本文關(guān)鍵詞： 推薦系統(tǒng) GPU CUDA 精確度　出處：《北京郵電大學(xué)》2013年碩士論文　論文類型：學(xué)位論文

【摘要】：隨著網(wǎng)絡(luò)信息的爆炸式發(fā)展而導(dǎo)致信息過(guò)載和搜索引擎系統(tǒng)本身的被動(dòng)性搜索過(guò)程,推薦引擎系統(tǒng)受到了越來(lái)越多的關(guān)注和研究。推薦系統(tǒng)當(dāng)前主要的研究方向是冷啟動(dòng)問(wèn)題,矩陣稀疏問(wèn)題以及推薦多樣性問(wèn)題等等,總體上是針對(duì)推薦結(jié)果的優(yōu)劣進(jìn)行研究和改進(jìn)。但是由于推薦系統(tǒng)本身的巨大規(guī)模和矩陣稀疏性問(wèn)題共同影響導(dǎo)致預(yù)測(cè)推薦結(jié)果需要耗費(fèi)大量時(shí)間所帶來(lái)的推薦系統(tǒng)滯后性問(wèn)題和推薦結(jié)果精度低所帶來(lái)的非智能性問(wèn)題上的研究則相對(duì)較少。商業(yè)上的解決方案是將推薦系統(tǒng)分為線下的計(jì)算模塊和線上實(shí)時(shí)推薦模塊。線下模塊通過(guò)提前計(jì)算預(yù)測(cè)推薦結(jié)果并存放在數(shù)據(jù)庫(kù)供用戶使用系統(tǒng)時(shí)再進(jìn)行實(shí)時(shí)推薦,這樣的解決方案能夠使用戶得到相對(duì)實(shí)時(shí)的推薦服務(wù),但是這樣的處理方式仍然不能解決由系統(tǒng)龐大規(guī)模帶來(lái)的海量計(jì)算的巨大時(shí)間消耗,推薦結(jié)果仍然存在滯后性,用戶得到的推薦都是系統(tǒng)過(guò)去的推薦結(jié)果,并不能盡可能地根據(jù)用戶的行為實(shí)時(shí)反饋。 GPU原本是一種應(yīng)用于圖形圖像處理的多核處理器,它專門為可并行化計(jì)算密集型的任務(wù)而設(shè)計(jì)的處理器,擁有非常高的計(jì)算能力和非常大的數(shù)據(jù)吞吐量,同樣的任務(wù)GPU往往以絕對(duì)的效率優(yōu)勢(shì)超越CPU的運(yùn)行表現(xiàn)。 推薦系統(tǒng)主要的耗時(shí)部分是線下計(jì)算模塊,而線下模塊主要的耗時(shí)任務(wù)是相似度模塊的計(jì)算任務(wù)。相似度模塊是可以實(shí)現(xiàn)并行化處理的過(guò)程,因此該部分進(jìn)行并行設(shè)計(jì)并移植到GPU上實(shí)現(xiàn)。為了達(dá)到更好的時(shí)間和空間優(yōu)化,本文使用CSR數(shù)據(jù)格式方式組織,GPU上的線程使用基于行并行的稀疏矩陣乘法處理算法。另外一個(gè)方面,由于矩陣稀疏性問(wèn)題,本文提出了基于信息關(guān)聯(lián)傳遞的用戶相似度算法,用戶之間的相似度為他們之間的直接相似度再加上他們共同好友之間的傳遞相似度的規(guī)則來(lái)衡量。實(shí)驗(yàn)表明該實(shí)現(xiàn)方案能夠帶來(lái)10倍加速并且新算法能夠提高20%的精度。實(shí)驗(yàn)結(jié)果也顯示數(shù)據(jù)越大,加速比就越顯著。
[Abstract]:With the explosive development of network information, the overload of information and the passive search process of search engine system, the recommendation engine system has received more and more attention and research. The main research direction of recommendation system is cold start problem. Matrix sparsity problem, recommendation diversity problem, etc., Generally speaking, it is to study and improve the merits and demerits of recommendation results. However, due to the huge scale of recommendation system and the problem of matrix sparsity, it is necessary to predict the recommended results in a large amount of time. There is relatively little research on the problem of system lag and the problem of non-intelligence caused by the low accuracy of recommendation results. The commercial solution is to divide the recommendation system into offline computing module and on-line real-time recommendation module. The module calculates the prediction recommendation results in advance and makes real-time recommendation when stored in the database for users to use the system. Such a solution can enable users to obtain a relatively real-time recommendation service, but this processing method still can not solve the huge time consumption of massive computing brought by the huge scale of the system, and the recommended results are still lagging behind. The recommendations received by users are all the past recommendations of the system, and can not be feedback in real time according to the user's behavior as much as possible. GPU was originally a multi-core processor for graphics and image processing. It is specially designed for parallelizing computationally intensive tasks with very high computing power and very large data throughput. The same task GPU often outperforms CPU performance with absolute efficiency advantage. The main time-consuming part of the recommendation system is the offline computing module, and the main time-consuming task of the offline module is the computation task of the similarity module. Therefore, this part is designed in parallel and implemented on GPU. In order to achieve better time and space optimization, This paper uses CSR data format to organize threads on CSR using row parallel sparse matrix multiplication algorithm. On the other hand, due to the problem of matrix sparsity, this paper proposes a user similarity algorithm based on information association transfer. The similarity between users is measured by the rules of the direct similarity between them and the transfer similarity between their common friends. Experiments show that the proposed scheme can bring 10 times acceleration and the new algorithm can improve 20%. The experimental results also show that the larger the data, The speedup is more significant.
【學(xué)位授予單位】：北京郵電大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2013
【分類號(hào)】：TP391.3

【參考文獻(xiàn)】

相關(guān)期刊論文 前8條

1 李聰;梁昌勇;楊善林;;電子商務(wù)協(xié)同過(guò)濾稀疏性研究:一個(gè)分類視角[J];管理工程學(xué)報(bào);2011年01期

2 焦良葆;陳瑞;;GPU核函數(shù)細(xì)化研究[J];計(jì)算機(jī)工程;2010年18期

3 曹小鵬;;基于GPU并行計(jì)算及在模式識(shí)別中的研究[J];計(jì)算機(jī)與數(shù)字工程;2011年08期

4 田緒紅;江敏杰;;GPU加速的神經(jīng)網(wǎng)絡(luò)BP算法[J];計(jì)算機(jī)應(yīng)用研究;2009年05期

5 周勇;王皓;程春田;郭禾;;基于GPU的多數(shù)據(jù)流相關(guān)系數(shù)并行計(jì)算方法研究[J];計(jì)算機(jī)應(yīng)用研究;2010年04期

6 馬超;韋剛;裴頌文;吳百鋒;;GPU上稀疏矩陣與矢量乘積運(yùn)算的一種改進(jìn)[J];計(jì)算機(jī)系統(tǒng)應(yīng)用;2010年05期

7 趙宏霞;王新海;楊皎平;;基于項(xiàng)目因子分析的Web客戶需求協(xié)同推薦算法[J];計(jì)算機(jī)系統(tǒng)應(yīng)用;2011年07期

8 黎明;徐德智;;一種結(jié)合基于項(xiàng)目和用戶的個(gè)性化推薦算法[J];小型微型計(jì)算機(jī)系統(tǒng);2011年04期

相關(guān)碩士學(xué)位論文 前1條

1 顏瑞;基于CUDA的立體匹配及去隔行算法[D];浙江大學(xué);2010年

，

本文編號(hào)：1543139