【摘要】：隨著互聯(lián)網(wǎng)的不斷發(fā)展,以產(chǎn)品為中心尋找潛在客戶已經(jīng)成為眾多領(lǐng)域的重要應(yīng)用之一,特別是Web 2.0時代的到來。本文旨在研究以商戶為中心的潛在客戶定向問題,提出了一個產(chǎn)品-客戶的匹配框架。一般來說,要完成用戶和產(chǎn)品之間的匹配主要包括兩個模塊,第一個模塊是構(gòu)造用戶的興趣偏好,第二個模塊是執(zhí)行以產(chǎn)品為驅(qū)動的查詢,也稱逆向排名查詢。因此,本文的框架主要包括了兩個階段：數(shù)據(jù)預(yù)處理和查詢處理。數(shù)據(jù)預(yù)處理階段設(shè)計了簡單的規(guī)則算法來學(xué)習(xí)用戶的興趣偏好。查詢處理階段,使用逆向k排名查詢和逆向Top-k-Ranks查詢兩種查詢來為給定的產(chǎn)品找到匹配的用戶。本文的主要貢獻如下：·基于規(guī)則方法的用戶興趣偏好學(xué)習(xí).隨著Web2.0應(yīng)用的發(fā)展,越來越多的用戶參與其中對消費過的產(chǎn)品進行打分評論。本文的研究場景主要是針對大眾點評、美團、Yelp和世紀佳緣等生活類平臺,由于其中的評論數(shù)據(jù)由于文本短、數(shù)據(jù)稀疏,要利用評論數(shù)據(jù)構(gòu)造偏好無疑具有很大的挑戰(zhàn)。幸運的是,用戶對產(chǎn)品各個屬性的打分可以直觀地提煉出用戶對產(chǎn)品的喜好程度。因此,本文主要利用打分信息來構(gòu)造用戶的興趣偏好,并在原有線性模型表示用戶喜好程度的基礎(chǔ)之上提出了擴展的線性模型。·逆向k排名查詢.本文提出了以產(chǎn)品為驅(qū)動的逆向k排名查詢。這個查詢是面向所有產(chǎn)品集合,使用的是線性模型,以每個用戶在給定產(chǎn)品下的排名作為排序的基準,為每一個產(chǎn)品找到排序靠前的k個用戶。本文在多維場景下為這個查詢提出了三種算法,包括基于樹的剪枝方法(Tree-based Pruning Approach, TPA),批量剪枝算法(Batch Pruning Approach, BPA)和標記的剪枝算法(Marked Pruning Approach, MPA).考慮到二維場景下平面幾何的特殊性質(zhì),本文設(shè)計了兩種算法,包括基于排序的算法(Sorting-based approach, SA)和基于樹的算法(Tree-based Approach, TBA).最后,論文分別使用人工數(shù)據(jù)集和真實數(shù)據(jù)集做了大量的實驗來對算法的有效性進行了驗證。·逆向Top-k-Ranks查詢.本文提出了面向產(chǎn)品的逆向Top-k-Ranks查詢。這個查詢整合了逆向top-k查詢和逆向k排名查詢,使用的是擴展的線性模型,通過計算用戶和產(chǎn)品的得分值得到每個用戶相對于查詢產(chǎn)品的排名,依據(jù)排名值,返回最喜歡這個查詢產(chǎn)品的k個用戶以及把這個產(chǎn)品列入top k集合的所有用戶。本文為這個查詢設(shè)計了兩種算法,包括擴展的RTA方法(Extended RTA, ERTA)和基于歷史信息的批量剪枝算法(History-Based Batch Pruning Approach, HBPA).最后,論文分別使用人工數(shù)據(jù)集和真實數(shù)據(jù)集做了大量的實驗來對算法的有效性進行了驗證。綜上所述,本文針對面向產(chǎn)品的客戶定向機制重點研究了如何構(gòu)建用戶興趣偏好向量、面向線性模型的逆向k排名查詢和面向擴展模型的逆向Top-k-Ranks查詢?nèi)齻�問題,提出了解決這個問題的通用匹配框架。本文對提出的模型及算法在理論分析的基礎(chǔ)之上,分別在真實數(shù)據(jù)集和人工數(shù)據(jù)集上進行了驗證,結(jié)果表明本文提出的解決方法有好的效果。
[Abstract]:With the development of the Internet, it has become one of the most important applications in many fields, especially the arrival of the Web 2.0 era. The purpose of this paper is to study the potential customer orientation problem with the merchant as the center, and put forward a product-customer matching framework. In general, the matching between the user and the product mainly includes two modules, the first module is the user's interest preference, and the second module is to execute the query driven by the product, also called the reverse ranking query. Therefore, the framework of this paper mainly includes two stages: data pre-processing and query processing. A simple rule algorithm is designed in the data pre-processing stage to study the user's interest preference. The query processing stage uses the reverse-k ranking query and the reverse Top-k-Ranks to query the two queries to find a matching user for a given product. The main contributions of this paper are as follows: 路 User interest preference learning based on the rule method. With the development of the Web 2.0 application, more and more users are involved in scoring the consumer products. The research scene of this paper is mainly focused on the public comment, the American League, the Yelp and the Jiayuan Jiayuan, because of the short text and the sparse data, it is undoubtedly a great challenge to make use of the preference of the comment data. Fortunately, the user's scoring of the individual attributes of the product can intuitively refine the user's degree of preference for the product. Therefore, this paper mainly uses the scoring information to construct the user's interest preference, and puts forward the extended linear model on the basis of the original linear model representing the user's preference degree. 路 Reverse-k ranking query. In this paper, a reverse-k ranking query based on product is presented. This query is for all product collections, using a linear model for each user to find the top-ranked k users for each product based on the rank of the given product as a sort reference. In this paper, three algorithms are proposed for this query in a multi-dimensional scene, including tree-based Pruning (TPA), batch pruning (BPA) and marked pruning (MPA). In view of the special properties of the plane geometry in the two-dimensional scene, two algorithms are designed, including the sort-based algorithm (SA) and the tree-based algorithm (TBA). Finally, the paper makes a lot of experiments to verify the validity of the algorithm by using the artificial data set and the real data set. 路閫嗗悜Top-k-Ranks鏌ヨ. In this paper, a reverse Top-k-Ranks query for products is presented. the query integrates a reverse top-k query and a reverse k ranking query, using an extended linear model, Returns the k users who most like this query product and all users who have this product in the top k set. In this paper, two algorithms, including extended RTA (RTA) and historical-based batch pruning (HBPA), are designed for this query. Finally, the paper makes a lot of experiments to verify the validity of the algorithm by using the artificial data set and the real data set. To sum up, this paper focuses on how to build a user interest preference vector, a reverse-k ranking query for a linear model and a reverse Top-k-Ranks query for an extended model, and proposes a universal matching framework to solve this problem. On the basis of the theoretical analysis of the proposed model and algorithm, this paper makes a verification on the real data set and the artificial data set, and the results show that the solution proposed in this paper has good effect.
【學(xué)位授予單位】：華東師范大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2015
【分類號】：TP311.13;TP393.09

【參考文獻】

相關(guān)期刊論文 前4條

1 周傲英;金澈清;王國仁;李建中;;不確定性數(shù)據(jù)管理技術(shù)研究綜述[J];計算機學(xué)報;2009年01期

2 周傲英;周敏奇;宮學(xué)慶;;計算廣告:以數(shù)據(jù)為核心的Web綜合應(yīng)用[J];計算機學(xué)報;2011年10期

3 曾春,邢春曉,周立柱;個性化服務(wù)技術(shù)綜述[J];軟件學(xué)報;2002年10期

4 張金增;孟小峰;;移動Web搜索研究[J];軟件學(xué)報;2012年01期

，

本文編號：2434971