APP下载

基于在线评价信息和消费者期望的商品选择方法

2018-01-02尤天慧樊治平

中国管理科学 2017年11期
关键词:损益排序矩阵

尤天慧, 张 瑾, 樊治平

(东北大学工商管理学院,辽宁 沈阳 110167)

基于在线评价信息和消费者期望的商品选择方法

尤天慧, 张 瑾, 樊治平

(东北大学工商管理学院,辽宁 沈阳 110167)

针对近年来许多电商网站涌现大量的有关商品在线评价信息,如何基于在线评价信息并考虑消费者给出属性评价期望进行商品选择,是一个值得关注的研究问题。本文提出了一种基于在线评价信息和消费者期望的商品选择方法。在该方法中,首先通过计算商品各属性的评价值相对于消费者给出的属性评价期望的损益值,进而确定关于各属性的评价损益结果的概率分布;然后,运用前景随机占优准则构建两两备选商品比较的前景随机占优关系矩阵,并采用PROMETHEE-II方法得到备选商品的排序结果。最后,以汽车之家网站提供的在线评价信息进行汽车选择为例说明了本文提出方法的可行性和有效性。

在线评价信息;消费者期望;前景随机占优;PROMETHEE-II;商品选择

1 引言

近年来,随着互联网和电子商务的迅猛发展,许多电商网站提供了关于商品或服务的在线评价/评论功能,如亚马逊(https://www.amazon.cn)、汽车之家(http://www.autohome.com.cn)网站等。如何使用在线评价/评论信息为广大消费者购买决策提供服务,已经引起了许多学者的关注。例如,有些学者针对在线评价/评论的机制[1, 2]、在线评价/评论对购买行为的影响[3-6]、在线评价/评论的有效性[7-10]、基于在线评价或在线评论的顾客满意度测评[11-13]和基于在线评价或在线评论的商品推荐[14-16]等方面进行了研究。已有研究表明:在线评价和在线评论对于消费者了解商品或服务质量并做出购买决策具有重要的作用[3-6]。目前,如何使用在线评价和在线评论信息来支持消费者进行购买决策,这是一个值得关注的研究课题,基于此,一些学者进行了基于在线评价或在线评论的商品排序方法研究[17-21]。例如:Li Yongli等[17]针对基于在线评价信息的商品排序问题,给出了一种基于社会网络分析的在线商品排序方法,该方法较好地解决了在线评价信息不完全且群体评价不一致的商品排序问题;Yang Xian等[18]利用在线评论、在线评级、比较语句和比较投票数4种类型的在线评价/评论信息给出了一种新的商品电子口碑排序方法,并针对该方法进行了系统实现以帮助消费者进行商品比较与选择;Chen Kun等[19]基于TOPSIS思想给出一种使用在线评论信息的商品排序方法;Najmi等[20]给出了一种基于商品在线评论和商品说明书的商品排序方法;Liu Yang等[21]给出了一种基于在线评论情感分析和直觉模糊集的商品排序方法。从已有的研究成果可以看到,大多没有从消费者的角度来研究如何基于在线评价信息进行商品的选择,同时也很少考虑消费者关于购买商品的偏好或期望信息。在现实中,例如,消费者有意愿购买汽车时,其会查看电商网站(如汽车之家)关于汽车空间、动力、操控等属性的在线评价信息,另外,消费者可能会对备选汽车的各属性有期望要求,如对汽车的油耗属性期望在线评价的评价值不低于4分等。鉴于此,本文则是研究基于在线评价信息且消费者给出属性在线评价期望的商品选择方法。在该方法中,首先将消费者给出的属性期望水平视为参考点,计算各属性的在线评价值相对于参考点的损益值,进而确定各属性在线评价损益结果的概率分布及累积分布函数;然后依据前景随机占优准则判断并确定两两备选商品之间比较所具有的占优关系,进而构建相应的前景随机占优关系矩阵;进一步地,依据该矩阵,运用PROMETHEE-II方法可以得到备选商品的排序结果。

2 问题描述

本文考虑基于商品多属性在线评价信息和消费者给出属性在线评价期望的商品选择问题,为了方便起见,下面的符号用来表示该问题中所涉及的集和量:

(1)A={A1,A2,…,Am}:消费者关注的m个备选商品的集合,其中Ai表示第i个备选商品,i=1,2,…,m。

(2)C={C1,C2,…,Cn}:消费者所关注的商品的n个属性的集合,其中Cj表示第j个属性,j=1,2,…,n。

(4)T={T1,T2,…,Tv}:在线评价过程中针对商品属性所使用的评价标度集合,其中Tε表示第ε个评价标度,ε=1,2,…,v,通常,ε越大,所对应的评价等级越高。例如,在汽车之家网站中,关于汽车属性的在线评价所使用的标度集是5分制等级的形式,即T={T1=1,T2=2,T3=3,T4=4,T5=5},其中1分表示最差,5分表示最好。

(5)Ki:参与商品Ai在线评价的用户数,i=1,2,…,m。

本文要解决的问题是:针对消费者关注的若干备选商品及相关属性,并依据相关电商网站提供的备选商品在线评价信息、消费者给出的属性期望和属性权重向量,如何使用一个决策分析方法对备选商品进行排序或者选择适合的商品。

3 原理与方法

针对上述商品选择问题,由于参与在线评价的众多用户依据评价标度集给出商品针对属性的评价值往往具有多种可能结果,所以商品针对属性的评价结果会呈现离散概率分布的形式,显然,评价结果可被视为一个离散型随机变量。为了解决上节提及的商品选择问题,本文给出一种基于在线评价信息和消费者期望的商品选择方法。该方法包括3个部分:计算评价损益结果及其概率分布、构建两两备选商品比较的前景随机占优关系矩阵和选择适合的商品。下面分别描述这3个部分的具体计算过程,进一步地给出商品选择方法的计算步骤。

1) 计算商品评价损益结果及其概率分布

(1)

其中,

(2)

(3)

(4)

依据式(4)可得到商品Ai针对属性Cj的评价损益结果yij的概率分布函数P(yij)为:

(5)

进一步地,由式(5)可得到商品Ai针对属性Cj的评价损益结果yij的累积分布函数Fij(y)为:

(6)

Fij(y)的期望为

(7)

2) 构建两两备选商品比较的前景随机占优关系矩阵

前景随机占优准则是一种基于前景理论的通过比较概率分布信息判断两两方案占优关系的准则[22-26],依据Levy等[24]和Levy等[25],关于前景随机占优准则的定义可表述如下:

依据上述随机占优准则的定义,下面给出针对本文需要解决的商品选择问题的随机占优准则的具体化描述。

由于备选商品Ai和Ah关于属性Cj的评价损益结果yij和yhj为离散型随机变量,那么,依据前景随机占优准则,对于∀y∈{Z1,Z2,…,Zv},若存在

(8)

则称商品Ai针对属性Cj的评价损益结果yij的累积分布函数Fij(y)前景随机占优于商品Ah针对属性Cj的评价损益结果yhj的累积分布函数Fhj(y),记作Fij(y)PSDFhj(y)。进一步地,依据式(5)和式(6),对于∀y∈{Z1,Z2,…,Zv},式(8)可进一步表示为如下形式:

(9)

这里,当上述随机占优准则应用于商品选择时,由于Fij(y)和Fhj(y)分别是商品Ai和Ah针对属性Cj的评价损益结果的累积分布函数,那么对于∀y∈{Z1,Z2,…,Zv},Fij(y)随机占优于Fhj(y)可以被视为商品Ai随机占优于商品Ah,这可记为AiPSDAh。

(10)

3) 选择适合的商品

对于属性Cj,任意两个备选商品Ai和Ah之间存在以下三种关系:

① 商品Ai严格优于商品Ah,即:Fij(y)PSDFhj(y)且uij≥uhj+αj;

② 商品Ai弱优于商品Ah,即:Fij(y)PSDFhj(y)且uhj

③ 商品Ai无差异于商品Ah,即:不存在Fij(y)PSDFhj(y)或Fhj(y)PSDFij(y)。

这里,uij和uhj分别表示备选商品Ai和Ah针对属性Cj的期望值;αj表示关于属性Cj的消费者偏好阈值,其与两两备选商品间的期望差值有关,其计算公式为:

(11)

其中

(12)

进一步地,可构造针对属性Cj的有序商品对(Ai,Ah)的偏好函数,即:

gj(Ai,Ah)=

(13)

其中,gj(Ai,Ah)∈[0,1]。gj(Ai,Ah)的值越接近0,商品Ai优于商品Ah的程度越小;当gj(Ai,Ah)=0时,商品Ai严格不占优于商品Ah;gj(Ai,Ah)的值越接近1,商品Ai优于商品Ah的程度越大。特别地,当gj(Ai,Ah)=1时,商品Ai严格占优于商品Ah。记Gj=[gj(Ai,Ah)]m×m为针对属性Cj的有序商品对(Ai,Ah)的占优程度矩阵,i,h=1,2,…,m,i≠h;j=1,2,…,n。

采用简单加权法则,可建立两两备选商品比较的总体优序度矩阵G=[g(Ai,Ah)]m×m,i,h=1,2,…,m,i≠h,其中,g(Ai,Ah)为有序商品对(Ai,Ah)的总体优序度,其计算公式为:

(14)

这里,g(Ai,Ah)可被视为备选商品Ai优于备选商品Ah的可信度,g(Ai,Ah)∈[0,1],g(Ai,Ah)越大,表示商品Ai优于商品Ah的程度越大。

依据总体优序度矩阵G=[g(Ai,Ah)]m×m,来计算商品Ai的“出流”Φ+(Ai)和“入流”Φ-(Ai),其计算公式分别为:

(15)

(16)

其中,Φ+(Ai)可被视为备选商品Ai优于其他所有备选商品的总可信度,Φ+(Ai)越大,商品Ai越好;Φ-(Ai)可被视为商品Ai劣于其他商品的总可信度,Φ-(Ai)越小,商品Ai越好。

依据Φ+(Ai)和Φ-(Ai),可计算备选商品Ai的排序值Φ(Ai),其计算公式为:

Φ(Ai)=Φ+(Ai)-Φ-(Ai),i=1,2,…,m

(17)

显然,Φ(Ai)越大,商品Ai越好。因此,依据排序值Φ(Ai)的大小,可对所有备选商品进行排序或进行商品的优选。

4) 商品选择的计算步骤

综上,考虑消费者给出属性在线评价期望的商品选择方法的具体计算步骤如下:

步骤4 依据式(11)~式(14)计算两两备选商品比较的总体优序度g(Ai,Ah),并建立总体优序度矩阵G=[g(Ai,Ah)]m×m;

步骤5 依据式(15)~式(17)计算每个备选商品的排序值,并依据排序值Φ(Ai)的大小选择适合的商品。

4 实例分析

第1步,使用八爪鱼采集器(https://www.bazhuayu.com/)从汽车之家网站获取4款备选汽车针对4个属性的在线评价信息,可确定参与4款备选汽车在线评价的用户数分别为:3268、3479、3750和3660,进一步地,依据式(1)和式(2)计算出4款备选汽车针对每个属性采用不同评价标度进行评价的用户数如表1所示。

表1 4款备选汽车针对每个属性采用不同评价标度进行评价的用户数

表2 4款备选汽车针对每个属性的评价损益结果的“概率”分布

然后,依据表2以及式(6)可分别计算每款备选汽车针对4个属性评价损益结果yij的累积分布函数Fij(y)及期望uij。这里以汽车A1(大众帕萨特)为例,针对4个属性的评价损益结果y1j的累积分布函数F1j(y)分别为:

进一步地,依据式(7),可计算上述累积分布函数的期望uij分别为u11=0.096,u12=1.422,u13=1.349,u14=-0.101。

第3步,运用前景随机占优准则(即依据式(9)和式(10)),建立针对每个属性的4款备选汽车两两比较的前景随机占优关系矩阵分别为:

第4步,依据式(11)和式(12)可计算得到各属性的偏好阈值分别为:

α1=0.644,α2=0.390,α3=0.372,α4=0.281

依据式(13)可计算得到针对每个属性的4款备选汽车两两比较的占优程度矩阵分别为:

进一步地,依据式(14)可建立两两备选汽车比较的总体优序度矩阵,即:

第5步,依据总体优序度矩阵G并由式(15)和式(16)可计算每款备选汽车的“出流”和“入流”分别为:

Φ+(A1)=1.433,Φ+(A2)=0.852,Φ+(A3)=1.443,Φ+(A4)=1.155

Φ-(A1)=1.191,Φ-(A2)=1.283,Φ-(A3)=1.608,Φ-(A4)=0.800

依据式(17)可计算每款汽车的排序值Φ(Ai),其结果为:

Φ(A1)=0.241,Φ(A2)=-0.432,Φ(A3)=-0.165,Φ(A4)=0.355

因此,依据得到的4款备选汽车的排序值Φ(Ai),可得到4款备选汽车的排序结果为:A4≻A1≻A3≻A2。

5 结语

本文给出了一种基于在线评价信息和消费者期望的商品选择方法。在该方法中,依据在线评价信息所呈现离散随机分布形式的特征,在考虑消费者给出属性评价期望的情况下,通过计算商品各属性的评价值相对于属性评价期望水平的损益值来确定关于各属性的评价损益结果的概率分布,在此基础上,通过运用前景随机占优准则给出了判断两两商品之间占优关系的计算公式,并依据此公式可构建两两备选商品比较的前景随机占优关系矩阵,进而运用PROMETHEE-II方法得到备选商品的排序结果。本文给出的方法具有概念清晰、计算过程简单、易于系统实现等特点,为解决现实中基于在线评价信息的商品选择问题提供了一种新途径。需要指出的是,本文给出的方法仅限于在线评价信息是多级评分的情形,今后的研究工作需要考虑具有文本形式的在线评论信息。

[1] 施晓菁, 梁循, 孙晓蕾. 基于在线评级和评论的评价者效用机制研究[J]. 中国管理科学, 2016, 24(5): 149-157.

[2] Qu Zhen, Zhang Han, Li Haizheng. Determinants of online merchant rating: Content analysis of consumer comments about Yahoo merchants [J]. Decision Support Systems, 2008, 46(1): 440-449.

[3] Casaló L V, Flavián C, Guinalíu M, et al. Do online hotel rating schemes influence booking behaviors? [J]. International Journal of Hospitality Management, 2015, 49: 28-36.

[4] 张紫琼, 叶强, 李一军. 互联网商品评论情感分析研究综述[J]. 管理科学学报, 2010, 13(6): 84-96.

[5] Ye Qiang, Law R, Gu Bin, et al. The influence of user-generated content on traveler behavior: An empirical investigation on the effects of e-word-of-mouth to hotel online bookings [J]. Computers in Human Behavior, 2011, 27(2): 634-639.

[6] Sparks B A, So K K F, Bradley G L. Responding to negative online reviews: The effects of hotel responses on customer inferences of trust and concern [J]. Tourism Management, 2016, 53: 74-85.

[7] Oh H K, Kim S W, Park S J, et al. Can you trust online ratings? A mutual reinforcement model for trustworthy online rating systems [J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2015, 45(12): 1564-1576.

[8] Liu Xiaolu, Guo Qiang, Hou Lei, et al. Ranking online quality and reputation via the user activity [J]. Physica A: Statistical Mechanics and its Applications, 2015, 436: 629-636.

[9] Liu Ying, Jin Jian, Ji Ping, et al. Identifying helpful online reviews: a product designer’s perspective [J]. Computer-Aided Design, 2013, 45(2): 180-194.

[10] Lee S, Choeh J Y. Predicting the helpfulness of online reviews using multilayer perceptron neural networks [J]. Expert Systems with Applications, 2014, 41(6): 3041-3046.

[11] Engler T H, Winter P, Schulz M. Understanding online product ratings: A customer satisfaction model [J]. Journal of Retailing and Consumer Services, 2015, 27: 113-120.

[12] Xu Xun, Li Yibai. The antecedents of customer satisfaction and dissatisfaction toward various types of hotels: A text mining approach [J]. International Journal of Hospitality Management, 2016, 55: 57-69.

[13] Ghasemaghaei M, Hassanein K. Online information quality and consumer satisfaction: The moderating roles of contextual factors-A meta-analysis [J]. Information & Management, 2015, 52(8): 965-981.

[14] Font F, Serrà J, Serra X. Class-based tag recommendation and user-based evaluation in online audio clip sharing [J]. Knowledge-Based Systems, 2014, 67(3): 131-142.

[15] Wang Wei, Wang Hongwei. Opinion-enhanced collaborative filtering for recommender systems through sentiment analysis [J]. New Review of Hypermedia and Multimedia, 2015, 21(3-4): 278-300.

[16] Li Shentun, Pham T T, Chuang Huichi, et al. Does reliable information matter? Towards a trustworthy co-created recommendation model by mining unboxing reviews [J]. Information Systems and e-Business Management, 2016, 14(1): 71-99.

[17] Li Yongli, Wu Chong, Luo Peng. Rating online commodities by considering consumers’ purchasing networks [J]. Management Decision, 2014, 52(10): 2002-2020.

[18] Yang Xian, Yang Guangfei, Wu Jiangning. Integrating rich and heterogeneous information to design a ranking system for multiple products [J]. Decision Support Systems, 2016, 84(C): 117-133.

[19] Chen Kun, Kou Gang, Shang J, et al. Visualizing market structure through online product reviews: Integrate topic modeling, TOPSIS, and multi-dimensional scaling approaches [J]. Electronic Commerce Research and Applications, 2015, 14(1): 58-74.

[20] Najmi E, Hashmi K, Malik Z, et al. CAPRA: A comprehensive approach to product ranking using customer reviews [J]. Computing, 2015, 97(8): 843-867.

[21] Liu Yang, Bi Jianwu, Fan Zhiping. Ranking products through online reviews: A method based on sentiment analysis technique and intuitionistic fuzzy set theory [J]. Information Fusion, 2017, 36: 149-161.

[22] Kahneman D, Tversky A. Prospect theory: An analysis of decision under risk [J]. Econometrica: Journal of the Econometric Society, 1979, 47(2): 263-291.

[23] Tversky A, Kahneman D. Advances in prospect theory: Cumulative representation of uncertainty [J]. Journal of Risk and Uncertainty, 1992, 5(4): 297-323.

[24] Levy H, Wiener Z. Stochastic dominance and prospect dominance with subjective weighting functions [J]. Journal of Risk and Uncertainty, 1998, 16(2): 147-163.

[25] Levy M, Levy H. Prospect theory: Much ado about nothing? [J]. Management Science, 2002, 48(10): 1334-1349.

[26] 张晓, 樊治平. 一种基于前景随机占优准则的随机多属性决策方法[J]. 控制与决策, 2010, 25(12): 1875-1879.

[27] Brans J P, Mareschal B. Multiple criteria decision analysis: State of the art surveys [M]. New York: Springer, 2005.

[28] Brans J P, VinckeP H. A preference ranking organization method: The PROMETHEE method for multiple criteria decision-making [J]. Management Science, 1985, 31(6): 647-656.

Method for Selecting Desirable Product(s) Based on Online Rating Informationand Customer’s Aspirations

YOUTian-hui,ZHANGJin,FANZhi-ping

(School of Business Administration, Northeastern University,Shenyang 110167, China)

In recent years, a large number of online ratings information about products has emerged on many ecommerce business websites, these online ratings information have significant impact on consumers’ understanding products and making purchase decisions. In reality, to make purchase decisions, consumer usually pour their attention to online ratings information of each attribute for products, and give their aspiration of online ratings information of each attribute for alternative products according to their demands. Accordingly, how to select desirable product(s) based on online ratings information and customer’s aspirations, it is a noteworthy research issue. On the basis, a method is proposed in this paper for the desirable product(s) selection considering online ratings information and customer’s aspiration based on the prospect stochastic dominance. In the method, first, the online ratings information of each attribute for alternative products is crawled by web crawler software, and the gain and loss for the alternative products can be calculated using the attribute rating value and the attribute aspiration, and then the probability distributions about the gain and loss results of each attribute for products are determined according to the obtained gain and loss values. On the basis, the cumulative distribution functions of gain and loss results and their expectations are obtained. Then, based on the obtained cumulative distribution functions of gain and loss values and their expectations, the prospect stochastic dominance relation matrices on pairwise comparisons of products with respect to each attribute are established according to the prospect stochastic dominance rule. Next, the degree of dominance on pairwise comparisons of each attribute of alternative products are calculated using PROMETHEE-II method, and the overall dominance degrees matrix for pairwise comparison of products is built using the simple additive weighting method, according to the obtained overall dominance degrees matrix, the “outflow”, “inflow” and the ranking values of each alternative products are calculated, respectively. Furthermore, a ranking of the alternative products is determined based on the obtained ranking values Finally, in order to illustrate the feasibility and validity of the proposed method, a case study about car selection is provided based on the online ratings information from the auto-home website.

online rating information; customer’s aspiration; prospect stochastic dominance; PROMET-HEE-II; product selection

1003-207(2017)11-0094-09

10.16381/j.cnki.issn1003-207x.2017.11.010

C934

A

2016-08-29;

2017-01-23

国家自然科学基金资助项目 (71271049,715710 39)

张瑾(1989-),女(汉族),山东莒县人,东北大学工商管理学院,博士研究生,研究方向:管理决策分析,E-mail: zjin0709@163.com.

猜你喜欢

损益排序矩阵
排序不等式
确认的公允价值变动损益需要转出吗?
恐怖排序
节日排序
刻舟求剑
初等行变换与初等列变换并用求逆矩阵
不同生态基流对水利枢纽发电损益的影响
矩阵
矩阵
矩阵