本文選題：社交網(wǎng)絡(luò) + 大數(shù)據(jù)��； 參考：《重慶郵電大學(xué)》2016年碩士論文

【摘要】：隨著社交網(wǎng)絡(luò)的飛速發(fā)展,互聯(lián)網(wǎng)用戶(hù)所面臨的信息過(guò)載問(wèn)題尤為嚴(yán)重,因此國(guó)內(nèi)外各大社交網(wǎng)絡(luò)、電子商務(wù)等平臺(tái)都相繼推出了個(gè)性化推薦系統(tǒng),以緩解海量數(shù)據(jù)帶來(lái)的“選擇困難癥”問(wèn)題。隨著推薦系統(tǒng)用戶(hù)關(guān)系愈加復(fù)雜化,傳統(tǒng)的推薦算法已經(jīng)不能滿(mǎn)足當(dāng)前多數(shù)平臺(tái)的推薦需求。因此,研究大數(shù)據(jù)時(shí)代背景下的推薦策略,在緩解數(shù)據(jù)壓力的同時(shí)提高推薦系統(tǒng)的綜合質(zhì)量具有重要的意義。論文的主要研究?jī)?nèi)容和應(yīng)用價(jià)值如下:通過(guò)調(diào)研推薦技術(shù)現(xiàn)狀,在研究各并行平臺(tái)實(shí)現(xiàn)機(jī)理基礎(chǔ)之上,分析了當(dāng)前大數(shù)據(jù)推薦算法,提出了面向大數(shù)據(jù)的推薦系統(tǒng)雙引擎。此外,根據(jù)具體算法的特點(diǎn),例如是否迭代、算法復(fù)雜度等,通過(guò)調(diào)節(jié)數(shù)據(jù)規(guī)模等因素進(jìn)行對(duì)比實(shí)驗(yàn),分析不同框架下特定算法的性能,設(shè)計(jì)了特定場(chǎng)景下最適配的面向大數(shù)據(jù)的雙擎推薦系統(tǒng)框架。雙引擎則作為組件按需裝配到該框架中,根據(jù)需求動(dòng)態(tài)提供單機(jī)或者分布式推薦服務(wù),通過(guò)對(duì)比和調(diào)用分析案例驗(yàn)證了該框架的有效性和實(shí)用性�；谏鲜雒嫦虼髷�(shù)據(jù)的雙擎推薦系統(tǒng)框架,深入研究協(xié)同過(guò)濾推薦算法存在的不足之處。針對(duì)用戶(hù)相似度計(jì)算方法單一的問(wèn)題,分析了社交網(wǎng)絡(luò)數(shù)據(jù)的可用性,引入社交網(wǎng)絡(luò)用戶(hù)關(guān)系作為計(jì)算依據(jù),通過(guò)構(gòu)建新的相似度計(jì)算規(guī)則,融合社交網(wǎng)絡(luò)與推薦系統(tǒng)的關(guān)鍵要素,提出了融合社交網(wǎng)絡(luò)多屬性的協(xié)同過(guò)濾推薦算法,為實(shí)施推薦的目標(biāo)用戶(hù)搜索到了較為準(zhǔn)確的鄰居集合,進(jìn)而提高了推薦的準(zhǔn)確度。實(shí)驗(yàn)結(jié)果表明,該推薦算法有效地利用了社交網(wǎng)絡(luò)元素,在相似度計(jì)算方法上較傳統(tǒng)方法更為準(zhǔn)確和個(gè)性化。在算法應(yīng)用研究中實(shí)現(xiàn)了個(gè)性化推薦,豐富了推薦系統(tǒng)的結(jié)果解釋,提高了推薦系統(tǒng)的綜合質(zhì)量。綜上所述,本文對(duì)融合社交網(wǎng)絡(luò)的協(xié)同過(guò)濾推薦算法的研究與應(yīng)用展開(kāi)了積極的探索和深入的研究,設(shè)計(jì)了一種面向大數(shù)據(jù)的雙擎推薦系統(tǒng)框架,提出了融合社交網(wǎng)絡(luò)多屬性的協(xié)同過(guò)濾推薦算法,并且設(shè)計(jì)和開(kāi)發(fā)了一個(gè)社交網(wǎng)絡(luò)推薦系統(tǒng),豐富了基于社交網(wǎng)絡(luò)的協(xié)同過(guò)濾推薦算法的理論研究。
[Abstract]:With the rapid development of social network, the problem of information overload faced by Internet users is particularly serious. Therefore, various platforms such as domestic and foreign social networks, e-commerce and other platforms have launched personalized recommendation system.To alleviate the mass of data caused by the "difficult to choose" problem.With the complexity of user relationship in recommendation systems, the traditional recommendation algorithms can not meet the current recommendation needs of most platforms.Therefore, it is of great significance to study the recommendation strategy under the background of big data, which can relieve the pressure of data and improve the comprehensive quality of recommendation system.The main contents and application value of this paper are as follows: based on the research of the status quo of recommendation technology and on the basis of studying the realization mechanism of each parallel platform, this paper analyzes the current big data recommendation algorithm, and puts forward the double engine of the recommended system for big data.In addition, according to the characteristics of the specific algorithm, such as iterative or not, the complexity of the algorithm, by adjusting the data scale and other factors to carry out comparative experiments, the performance of the specific algorithm under different frameworks is analyzed.This paper designs the most suitable double-platform recommendation system framework for big data in a specific scenario.The dual engine is assembled into the framework as a component on demand and provides a single machine or distributed recommendation service dynamically according to the requirements. The effectiveness and practicability of the framework are verified by comparison and call analysis cases.Based on the framework of big data-oriented recommendation system, the shortcomings of collaborative filtering recommendation algorithm are studied.In order to solve the problem of simple similarity calculation method, this paper analyzes the usability of social network data, introduces the user relationship of social network as the basis of calculation, and constructs a new similarity calculation rule.Combining the key elements of social network and recommendation system, a collaborative filtering recommendation algorithm based on multiple attributes of social network is proposed, which can search the accurate neighbor set for the target user of recommendation, and then improve the accuracy of recommendation.Experimental results show that the proposed recommendation algorithm is more accurate and individualized than the traditional method in computing similarity by using the elements of social network effectively.The personalized recommendation is realized in the research of algorithm application, which enriches the interpretation of the recommendation system and improves the comprehensive quality of the recommendation system.To sum up, this paper has carried out active exploration and deep research on the research and application of collaborative filtering recommendation algorithm based on social network, and designed a framework of dual-level recommendation system for big data.A collaborative filtering recommendation algorithm based on multiple attributes of social network is proposed, and a recommendation system of social network is designed and developed, which enriches the theoretical research of collaborative filtering recommendation algorithm based on social network.
【學(xué)位授予單位】：重慶郵電大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2016
【分類(lèi)號(hào)】：TP391.3

【參考文獻(xiàn)】

相關(guān)期刊論文 前4條

1 LIU Qingwen;XIONG Yan;HUANG Wenchao;;Combining User-Based and Item-Based Models for Collaborative Filtering Using Stacked Regression[J];Chinese Journal of Electronics;2014年04期

2 王s，

本文編號(hào)：1765032