發(fā)布時(shí)間：2018-02-27 00:32

  本文關(guān)鍵詞： 網(wǎng)絡(luò)購(gòu)物 相似離度 加權(quán) RA 鏈路預(yù)測(cè) 共同趨向性　出處：《首都經(jīng)濟(jì)貿(mào)易大學(xué)》2015年碩士論文　論文類型：學(xué)位論文

【摘要】：電子商務(wù)的迅猛發(fā)展,讓網(wǎng)絡(luò)購(gòu)物成為一種趨勢(shì),但身處大數(shù)據(jù)時(shí)代,面對(duì)海量的商品信息,很容易讓顧客的購(gòu)物興趣減弱。推薦系統(tǒng)的出現(xiàn)在一定程度上解決了這個(gè)問(wèn)題,而推薦算法在推薦系統(tǒng)中扮演著重要的角色。本文運(yùn)用集合相似度理論、復(fù)雜網(wǎng)絡(luò)的相關(guān)理論及鏈路預(yù)測(cè)的知識(shí),從顧客的角度出發(fā),利用歷史購(gòu)買記錄對(duì)顧客關(guān)系進(jìn)行分析,運(yùn)用加權(quán)RA消除非常規(guī)顧客對(duì)相似性計(jì)算的影響。通過(guò)進(jìn)行鏈路預(yù)測(cè)的方法,構(gòu)建顧客共同趨向性獲取算法,進(jìn)而基于相關(guān)顧客購(gòu)物共同趨向性得到目標(biāo)顧客最有可能購(gòu)買的商品。通過(guò)將所建模型的求解結(jié)果與實(shí)驗(yàn)驗(yàn)證結(jié)果進(jìn)行對(duì)比分析,得出本文算法的可行性。對(duì)于模型的構(gòu)建主要運(yùn)用以下方法:(1)集合運(yùn)算。通過(guò)讓大學(xué)生作為顧客進(jìn)行商品購(gòu)買,從而得到顧客 商品集合,集合運(yùn)算得到顧客 商品關(guān)系矩陣。(2)相似性算法。針對(duì)顧客是否為孤立點(diǎn)這兩種情況,綜合運(yùn)用余弦相似系數(shù)和相對(duì)歐式距離系數(shù)進(jìn)行顧客相似離度的求解,既考慮了樣本內(nèi)數(shù)據(jù)變化規(guī)律的差異也考慮了樣本數(shù)據(jù)的數(shù)值差異。為了消除主流顧客對(duì)顧客相似性計(jì)算的影響,在此基礎(chǔ)上將相似離度值作為權(quán)重進(jìn)行加權(quán)RA的計(jì)算,運(yùn)用pajek構(gòu)建顧客相似關(guān)系網(wǎng)絡(luò)。(3)共同趨向性獲取算法。針對(duì)最相似顧客數(shù)量的不同分別進(jìn)行推薦,基于相關(guān)顧客購(gòu)物共同趨向性得到目標(biāo)顧客最有可能購(gòu)買的商品。
[Abstract]:With the rapid development of electronic commerce, online shopping has become a trend, but in the era of big data, it is easy to weaken the customer's interest in shopping in the face of massive commodity information. The appearance of recommendation system solves this problem to a certain extent. The recommendation algorithm plays an important role in the recommendation system. Using the theory of set similarity, the related theory of complex network and the knowledge of link prediction, from the customer's point of view, using the historical purchase record to analyze the customer relationship, this paper analyzes the relationship of customer by using the theory of set similarity, the theory of complex network and the knowledge of link prediction. The weighted RA is used to eliminate the influence of unconventional customers on similarity calculation. Through the method of link prediction, a customer common trend acquisition algorithm is constructed. Then, based on the common tendency of relevant customers, we get the most likely items to be purchased by the target customers. The results of the model are compared with the experimental results. The feasibility of this algorithm is obtained. The following method is mainly used to construct the model: 1) set operation. By making college students buy goods as customers, we can get the set of customers. The similarity algorithm of customer's merchandise relation matrix is obtained by set operation. In view of whether the customer is an outlier or not, the similarity degree of customer is solved by using cosine similarity coefficient and relative Euclidean distance coefficient synthetically. In order to eliminate the influence of mainstream customers on customer similarity calculation, the similarity deviation value is used as the weight to calculate the weighted RA in order to eliminate the influence of mainstream customers on customer similarity calculation. Using pajek to construct customer similarity relationship network. (3) Common trend acquisition algorithm. According to the different number of most similar customers, we recommend the most likely products for the target customers based on the common trend of customer shopping.
【學(xué)位授予單位】：首都經(jīng)濟(jì)貿(mào)易大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類號(hào)】：TP391.3;F724.6

【參考文獻(xiàn)】

相關(guān)博士學(xué)位論文 前1條

1 任磊;推薦系統(tǒng)關(guān)鍵技術(shù)研究[D];華東師范大學(xué);2012年

，

本文編號(hào)：1540380