【摘要】：隨著Web2.0的飛速發(fā)展,網(wǎng)絡(luò)中的信息也在以前所未有的規(guī)模增長(zhǎng)。尤其在電子商務(wù)領(lǐng)域,海量商品出現(xiàn)在各種電子商務(wù)網(wǎng)站上。普通用戶需要的是從大量的信息流中實(shí)時(shí)獲取到對(duì)自身有用的信息,讓生活更加簡(jiǎn)單;商家需要的是將自家的產(chǎn)品有效地推薦給適合的群體,使得商家獲得最大的利益。在這樣的背景下,推薦算法的研究,尤其是基于用戶消費(fèi)習(xí)慣的推薦算法研究就顯得十分重要。本文的主要工作有以下兩個(gè)方面:1)基于用戶消費(fèi)價(jià)格的評(píng)分預(yù)測(cè)算法研究。評(píng)分預(yù)測(cè)問(wèn)題是推薦系統(tǒng)研究中的一個(gè)熱點(diǎn)問(wèn)題,評(píng)分預(yù)測(cè)算法使用各種輔助信息來(lái)提升推薦效果。電子商務(wù)網(wǎng)站存在大量的用戶消費(fèi)價(jià)格信息,本文利用用戶的消費(fèi)價(jià)格信息來(lái)提升評(píng)分預(yù)測(cè)的效果。通過(guò)對(duì)于這類(lèi)數(shù)據(jù)的分析,本文發(fā)現(xiàn)了用戶評(píng)分與消費(fèi)價(jià)格信息之間的三條關(guān)聯(lián),同時(shí)本文將這三條關(guān)聯(lián)與矩陣分解算法相結(jié)合,提出了基于價(jià)格敏感的評(píng)分推薦算法。除此之外,本文通過(guò)增廣矩陣和價(jià)格離散化的方法很好地解決了數(shù)據(jù)稀疏性問(wèn)題以及價(jià)格信息的噪音問(wèn)題�；诖蟊婞c(diǎn)評(píng)的數(shù)據(jù)集,本文針對(duì)提出的基于價(jià)格敏感的評(píng)分推薦算法做了大量實(shí)驗(yàn)。通過(guò)與眾多經(jīng)典算法相比較,本文驗(yàn)證了基于價(jià)格敏感的評(píng)分推薦算法的有效性。同時(shí),通過(guò)研究不同參數(shù)的物理意義,深入地研究了評(píng)分與價(jià)格間的關(guān)系。2)基于用戶消費(fèi)行為的潛在重復(fù)購(gòu)買(mǎi)用戶推薦算法研究。很多商家都會(huì)通過(guò)在電子商務(wù)網(wǎng)站上打折的方式吸引新用戶。商家希望吸引到的新用戶可以轉(zhuǎn)化為重復(fù)購(gòu)買(mǎi)用戶,而大多數(shù)新用戶只是一次性購(gòu)買(mǎi)用戶。本文提出了一個(gè)多分類(lèi)模型融合的推薦算法,利用這個(gè)算法為商家推薦潛在的重復(fù)購(gòu)買(mǎi)用戶。在這個(gè)算法中,本文設(shè)計(jì)了一種二層級(jí)聯(lián)的多分類(lèi)模型融合框架。在第一層中,本文通過(guò)使用多種不同類(lèi)型的分類(lèi)器以及特征組合,產(chǎn)生了大量不同的分類(lèi)結(jié)果。在第二層中,本文使用邏輯斯特回歸的分類(lèi)方法將第一層的多種分類(lèi)結(jié)果相互融合,達(dá)到更好的推薦效果。通過(guò)在天貓商城真實(shí)數(shù)據(jù)集上的實(shí)驗(yàn),本文驗(yàn)證了算法的有效性。同時(shí),為了將算法部署到云計(jì)算平臺(tái),本文在特征提取階段,對(duì)算法做了相應(yīng)的并行化設(shè)計(jì)。
[Abstract]:With the rapid development of Web2.0, the information in the network is also growing on an unprecedented scale. Especially in the field of e-commerce, a large number of goods appear on a variety of e-commerce websites. What ordinary users need is to obtain useful information for themselves in real time from a large number of information flows, so that life is simpler; what businesses need is to effectively recommend their own products to the appropriate groups, so that businesses can get the greatest benefits. In this context, the research of recommendation algorithm, especially the recommendation algorithm based on user consumption habit, is very important. The main work of this paper is as follows: 1) Research on scoring prediction algorithm based on user consumer price. Scoring prediction problem is a hot issue in the research of recommendation system. Scoring prediction algorithm uses a variety of auxiliary information to improve the recommendation effect. There are a lot of consumer price information in e-commerce website. This paper uses the consumer price information of users to improve the effect of rating prediction. Through the analysis of this kind of data, this paper finds three associations between user score and consumer price information. At the same time, this paper combines these three associations with matrix decomposition algorithm, and proposes a rating recommendation algorithm based on price sensitivity. In addition, this paper solves the problem of data sparsity and the noise of price information by means of augmented matrix and price discretization. Based on Dianping's dataset, a large number of experiments are carried out on the proposed price-sensitive rating recommendation algorithm. Compared with many classical algorithms, this paper verifies the effectiveness of the price-sensitive rating recommendation algorithm. At the same time, by studying the physical meaning of different parameters, the relationship between score and price is deeply studied. 2) Research on potential repeat purchase user recommendation algorithm based on user consumption behavior. Many businesses attract new users by offering discounts on e-commerce sites. Businesses want to attract new users who can be converted to repeat buyers, while most new users buy users at once. In this paper, a recommendation algorithm based on multi-classification model fusion is proposed, which is used to recommend potential repeat purchase users for businesses. In this algorithm, a two-layer cascade multi-classification model fusion framework is designed. In the first layer, many different types of classifiers and feature combinations are used to produce a large number of different classification results. In the second layer, this paper uses the logical regression classification method to merge the classification results of the first layer with each other to achieve better recommendation results. The effectiveness of the algorithm is verified by experiments on the real data set of Tmall Mall. At the same time, in order to deploy the algorithm to cloud computing platform, this paper makes the corresponding parallelization design of the algorithm in the feature extraction stage.
【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】：2016
【分類(lèi)號(hào)】：TP391.3

【參考文獻(xiàn)】

相關(guān)期刊論文 前3條

1 高全力;高嶺;楊建鋒;王海;;上下文感知推薦系統(tǒng)中基于用戶認(rèn)知行為的偏好獲取方法[J];計(jì)算機(jī)學(xué)報(bào);2015年09期

2 劉樹(shù)棟;孟祥武;;基于位置的社會(huì)化網(wǎng)絡(luò)推薦系統(tǒng)[J];計(jì)算機(jī)學(xué)報(bào);2015年02期

3 劉紅巖,陳劍,陳國(guó)青;數(shù)據(jù)挖掘中的數(shù)據(jù)分類(lèi)算法綜述[J];清華大學(xué)學(xué)報(bào)(自然科學(xué)版);2002年06期

相關(guān)碩士學(xué)位論文 前2條

1 葛艷艷;基于社會(huì)標(biāo)簽系統(tǒng)的推薦技術(shù)研究[D];大連理工大學(xué);2011年

2 李新煜;旅游移動(dòng)商務(wù)中基于位置感知的推薦系統(tǒng)研究[D];中國(guó)科學(xué)技術(shù)大學(xué);2011年

，

本文編號(hào)：2492278