personalized recommendation service to users according to their preference information.
【學位授予單位】：北京交通大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.3
【參考文獻】
中國期刊全文數(shù)據(jù)庫 前10條
1 王宇飛;宋俊典;戴炳榮;;基于用戶評分和項目類偏好的協(xié)同過濾推薦算法[J];軟件導刊;2016年12期
2 陳功平;王紅;;改進Pearson相關系數(shù)的個性化推薦算法[J];山東農(nóng)業(yè)大學學報(自然科學版);2016年06期
3 張栩晨;;利用Tri-training算法解決推薦系統(tǒng)冷啟動問題[J];計算機科學;2016年12期
4 魏慧娟;戴牡紅;寧勇余;;基于最近鄰居聚類的協(xié)同過濾推薦算法[J];中國科學技術大學學報;2016年09期
5 王興茂;張興明;吳毅濤;潘俊池;;基于啟發(fā)式聚類模型和類別相似度的協(xié)同過濾推薦算法[J];電子學報;2016年07期
6 黃濤;黃仁;張坤;;一種改進的協(xié)同過濾推薦算法[J];計算機科學;2016年S1期
7 邱爽;葛萬成;汪亮友;林佳燕;;個性化推薦中基于用戶協(xié)同過濾算法的優(yōu)化[J];信息技術;2016年03期
8 趙宏晨;翟麗麗;張樹臣;;基于灰色關聯(lián)度聚類與標簽重疊因子結(jié)合的協(xié)同過濾推薦方法研究[J];計算機工程與科學;2016年01期
9 原福永;馬琳;梁順攀;;融合用戶相似度和信任傳播重組信任矩陣算法[J];燕山大學學報;2015年06期
10 李艷萍;劉明;于麗梅;;個性化信息服務網(wǎng)絡系統(tǒng)架構(gòu)研究[J];數(shù)字技術與應用;2015年09期
中國碩士學位論文全文數(shù)據(jù)庫 前1條
1 蒲彬;基于社交信號的個性化新聞推薦系統(tǒng)的設計與實現(xiàn)[D];中國科學院大學（工程管理與信息技術學院）;2015年
，本文編號：2436505
本文鏈接：http://sikaile.net/jingjilunwen/dianzishangwulunwen/2436505.html
最近更新
教材專著
