張 瑾，尤天慧

考慮消費(fèi)者期望的多屬性在線評論商品選擇方法

張 瑾，尤天慧

(東北大學(xué) 工商管理學(xué)院，遼寧 沈陽 110169)

電子商務(wù)網(wǎng)站涌現(xiàn)的關(guān)于商品的大量在線評論對于消費(fèi)者了解商品并做出購買決策具有重要的影響，如何基于商品的多屬性在線評論并考慮消費(fèi)者給出屬性評論期望進(jìn)行商品選擇，是一個值得關(guān)注的研究問題。本文提出一種基于情感分析和前景理論的商品選擇方法。在該方法中，首先，依據(jù)獲取的備選商品的屬性在線評論構(gòu)建屬性領(lǐng)域情感詞典，并基于屬性領(lǐng)域情感詞典識別每條在線評論的情感傾向；其次，通過統(tǒng)計(jì)正向、負(fù)向和中性在線評論的比率確定備選商品各屬性的正、負(fù)向評價值；然后，依據(jù)前景理論，分別計(jì)算正、負(fù)向評價值相對于消費(fèi)者給出的屬性評論正、負(fù)向期望水平的正、負(fù)向損益值，進(jìn)而計(jì)算關(guān)于各屬性正、負(fù)向評價值所對應(yīng)的正、負(fù)向前景價值；進(jìn)一步地，采用簡單加權(quán)法則，計(jì)算各備選商品的正、負(fù)向綜合前景價值；在此基礎(chǔ)上，計(jì)算每個備選商品的綜合前景價值得到備選商品的排序結(jié)果。為了說明本文提出方法的可行性和有效性，依據(jù)汽車之家網(wǎng)站提供的在線評論，給出汽車商品選擇的實(shí)例分析。

在線評論；消費(fèi)者期望；情感分析；前景理論；商品選擇

0 引言

隨著互聯(lián)網(wǎng)和電子商務(wù)迅速發(fā)展，許多網(wǎng)站如汽車之家、淘寶和亞馬遜等網(wǎng)站均為消費(fèi)者提供了分享購物經(jīng)驗(yàn)的平臺，消費(fèi)者可通過網(wǎng)站發(fā)表對已購商品的評論信息[1, 2]。與賣方提供的商品描述信息相比，這些由消費(fèi)者提供的商品在線評論能夠更客觀的反映商品的真實(shí)情況[3]。但由于在線評論信息體量非常大，消費(fèi)者很難查看其關(guān)注的商品的所有在線評論信息并得到綜合評價結(jié)果，因此，如何基于大量商品在線評論信息給出消費(fèi)者關(guān)注的備選商品排序輔助其選擇成為重要研究問題。

目前，針對基于在線評論的商品選擇/排序問題的研究已引起了學(xué)者的關(guān)注[4-14]。例如，Wang等通過情感分析提取在線評論中商品特征-意見對來構(gòu)建計(jì)量經(jīng)濟(jì)學(xué)模型進(jìn)而對商品特征進(jìn)行排序[4]；Peng等基于在線評論通過情感分析來提取競爭商品的特征及確定特征權(quán)重，并依據(jù)專家給出的商品關(guān)于特征的模糊決策矩陣使用模糊PROMETHEE方法對競爭商品進(jìn)行排序[5]；Chen等利用主題建模方法確定正向評論和負(fù)向評論的商品屬性及屬性初始權(quán)重矩陣，然后利用WVAP方法獲得最終商品正向評論和負(fù)向評論的屬性權(quán)重矩陣，并使用TOPSIS方法對可視化產(chǎn)品市場結(jié)構(gòu)中競爭商品進(jìn)行排序[6]；Najmi等依據(jù)商品在線評論及商品說明書確定商品特征（屬性）及屬性權(quán)重，采用改進(jìn)Page-Rank排序算法對商品品牌排序并利用支持向量機(jī)和有用性分析計(jì)算結(jié)果對商品在線評論進(jìn)行排序，最終得到商品的排序結(jié)果[7]；Yang等基于在線評論、在線評級、比較語句和比較投票數(shù)4種類型的在線評論/評價信息給出了一種新的商品電子口碑排序方法，并進(jìn)行了系統(tǒng)實(shí)現(xiàn)以幫助消費(fèi)者進(jìn)行商品比較與選擇[8]；Liu等通過對在線評論進(jìn)行情感分析并使用直覺模糊集形式表示商品屬性值信息，采用PROMETHEE-II方法對備選商品進(jìn)行排序[9]；梁霞等利用商品的在線評論信息，首先從在線評論中提取商品屬性并確定屬性權(quán)重，然后通過對評論中的情感詞進(jìn)行分析，將消費(fèi)者情感傾向表示為關(guān)于評價標(biāo)度的概率分布，進(jìn)而采用隨機(jī)占優(yōu)準(zhǔn)則和PROMETHEE II對商品進(jìn)行排序[10]；Kang等通過對移動服務(wù)的在線評論進(jìn)行情感分析，提取移動服務(wù)的特征（屬性）并確定特征權(quán)重，利用VIKOR方法對備選移動APP服務(wù)客戶滿意度進(jìn)行排序[11]；Zhang等采用情感分析方法確定在線評論的情感極性，然后依據(jù)商品評論的情感極性建立有向加權(quán)特征圖確定商品排序[12]；Zhang等通過分析商品在線報(bào)告建立商品特征集以及標(biāo)注各商品特征在線評論的情感極性，并依據(jù)商品評論的情感極性建立有向加權(quán)特征圖確定商品排序[13]，在此基礎(chǔ)上，Zhang等通過引入評論幫助性投票和評論發(fā)布日期等信息確定在線評論的重要性權(quán)重進(jìn)而對文獻(xiàn)[13]的方法進(jìn)行了擴(kuò)展[14]。

從已有的研究成果可以看到，很少考慮消費(fèi)者關(guān)于商品選擇的偏好或期望信息。在現(xiàn)實(shí)中，消費(fèi)者通過查看在線評論進(jìn)行商品選擇時，常會針對預(yù)選商品屬性的在線評論有期望，例如，消費(fèi)者有意愿購買汽車時，其會查看相關(guān)網(wǎng)站（如汽車之家）上關(guān)于汽車油耗、動力、舒適性等屬性的在線評論，并期望針對汽車油耗屬性的在線評論好評率不低于85%，可能同時期望針對汽車油耗屬性的在線評論差評率不高于10%等。鑒于此，本文研究基于在線評論且考慮消費(fèi)者給出屬性評論期望的商品選擇方法。該方法的研究思路是：首先，利用網(wǎng)絡(luò)爬蟲軟件獲取備選商品關(guān)于各屬性的在線評論并進(jìn)行預(yù)處理，進(jìn)而構(gòu)建商品屬性領(lǐng)域情感詞典；其次，依據(jù)構(gòu)建的屬性領(lǐng)域情感詞典識別每條在線評論的情感傾向，并計(jì)算各屬性在線評論的正向、負(fù)向和中性情感傾向的比率；然后，依據(jù)前景理論，計(jì)算備選商品各屬性相對于消費(fèi)者給出的屬性評論正向期望水平的正向評價損益值和相對于消費(fèi)者給出的屬性評論負(fù)向期望水平的負(fù)向評價損益值，并依據(jù)得到的正向評價損益值和負(fù)向評價損益值，計(jì)算備選商品關(guān)于各屬性的正向前景價值和負(fù)向前景價值；進(jìn)一步地，采用簡單加權(quán)法則，計(jì)算每個備選商品的正向綜合前景價值和負(fù)向綜合前景價值；在此基礎(chǔ)上，計(jì)算每個備選商品的綜合前景價值并依據(jù)綜合前景價值的大小得到備選商品的排序結(jié)果。

1 問題描述

針對考慮消費(fèi)者期望的多屬性在線評論商品選擇問題，為方便起見，采用下面的符號表示該問題中所涉及的集和量：

本文要解決的問題是：針對消費(fèi)者關(guān)注的若干備選商品及相關(guān)屬性，依據(jù)相關(guān)電商網(wǎng)站提供的備選商品的在線評論、消費(fèi)者給出的屬性評論期望和屬性權(quán)重向量，如何使用一個決策分析方法對備選商品進(jìn)行排序或者選擇適合的商品。

2 商品選擇方法

為了解決上面提及的問題，這里闡述考慮消費(fèi)者期望的多屬性在線評論商品選擇方法。該方法包括4個部分，即備選商品在線評論的采集與預(yù)處理，商品屬性領(lǐng)域情感詞典的構(gòu)建，基于商品屬性領(lǐng)域情感詞典的在線評論情感傾向識別以及考慮消費(fèi)者屬性評論期望的商品選擇。下面分別給出該方法中每個部分的計(jì)算過程描述，最后給出該方法的計(jì)算步驟。

2.1 備選商品在線評論的采集與處理

2.2 商品屬性領(lǐng)域情感詞典的構(gòu)建

考慮到針對不同屬性的評論情感詞集合可能不同，且同一情感詞出現(xiàn)在針對不同屬性的評論中可能表達(dá)了不同的情感傾向，如“低”在描述價格時表達(dá)為正情感傾向，而在描述性價比時表達(dá)為負(fù)情感傾向。因此，為了提高在線評論情感分析的準(zhǔn)確性，本文針對商品不同屬性分別建立屬性領(lǐng)域情感詞典。下面給出屬性領(lǐng)域情感詞典構(gòu)建的具體過程。

2.3 基于商品屬性領(lǐng)域情感詞典的在線評論情感傾向識別

2.4 考慮消費(fèi)者給出屬性評論期望的商品選擇方法

3 實(shí)例分析

圖1 在線評論情感傾向的確定過程

Figure 1 The process of identifying sentiment orientation of online reviews

（1）備選轎車屬性領(lǐng)域情感詞典的構(gòu)建

（2）基于屬性領(lǐng)域情感詞典的備選汽車在線評論情感傾向識別

（3）考慮消費(fèi)者給出屬性評論期望的轎車選擇

本文針對從汽車之家網(wǎng)站獲取的4款備選轎車關(guān)于4個屬性的所有在線評論通過第(2)部分的步驟，得到每款備選轎車針對每個屬性的在線評論的情感傾向向量指示值后，運(yùn)用式(9)~式(11)可計(jì)算每款備選轎車關(guān)于各屬性的正向評論、負(fù)向評論和中性評論的評論數(shù)量，計(jì)算結(jié)果如表1所示。運(yùn)用式(12)和式(13)，計(jì)算得到每款備選轎車關(guān)于各屬性的正向評價值和負(fù)向評價值，如表2所示。

表1 4款備選轎車關(guān)于各屬性不同情感傾向的評論數(shù)量

表3 4款備選汽車針對每個屬性的評價損益值

表4 4款備選汽車針對每個屬性的正向前景價值和負(fù)向前景價值

表5 取值及相應(yīng)的備選轎車排序結(jié)果

4 結(jié)論

本文給出了一種基于多屬性在線評論和消費(fèi)者期望的商品選擇方法。該方法通過提取備選商品屬性在線評論中情感詞構(gòu)建屬性領(lǐng)域情感詞典，依據(jù)構(gòu)建的屬性領(lǐng)域情感詞典識別每條在線評論的情感傾向，并計(jì)算各屬性評論的正向、負(fù)向和中性情感傾向的比率，在此基礎(chǔ)上，給出基于前景理論的考慮消費(fèi)者期望的產(chǎn)品選擇方法。本文給出的方法概念清晰，具有較強(qiáng)的可操作性，有實(shí)際應(yīng)用價值。

[1] 王祖輝, 姜維, 李一軍. 在線評論情感分析中固定搭配特征提取方法研究[J]. 管理工程學(xué)報(bào), 2014, 28(4): 180-186.

Wang Z H, Jiang W, Li Y J. Regular collocation features extraction method in online reviews sentiment analysis [J]. Journal of Industrial Engineering and Engineering Management, 2014, 28(4): 180-186.

[2] Chen H, Chiang RHL, Storey VC. Business intelligence and analytics: From big data to big impact [J]. MIS Quarterly, 2012, 36(4): 1165-1188.

[3] Liu Y, Bi JW, Fan ZP. A Method for ranking products through online reviews based on sentiment classification and interval-valued intuitionistic fuzzy TOPSIS [J]. International Journal of Information Technology & Decision Making, 2017, 16(6): 1497-1522.

[4] Wang W, Wang HW, Song Y. Ranking product aspects through sentiment analysis of online reviews [J]. Journal of Experimental & Theoretical Artificial Intelligence, 2017, 29(2): 227-246.

[5] Peng Y, Kou G, Li J. A fuzzy PROMETHEE approach for mining customer reviews in Chinese [J]. Arabian Journal for Science and Engineering, 2014, 39(6): 5245-5252.

[6] Chen K, Kou G, Shang J,. 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.

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

[8] Yang X, Yang GF, Wu JN. Integrating rich and heterogeneous information to design a ranking system for multiple products [J].Decision Support Systems, 2016, 84: 117-133.

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

[10] 梁霞, 姜艷萍, 高夢. 基于在線評論的產(chǎn)品選擇方法[J]. 東北大學(xué)學(xué)報(bào)(自然科學(xué)版), 2017, 38(1): 143-147.

Liang X, Jiang Y P, Gao M. Product selection methods based on online reviews [J]. Journal of Northeastern University (Natural Science), 2017, 38(1): 143-147.

[11] Kang D, Park Y. Review-based measurement of customer satisfaction in mobile service: Sentiment analysis and VIKOR approach [J]. Expert Systems with Applications, 2014, 41(4): 1041-1050.

[12] Zhang K, Narayanan R, Choudhary A. Mining online customer reviews for ranking products [R]. Technical Report, EECS department, Northwestern University, 2009.

[13] Zhang K, Narayanan R, Choudhary A. Voice of the customers: mining online customer reviews for product feature-based ranking [A]. In: Proceedings of the 3rd Conference on Online Social Networks [C].Boston: Workshop on Online Social Networks, 2010, 11-11.

[14] Zhang K, Cheng Y, Liao WK,. Mining millions of reviews: a technique to rank products based on importance of reviews [A]. In: Proceedings of the 13th International Conference on Electronic Commerce [C]. Liverpool: Conf. Electronic Commerce, 2011, 121-128.

[15] 王偉, 王洪偉, 盛小寶. 中文在線評論的產(chǎn)品特征與觀點(diǎn)識別: 跨領(lǐng)域的比較研究[J]. 管理工程學(xué)報(bào), 2017, 31(4): 52-62.

Wang W, Wang H W, Sheng X B. Extracting product features and opinions from Chinese online reviews: A comparative study on multi-domains [J]. Journal of Industrial Engineering and Engineering Management, 2017, 31(4): 52-62.

[16] 中國科學(xué)院. ICTCLAS漢語分詞系統(tǒng) [EB/OL]. [2015-12-30]. http://www.ictclas.org.

Chinese Academy of Sciences. ICTCLAS Chinese word segmentation system [EB/OL]. [2015-12-30]. http://www.ictclas.org.

[17] 董振東. HowNet [EB/OL]. [2013-10-16]. http://www.keenage.com.

Dong Z D. HowNet [EB/OL]. [2013-10-16]. http://www.keenage.com.

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

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

Method for selecting desirable product(s) through multiple attribute online reviews considering customer's aspirations

ZHANG Jin, YOU Tianhui

(School of Business Administration, Northeastern University, Shenyang 110169, China)

In recent years, with the rapid development of the Internet and E-commerce, a large number of online reviews of product(s) have emerged on many E-commerce websites. Online reviews have a significant impact on consumer's understanding of products and have become critical information for consumers to select the desired product(s). When a customer wants to buy some products, he or she will view online reviews on the attribute of alternative products on the one hand, and has aspirations for online reviews of alternative product attributes on the other. Present researches have investigated the ranking of products based on online reviews. However, the existing studies on the product ranking method based on online reviews have not taken into account the consumer's aspirations. Therefore, how to select the desired product(s) through multiple attribute online reviews considering consumer's aspirations, it is a noteworthy research issue.

A method based on the sentiment analysis and the prospect theory is proposed to solve the problem of selecting the desired product(s) through multiple attribute online reviews considering consumer's aspirations in this paper. In the method, first, the online reviews on alternative product attributes concerned by consumers are crawled by web crawler software from the related website, and the online reviews are preprocessed by using the ICTCLAS 2015. Second, according to the preprocessed online reviews and the HowNet sentiment dictionary, the domain sentiment dictionary concerning each attribute for the products is constructed. The sentiment orientation of each online review is identified by using an algorithm of sentiment analysis afterward. Then, the positive and negative evaluation values of each attribute for alternative products are calculated by counting the proportion of online reviews with positive, negative, and neutral sentiment orientation. Further, according to the prospect theory, the positive prospect value of each attribute for the alternative products is obtained by calculating the positive gains and losses using the attribute positive evaluation value relative to the attribute positive aspiration, similarly, for the negative prospect value of each attribute for the alternative products is obtained by calculating the negative gains and losses using the attribute negative evaluation value relative to the attribute negative aspiration. On the basis, using the simple additive weighting method, the overall positive prospect value, and the overall negative prospect valueof each attribute for the alternative products are calculated. Furthermore, the overall prospect value of each alternative product is calculated based on the overall positive prospect value, and the overall negative prospect value, an order ranking of all alternative products, or select the desirable alternative product(s) is determined. Finally, to illustrate the feasibility and validity of the proposed method, a case study about car selection is provided based on the online reviews from the auto-home website.

The proposed method improves the existing research methods to effectively solve the problem of selecting a desirable product(s) through multiple attribute online reviews considering customer's aspirations. The method proposed in this paper has a clear concept, strong operability, and practical application value.

Online reviews; Customer's aspiration; Sentiment analysis; Prospect theory; Product selection

2018-02-09

2018-05-31

Supported by t the Natural Science Foundation of China (71271049, 71771043)

2018-02-09

2018-05-31

國家自然科學(xué)基金資助項(xiàng)目（71271049、71771043）

張瑾（1989—），女，山東莒縣人；東北大學(xué)工商管理學(xué)院博士研究生；研究方向：管理決策分析、電子商務(wù)。

C934

A

1004-6062(2020)05-0024-008

10.13587/j.cnki.jieem.2020.05.003

中文編輯：杜 ?。挥⑽木庉嫞篊harlie C. Chen