JURNAL OF CONSUMER BEHAVIOR

Examining the Impact of Ranking on Consumer Behavior and Search Engine

Revenue Anindya Ghose Department of Information, Operations and Management Sciences, Department of Marketing, Stern School of Business, New York University, New York, New York 10012, aghose@stern.nyu.edu Panagiotis G. Ipeirotis Department of Information, Operations and Management Sciences, Stern School of Business, New York University, New York, New York 10012, panos@stern.nyu.edu Beibei Li Information Systems and Management, Heinz College, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, beibeili@andrew.cmu.edu

I n this paper, we study the effects of three different kinds of search engine rankings on consumer behavior and search engine revenues: direct ranking effect, interaction effect between ranking and product ratings, and personalized ranking effect. We combine a hierarchical Bayesian model estimated on approximately one million online sessions from Travelocity, together with randomized experiments using a real-world hotel search engine application. Our archival data analysis and randomized experiments are consistent in demonstrating the following: (1) A consumer-utility-based ranking mechanism can lead to a significant increase in overall search engine revenue. (2) Significant interplay occurs between search engine ranking and product ratings. An inferior position on the search engine affects “higher-class” hotels more adversely. On the other hand, hotels with a lower customer rating are more likely to benefit from being placed on the top of the screen. These findings illustrate that product search engines could benefit from directly incorporating signals from social media into their ranking algorithms. (3) Our randomized experiments also reveal that an “active” personalized ranking system (wherein users can interact with and customize the ranking algorithm) leads to higher clicks but lower. 
 purchase propensities and lower search engine revenue compared with a “passive” personalized ranking system (wherein users cannot interact with the ranking algorithm). This result suggests that providing more information during the decision-making process may lead to fewer consumer purchases because of information overload. Therefore, product search engines should not adopt personalized ranking systems by default. Overall, our study unravels the economic impact of ranking and its interaction with social media on product search engines. Keywords: travel search engine; randomized experiments; hierarchical Bayesian methods; information systems; IT policy and management; electronic commerce History: Received May 28, 2012; accepted May 12, 2013, by Lorin Hitt, information systems. Published online in Articles in Advance. 1. Introduction Over the last few decades, search engines have emerged as a significant channel for promoting and selling products. In information search engines (e.g., Google) the ranking of the search results is an immediate signal of the relevance of the result to the query. However, in product search engines, the ranking of the displayed products is often based on criteria such as price, product rating, etc. In such a setting, we may often have multiple, potentially conflicting signals given to the customer about the products’ rankings. For example, if we rank by price, then the cheapest products sometimes have low product ratings, or products appearing on top of the list may be too expensive for the customer. Effectively consumers have to observe multiple, competing ranking signals and come up with their own ranking in their minds; in some settings, the product search engine will also generate personalized results, trying to rank the products according to the preferences of the consumer. In such an environment, we want to understand which factors influence the decision-making process of the customers and the magnitude of that influence. Are consumers influenced by the display ranking order, by the product rating, by price, and to what degree? How does this interplay affect the revenue that a search engine can generate?

1.1.  Related Work

In the last 10 years, the literature in e-commerce has shown the existence of a strong primacy effect in environments wherein consumers make choices among offers displayed in information search engines such as Google, Yahoo, or Bing. Specifically, we have learned that an online position effect exists and that rank order has a significant impact on the clickthrough rates and conversion rates (e.g., Ghose and Yang 2009, Yang and Ghose 2010, Agarwal et al. 2011, Jerath et al. 2011, Rutz and Trusov 2011, Narayanan and Kalyanam 2011, Animesh et al. 2011, Baye et al. 2012, Jeziorski and Segal 2012, Rutz et al. 2012, Abhishek et al. 2013, Ghose et al. 2013a). These papers focused primarily on evaluating the effect of screen position on user behavior, controlling for the quality of the advertisement. However, in product search engines, the observed demand patterns can be influenced by the joint variation in product ratings (either professional rating or user rating) and online screen position. The first goal of our study is to examine the position effect in product search engines, conditional on its interaction with product ratings. Search engines are beginning to adopt signals from social media sites directly into their ranking mechanism design (e.g., Bing Social Search, TripAdvisor). Recent work has found that a utility-based ranking mechanism on product search engines that incorporates multidimensional consumer preferences and social media signals can lead to significant surplus gain for consumers (Ghose et al. 2012). However, given that price was not the top priority considered in the ranking recommendation, whether such a mechanism can actually benefit product search engines is unclear because their revenues are normally commission based. Therefore, the second goal of our study is to examine the effect of different ranking mechanisms on product search engine revenue. Outside of search, one of the most important ways for shoppers to discover products has been through recommendation engines (Chittor 2010). However, although some online retailers use recommendation systems, many product-specific search engines (e.g., travel search engines) still do not provide personalized ranking results in response to consumer queries, presumably because these product search engine companies are unsure whether providing extra information to consumers will lead to an increase in profit. Existing research holds two different opinions on the effects of  personalization. One stream of work is supportive of personalization (e.g., Rossi et al. 1996, Ansari and Mela 2003, Arora and Henderson 2007, Yao and Mela 2011), whereas another stream of work is a bit more skeptical (e.g., Zhang and Wedel 2009, Aral and Walker 2011, Goldfarb and Tucker 2011, Lambrecht and Tucker 2013), suggesting that although personalization can lead to higher customer satisfaction and profits, it will not work as well universally.1 However, none of these papers have examined the effect of information availability and personalization in a search engine context. Koulayev (2014) examines consumers’ costly search behavior on travel search engines through the formation of consideration sets. Ghose et al. (2013b) build a structural econometric model to predict individual consumers’ online footprints on product search engines to improve user experiences under the context of social media overload. Chen and Yao (2012) use secondary data to examine how the sorting and filtering tools on travel search engines influence consumer hotel search. They find these tools result in a significant increase in total search activities, but they also lead to lower overall welfare because of the disproportional engagement induced by the refinement tools. With these findings in mind, our third goal is to examine how different kinds of personalized ranking mechanisms in product search engines affect consumer behavior and search engine revenues. Specifically, does allowing users to interact with the ranking algorithm to proactively personalize their search results lead to more or fewer purchases?

1.1.  Contributions and Results

 We situate our study in a travel search engine context, looking specifically at consumer selection of a hotel. We first apply archival data analysis to gain insights into the product-rating effects and ranking effects on consumers’ click and purchase behaviors. Using a panel data set from November 2008 to January 2009 containing approximately one million online user search sessions—including detailed information on consumer searches, clicks, and transactions obtained from Travelocity—we propose a hierarchical Bayesian framework in which we build a simultaneous equation model to jointly examine the interrelationship between consumers’ click and purchase behavior, search engine ranking decisions, and customers’ ratings. Toward the first goal, we examine the variation in the ratings of different hotels (both hotel “class” rating and customer rating) at the same rank on the travel search engine over time. In addition, our data setting has variation in the rank of the same hotel over time because the same hotel appears at different positions at different points in time. Controlling for room prices, such variation allows us to model the interaction effect of hotel class and customer ratings with rank and to measure its effect on demand. Toward the second goal, we examine how different ranking mechanisms affect the search engine revenue. We achieve this goal by conducting a set of policy experiments. We consider six different ranking designs: utility, conversion rate (CR), clickthrough rate (CTR), price, customer rating, and the Travelocity default algorithms. Then we estimate our model and predict future search engine revenues under each ranking mechanism. Toward our third goal, we examine how different levels of personalized ranking mechanisms affect consumer behavior and search engine revenue. Particularly, we compare two types of personalization mechanisms used to drive the ranking of results in response to a query: active personalized ranking and passive personalized ranking. In our context, a ranking system that allows consumers to proactively interact with the recommendation algorithm prior to the display of results from a search query is classified as “active.” By contrast, a ranking system that does not allow customers to interact with the recommendation algorithm is classified as “passive.” As of today, no hotel search engine has explicitly adopted a personalization-based approach to hotel ranking because they are still grappling with the issue of whether such an approach is useful.2 Hence, to our knowledge, no archival data in any product search engine have information on the effect of personalized ranking on user behavior. Therefore, we designed randomized experiments using a hotel search engine application that we built. Our randomized experimental results are based on a total of 900 unique user responses over a two-week period via the Amazon Mechanical Turk (AMT) crowdsourcing platform. We use a customized behavior-tracking system to observe the detailed information of consumers’ search, evaluation, and purchase decisionmaking process. By manipulating the default ranking method and by enabling or disabling a variety of personalization features on the hotel search engine website, we are able to study the effect of personalized ranking on consumer behavior. Our archival data analysis and randomized experiments are consistent in demonstrating the following: (1) A utility-based ranking mechanism can lead to a significant increase in the overall search engine revenue. (2) Significant interplay occurs between search engine ranking and product ratings. An inferior rank affects “higher-class” hotels more adversely. On the other hand, hotels with a lower customer rating are more likely to benefit from being placed on the top of the screen. These findings illustrate that product search engines could benefit from directly incorporating signals from online social media into the ranking algorithms. (3) Our randomized experiments also reveal that an active personalized ranking mechanism that enables consumers to specify both search context and individual preferences leads to more clicks but lower purchase propensities and lower search engine revenue, compared with passive personalized ranking mechanisms. A plausible explanation is related to theories of consumer cognitive cost. Prior theoretical work has shown that information overload and nonnegligible search costs can discourage decision makers from evaluating choices, leading to a scenario where they make no choices at all (Kuksov and Villas-Boas 2010). Our empirical finding dovetails with the theoretical conclusion by Kuksov and Villas-Boas that providing more information can actually lead to fewer purchases. It is also consistent with Dzyabura (2014), who shows that consumers who do not have well-formed preferences at the start of their search may be better off with uncertainty about product attribute levels rather than perfect knowledge of the attributes of all available products. Therefore, although an active personalized ranking recommendation may help consumers discover what they want to buy, product search engines should not ubiquitously adopt it. Two recent studies that are closely related to the current paper are Ghose et al. (2012) and Ghose and Yang (2009). However, this current paper distinguishes itself from the two previous studies in the following ways: (1) Ghose et al. (2012) do not focus on how rankings can benefit the search engine companies (in addition to the customers)—i.e., is it profitable for a search engine company to implement a utility-based ranking mechanism? In particular, we focus on examining whether the utility-based ranking leads to a significant improvement in the CTR, CR, and the total revenue for search engines. (2) Ghose et al. (2012) examine the direct WOM (word-ofmouth) effect on demand, without considering the impact of rank on the search engine. However, in our paper, we examine the WOM effect conditional on the ranking position of the product on search engines. We focus on examining the interaction effect between ranking and product ratings (both professional hotel class rating and online customer rating). Our findings illustrate that product search engines could benefit from directly incorporating signals from online social media into the ranking algorithms. (3) In Ghose et al. (2012), the authors did not focus on personalized rankings. Our randomized experiments reveal that an active personalized ranking mechanism that enables consumers to specify both search context and individual preferences leads to more clicks but lower purchase propensities and lower search engine revenues, compared with a passive personalized ranking mechanism. Therefore, although active personalized ranking recommendation may help consumers discover what they want to buy, it should not be adopted ubiquitously. (4) Ghose and Yang (2009) study the effect of keyword ranking on CTR and CR in the sponsored search context. However, our current paper looks at a different research context of ranking in product search engines. Compared with Ghose and Yang (2009), we make two method-based improvements in this paper. First, in addition to CTR, CR, and ranking, we model customer rating as a fourth dependent variable in the simultaneous model framework. Second, we allow for unobserved heterogeneity in all time-varying covariates. Our model fitness comparison results show that the model with full heterogeneity on all time-varying variables provides the overall best performance.

1.       Data

 Our data set consists of detailed information on a total of 969,033 online sessions from Travelocity.com, including consumer searches, clicks, and conversions that occurred within these sessions between November 2008 and January 2009. In addition, we have hotelrelated information, such as hotel class, brand, online reviewer rating, and number of reviews. We collected customer reviews from Travelocity.com. We collected the online reviews and reviewers’ information on a daily basis up to January 31, 2009 (the last date of transactions in our database). This process provides us with a final data set containing 29,222 weekly observations for 2,117 hotels in the United States.3 We define an “online session” to capture a set of activities by an online user, identified by a unique cookie. In our data, a starting indicator and an ending indicator with a corresponding time stamp (provided by the company) can characterize each unique online session. More specifically, a typical online session involves the initialization of the session, the search query, the results (in a particular rank order) returned from that search query, the sorting method, the click(s) on hotel(s) if any exist, the login and actual transaction(s) if any conversion occurs, and the termination of the session. The ending indicator marks the termination of a session.

We count a “display” for a hotel if that hotel appears visible to a consumer on the webpage in an online search session. Meanwhile, we count a “click” if a consumer selects the hotel and a “conversion” if a consumer has completed the payment in that online session. We only consider sessions with at least one display.4 A display can lead to a click, but it may not lead to a purchase. Each hotel that counts for a display is associated with a page number and a screen position, which capture the corresponding page order and (within-page) rank order of that hotel in the search results. Note that Travelocity only shows 25 hotels per page when it displays the hotel search results on a webpage.5 This design restricts the rank order for each hotel within the range from 1 to 25. Meanwhile, to facilitate consumer search, Travelocity provides a sorting criterion called “Travelocity Pick” by default. It also provides multiple alternative sorting criteria: price, hotel class, hotel name, and customer review rating. To capture consumers’ particular sorting preferences that may potentially influence the position effect, we include a set of control variables in our study to indicate how frequently a hotel appears in a result list under different sorting criteria. In particular, we use a vector (SpecialSort) that contains six control variables to capture the frequency of six sorting criteria that consumers use during their searches: default (DFT), price ascending (PRA), class descending (CLD), class ascending (CLA), city name ascending (CNA), and hotel name ascending (HNA). In summary, each observation in our data set contains the hotel ID, week ID, number of competing hotels, number of displays, number of clicks, number of conversions, average screen position (i.e., rank on the result page), average page number, and corresponding hotel characteristics in that week. For a better understanding of the variables in our setting, we present the definitions and the summary statistics of our data variables in Table 1.

1.       Empirical Model In this section,

 we discuss how we develop our simultaneous model in a hierarchical Bayesian framework. Then we describe how we apply the Markov chain Monte Carlo (MCMC) methods (Rossi and Allenby 2003) to empirically identify the effects of product quality and ranking position on consumer search and purchase behavior. More specifically, our model is motivated by the work of Ghose and Yang (2009). The general idea is as follows: We propose to build a simultaneous equations model of clickthrough, conversion, rank, and customer rating. We model the clickthrough and conversion behavior as a function of hotel brand, price, rank, page, sorting criteria, customer rating, and hotel characteristics. The rank of a hotel is modeled as a function of hotel brand, price, sorting criteria, customer rating, hotel characteristics, and performance metrics such as previous conversion rate. The customer rating of a hotel is modeled as a function of hotel brand, price, rank, page, sorting criteria, and hotel characteristics. Each function contains an unobserved error that is normally distributed with mean zero. To capture the unobserved covariation among clickthroughs, conversions, rank, and customer rating, we assume the four error terms are correlated and follow the multivariate normal distribution with mean zero. We describe our model next.

1.       Empirical Analyses and Results

To estimate our model, we applied the MCMC methods using a Metropolis–Hastings algorithm with a random walk chain (Chib and Greenberg 1995). In particular, we ran the MCMC chain for 80,000 iterations and used the last 40,000 iterations to compute the mean and standard deviation of the posterior distribution of the model parameters. We provide more details on the MCMC estimation algorithm in Online Appendix D.

rank, suggesting the negative effect of rank order on conversion rate also increases at a decreasing rate. As expected, Price has a negative effect on hotel demand, whereas Class has a positive effect on hotel demand. The online WOM-related variables, Rating and ReviewCnt, have a statistically significant and positive effect on hotel demand. We also found similar trends in the interaction effects between Ranking and Price/Class/Rating, suggesting higher-class hotels and more expensive hotels are more sensitive to the online ranking effect. And hotels that receive lower ratings from users benefit more when placed on the top of the screen. The total number of hotels in a certain market, H, has a negative effect on hotel-level conversion rate. Intuitively, the higher the number of choices there are available to consumers, lower the probability of buying from any given hotel. Thus, on average, the conversion rate for each hotel decreases.

1.       Conclusions and Implications

In this paper, we focus on investigating three major issues that product search engines are increasingly facing: the direct effect of ranking mechanism on consumer behavior and search engine revenue; the interaction effect of ranking and product ratings; and what kind of personalized ranking mechanism, if any, to adopt. Toward these objectives, we combine archival data analysis with randomized experiments based on a hotel search engine application that we designed. By manipulating the default ranking method and enabling or disabling a variety of active personalization features on the hotel search engine website, we are able to analyze consumer behavior and search engine revenue under different scenarios. Our experimental results on ranking are consistent with those from the Bayesian model-based archival data analysis, suggesting a significant and causal effect of search engine ranking on consumer click and purchase behavior. In addition to a significant surplus gain found by a previous study (Ghose et al. 2012), a consumer-utility-based ranking mechanism yields the highest purchase propensity and the highest search engine overall revenue compared with existing benchmark systems, such as ranking based on price or star ratings. Moreover, an inferior screen position tends to more adversely affect luxury hotels and more expensive hotels. Hotels with lower reputations benefit more from being placed at the top of the search results. This finding illustrates the need for product search engines to directly incorporate signals from online social media into the ranking algorithms. We are beginning to see much of this interplay between search and social media happening in information search engines. Google began to incorporate tweets and other social media status updates into its real-time search function and then decided to create its own version of the Facebook “Like” button— the Google +1—and have it show up in search results. In another example of the interplay between social media and search, Microsoft’s search engine Bing is now incorporating Facebook updates in its results. Our experimental results on personalized ranking show the availability of excess personalization capabilities during the decision-making process may discourage consumers from searching, evaluating, and making final choices. In particular, we find that although active personalized ranking, compared with passive personalized ranking, can attract more online attention from consumers, it leads to a lower purchase propensity and lower search engine revenue. This finding suggests that personalized ranking should not be adopted blindly and the level of personalization should be carefully designed based on the search context. Our research sheds light on how consumers search, evaluate choices, and make purchase decisions in response to differences in product search engine designs. We provide empirical and experimental evidence for future studies to build on when designing an efficient ranking system and dynamically modeling consumer behavior on product shopping sites. A good ranking mechanism can reduce consumers’ search costs, improve clickthrough rates and conversion rates of products, and improve revenue for search engines. Our work has some limitations, some of which we are striving to address in our ongoing work. First, although the AMT platform provides an efficient and cost-friendly framework for randomized experimental design, the inherent heterogeneity in the Internet population makes controlling for subject characteristics across different treatment groups difficult. The randomization process can alleviate this concern to a large extent. However, robustness tests based on offline subjects as well would be helpful. Our current experiments focus on the type of consumers who can make, at most, one purchase in each online shopping session. To better understand the counterintuitive finding that an active personalized ranking mechanism leads to lower conversion rates, one can extend our experimental design to make a comparison with consumers who are allowed to make multiple purchases in a given session. In addition, a study of how the content of consumer search, such as the length and type of search keyword, interacts with the ranking effect would be interesting. Moreover, with regard to examining the ranking mechanism, one can expand the research scope by taking into account consumers’ social network neighbors’ search and purchase behavior. This expansion would allow one to test the impact of social-signal-based ranking mechanisms on product search engines. Notwithstanding these limitations, we believe our paper paves the way for future research in this exciting area at the intersection of social media and search engines.

Acknowledgments

 For helpful comments, the authors thank Jason Chan, Sam Hui, Prasanna Tambe, Duncan Simester, Yong Tan, and Russ Winer, as well as seminar participants at Carnegie Mellon University, Cheung Kong Graduate School of Business, Columbia University, Harvard Business School, the Massachusetts Institute of Technology, University of Texas at Austin, University of Connecticut, University of Texas at Dallas Marketing Conference, the Wharton School, University of California, Los Angeles, and New York University. Anindya Ghose and Panagiotis G. Ipeirotis acknowledge funding from the Google Faculty Research Award in 2012.

      References

 Abhishek V, Hosanagar K, Fader P (2013) Aggregation bias in sponsored search data: The curse and the cure. Working paper, Carnegie Mellon University, Pittsburgh. Agarwal A, Hosanagar K, Smith M (2011) Location, location, location: An analysis of profitability of position in online advertising markets. J. Marketing Res. 48(6):1057–1073. Animesh A, Viswanathan S, Agarwal R (2011) Competing “creatively” in sponsored search markets: The effect of rank, differentiation strategy, and competition on performance. Inform. Systems Res. 22(1):153–169. Ansari A, Mela CF (2003) E-customization. J. Marketing Res. 40(2): 131–145. Aral S, Walker D (2011) Creating social contagion through viral product design: A randomized trial of peer influence in networks. Management Sci. 57(9):1623–1639. Arora N, Henderson T (2007) Embedded premium promotion: Why it works and how to make it more effective. Marketing Sci. 26(4):514–531. Baye MR, De los Santos B, Wildenbeest MR (2012) What’s in a name? Measuring prominence, and its impact on organic traf- fic from search engines. Working paper, Indiana University, Bloomington. Baye MR, Gatti JRJ, Kattuman P, Morgan J (2009) Clicks, discontinuities, and firm demand online. J. Econom. Management Strategy 18(4):935–975. Chen Y, Yao S (2012) Search with refinement. Working paper, Northwestern University, Evanston, IL.Chevalier J, Mayzlin D (2006) The effect of word of mouth on sales: Online book reviews. J. Marketing Res. 43(3):345–354. Chib S, Greenberg E (1995) Understanding the Metropolis–Hastings algorithm. Amer. Statistician 49(4):327–335. Chittor V (2010) Online retail: Getting the right product in front of your customers. Inc. (March 8), http://www.inc.com/internet/ articles/201002/chittoor.html. Dellarocas C (2012) Double marginalization in performancebased advertising: Implications and solutions. Management Sci. 58(6):1178–1195. De los Santos B, Koulayev S (2013) Optimizing click-through in online rankings for partially anonymous consumers. Working paper, Indiana University, Bloomington. Dodds WB, Monroe KB, Grewal D (1991) Effects of price, brand, and store information on buyers’ product evaluations. J. Marketing Res. 28(3):307–319. Dzyabura D (2014) Preference discovery in product search: When novice consumers don’t know what they don’t know. Working paper, New York University, New York. Ghose A, Yang S (2009) An empirical analysis of search engine advertising: Sponsored search in electronic markets. Management Sci. 55(10):1605–1622. Ghose A, Goldfarb A, Han S (2013a) How is the mobile Internet different? Search costs and local activities. Inform. Systems Res. 24(3):613–631. Ghose A, Ipeirotis PG, Li B (2012) Designing ranking systems for hotels on travel search engines by mining user-generated and crowd-sourced content. Marketing Sci. 31(3):493–520. Ghose A, Ipeirotis PG, Li B (2013b) Surviving social media overload: Predicting consumer footprints on product search engines. Working paper, Carnegie Mellon University, Pittsburgh. Goldfarb A, Tucker C (2011) Online display advertising: Targeting and obtrusiveness. Marketing Sci. 30(3):389–404. Greene W (1999) Econometric Analysis, 4th ed. (Prentice Hall, Upper Saddle River, NJ). Hausman JA (1975) An instrumental variable approach to full information estimators for linear and certain nonlinear econometric models. Econometrica 43(4):727–738. Jerath K, Ma L, Park Y, Srinivasan K (2011) A “position paradox” in sponsored search auctions. Marketing Sci. 30(4):612–627.

Latar belakang Perilaku konsumen

Selama beberapa dekade terakhir, search engine memiliki muncul sebagai saluran yang signifikan untuk mempromosikan dan menjual produk. Di mesin pencari informasi (misalnya, Google) peringkat hasil pencarian adalah langsung sinyal relevansi hasil untuk query. Namun, dalam mesin pencari produk, peringkat produk yang ditampilkan sering didasarkan pada kriteria seperti seperti harga, wisatawan produk, dll Dalam kondisi seperti ini, kami mungkin sering memiliki beberapa, sinyal berpotensi bertentangan diberikan kepada pelanggan tentang peringkat produk '. Sebagai contoh, jika kita peringkat berdasarkan harga, maka produk termurah kadang-kadang memiliki peringkat produk rendah, atau produk muncul di atas daftar mungkin terlalu mahal bagi pelanggan. Efektif konsumen harus mengamati beberapa, bersaing sinyal Peringkat dan datang dengan peringkat mereka sendiri dalam pikiran mereka; dalam beberapa pengaturan, mesin pencari produk juga akan menghasilkan hasil personalisasi, mencoba untuk peringkat produk menurut preferensi konsumen.

Dalam lingkungan seperti itu, kita ingin memahami faktor yang mempengaruhi proses pengambilan keputusan pelanggan dan besarnya pengaruh itu. Adalah konsumen dipengaruhi oleh peringkat display order, dengan rating produk, berdasarkan harga, dan untuk apa Gelar? Bagaimana interaksi ini mempengaruhi pendapatan bahwa sebuah mesin pencari dapat menghasilkan?

Metode Perilaku Konsumen

Pada bagian ini, kita membahas bagaimana kita mengembangkan simultan kami model dalam kerangka Bayesian hirarkis. Kemudian kita menggambarkan bagaimana kita menerapkan rantai Markov Monte Carlo (MCMC) metode (Rossi dan Allenby 2003) secara empiris mengidentifikasi efek dari produk kualitas dan posisi peringkat pada pencarian konsumen dan membeli perilaku. Lebih khusus, model kami 4 Dalam beberapa kasus, pengguna dapat memulai sesi dan mencari umum informasi wisata, seperti kawasan kota, daripada pencarian untuk setiap hotel; dengan demikian, tidak ada hotel akan ditampilkan pada halaman Web manapun. Kita mengecualikan sesi tersebut dari analisis kami. 5 Baru-baru ini, Travelocity upgrade desain halaman web dengan menunjukkan 10 hotel per halaman. Namun, selama periode waktu pemeriksaan kami, Jumlah itu masih 25. Termotivasi oleh karya Ghose dan Yang (2009). Ide umum adalah sebagai berikut: Kami mengusulkan untuk membangun sebuah model persamaan simultan klik per tayang, konversi, rank, dan dari pelanggan. Kita model klik per tayang dan konversi perilaku sebagai fungsi merek hotel, harga, peringkat, halaman, kriteria penyortiran, pelanggan Peringkat, dan hotel karakteristik. Pangkat Hotel dimodelkan sebagai fungsi hotel merek, harga, pemilahan kriteria, dari pelanggan, karakteristik Hotel, dan metrik kinerja seperti konversi sebelumnya tingkat. Rating pelanggan dari sebuah hotel dimodelkan sebagai fungsi dari hotel merek, harga, peringkat, halaman, menyortir kriteria, dan karakteristik Hotel. Setiap fungsi mengandung kesalahan tidak teramati yang biasanya didistribusikan dengan maksud nol. Untuk menangkap covariation teramati antara jumlah klik, konversi, peringkat, dan pelanggan peringkat, kita asumsikan empat hal error berkorelasi dan mengikuti distribusi normal multivariat.

Kesimpulan Perilaku Konsumen

Dalam tulisan ini, kita fokus pada menyelidiki tiga besar isu bahwa mesin pencari produk yang semakin menghadapi: efek langsung dari mekanisme pada konsumen perilaku dan mesin pencari pendapatan; interaksi pengaruh penilaian peringkat dan produk; dan apa jenis personalisasi mekanisme peringkat, jika ada, untuk mengadopsi. Menuju tujuan ini, kami menggabungkan arsip analisis data dengan percobaan acak berdasarkan aplikasi mesin pencari hotel yang kami dirancang. Dengan memanipulasi metode peringkat default dan mengaktifkan atau menonaktifkan berbagai personalisasi aktif fitur pada hotel Situs Web mesin pencari, kita mampu menganalisis perilaku konsumen dan mencari pendapatan mesin di bawah skenario yang berbeda.

            Hasil eksperimen kami pada personalisasi peringkat menunjukkan ketersediaan kemampuan personalisasi kelebihan selama proses pengambilan keputusan dapat mencegah konsumen dari mencari, mengevaluasi, dan membuat pilihan akhir. Secara khusus, kami menemukan bahwa meskipun aktif peringkat pribadi, dibandingkan dengan pasif pribadi ranking, dapat menarik lebih banyak secara online perhatian dari konsumen, itu mengarah ke pembelian lebih rendah kecenderungan dan pendapatan mesin pencari yang lebih rendah. Ini temuan menunjukkan bahwa peringkat pribadi seharusnya tidak diadopsi membabi buta dan tingkat personalisasi harus hati-hati dirancang berdasarkan pencarian konteks. Penelitian kami menyoroti bagaimana konsumenpencarian, mengevaluasi pilihan, dan membuat keputusan pembelian dalam menanggapi perbedaan mesin pencari produk desain. Kami memberikan bukti empiris dan eksperimental untuk studi masa depan untuk membangun ketika merancang sistem ranking yang efisien dan dinamis pemodelan perilaku konsumen di situs belanja produk. Mekanisme peringkat yang baik dapat mengurangi konsumen biaya pencarian, meningkatkan rasio klik per tayang dan konversi tingkat produk, dan meningkatkan pendapatan untuk mesin pencari.

            Pekerjaan kami memiliki beberapa keterbatasan, beberapa di antaranya kami berusaha untuk mengatasi dalam pekerjaan berkelanjutan kami. pertama, meskipun platform AMT menyediakan efisien dan Kerangka ramah-biaya untuk acak eksperimental desain, heterogenitas yang melekat di Internet . Populasi membuat mengendalikan karakteristik subjek di kelompok perlakuan yang berbeda sulit. Proses pengacakan dapat meringankan kekhawatiran ini untuk sebagian besar. Namun, tes ketahanan berdasarkan pada mata pelajaran offline juga akan sangat membantu. Kami eksperimen saat fokus pada jenis konsumen yang dapat membuat, paling banyak, satu pembelian secara online di setiap sesi belanja. Untuk lebih memahami berlawanan menemukan bahwa peringkat pribadi yang aktif. Mekanisme mengarah untuk menurunkan tingkat konversi, satu dapat memperpanjang desain eksperimental kami untuk membuat perbandingan dengan konsumen yang diperbolehkan untuk membuat beberapa pembelian dalam sesi yang diberikan. Selain itu, studi bagaimana isi pencarian konsumen, seperti panjang dan jenis pencarian kata kunci, berinteraksi dengan efek peringkat akan menarik. Selain itu, dengan Berkenaan dengan memeriksa mekanisme peringkat, satu kaleng memperluas cakupan penelitian dengan memperhatikan konsumen akun ' jaringan sosial tetangga 'pencarian dan pembelian perilaku. Ekspansi ini akan memungkinkan seseorang untuk menguji dampak sosial-sinyal berbasis mekanisme peringkat di search engine produk. Meskipun keterbatasan ini, kami percaya kami kertas membuka jalan untuk penelitian masa depan di daerah ini menarik di persimpangan media sosial dan mesin pencari.