Profile of User Search Behavior and Advertising on Alibaba’s Taobao Platform

International Journal of Business Marketing and Management (IJBMM)
Volume 4 Issue 9 September 2019, P.P. 50-60
ISSN: 2456-4559
www.ijbmm.com
International Journal of Business Marketing and Management (IJBMM) Page 50
Profile of User Search Behavior and Advertising on Alibaba’s
Taobao Platform
Min Li1
, Joseph Richards2
1
(Department of Information Systems and Business Analytics, CSU-Sacramento, USA)
2
(Department of Marketing, CSU-Sacramento, USA)
ABSTRACT : This paper uses the rich features of two data sets obtained from the Alibaba’s Taobao online
shopping site taobao.com. The page view log data file contains the information about the advertisements
displayed after the customer’s search using keyword(s). The click log data file contains the information about
the advertisements clicked by customers. Both data files contain information specific to customers including
their geographic location, the time of search, and the time of clicking an advertisement. Exploratory data
analysis reveals the profile of search behavior and advertising strategy employed by Taobao. This analysis will
be valuable for replicating the success of Taobao platform in other emerging markets.
KEYWORDS - online consumer behavior, online search, sponsored search, taobao
I. Introduction
Today, China’s e-commerce market is the world’s most voracious, the Single’s Day promotion sales on
Alibaba’s online platforms exceeded $30 billion in 2018. According to CNBC, Alibaba’s Singles Day sales haul
easily exceeded the spending by consumers during any single U.S. shopping holiday. By 2020, China’s e-
commerce market is forecast to be bigger than the size of markets in America, Britain, Japan, Germany, and
France combined (“The Alibaba Phenomenon”, 2013). Using the Taobao search platform, consumers often
obtain information before shopping online, such as product prices, the origin of the goods, product quality, and
the seller’s credit evaluation. The search engine also records the query process and any clicks by the customer
on advertisements or search links displayed (Wei et al., 2014).
The plethora of innovations that are emerging from China calls for a closer look to study their specific
features, and perhaps by doing so will offer valuable lessons for other emerging markets. This paper uses the
customer search data from Taobao.com, a consumer-to-consumer portal and one of the principal arms of
Alibaba, for analyzing the search advertising business carried out by Taobao. Customer search behavior in
conjunction with the search advertising carried out by the portal in response to search and browsing, provide a
rich source of data for analysis and important insights.
II. The Taobao platform
The users search for products on Taobao (taobao.com) by entering keywords. After a user searches for a product
by entering a keyword, four columns of pictures and texts relevant to this product are displayed in the central
area of the webpage. Two additional columns, on the right and at the bottom, display product advertisements
from various merchants who have bid this keyword entered by the customer. Discounted prices for the
advertised products are usually shown in these columns with the original list (usually higher) prices crossed out.
Taobao’s search engine matches the keywords entered by customers to the keywords (bidwords in Taobao’s
terminology) merchants have bid for online advertisements. The search engine must determine which
advertisements to display on the right side and at the bottom of the page, and the order and position of the
advertisements. The advertisement receiving the highest bid from the merchants is usually placed at the top of
the list. Merchants pay Taobao the bid price each time a potential customer clicks the advertisement. Thus, the
advertising mechanism on Taobao.com is similar to Google’s AdWords. In addition to sponsored search
advertising, Taobao also provides other services including display advertising (contextual advertising) and

Profile of User Search Behavior and Advertising on Alibaba’s Taobao Platform
behavioral targeting. Display advertising involves showing advertisements such as banners on a website aiming
to drive the traffic to advertisers’ own site. Behavioral targeting predicts customers’ interests based on their past
online behavior and other features (see Svensen, et al. 2011). Then customers are divided into segments of
similar interest for targeted advertising campaigns. All these services allow merchants and advertisers to reach a
large number of target customers with the highest purchase intent for the advertised products.
This paper is exploratory in nature and complements the work done by the authors in a related paper
(see Richards and Li, 2018) where the authors’ focus was to infer strategic insights that could be tested by the
data. This paper presents exploratory analysis of the data and suggests potential managerial implications which
are of immediate concern for e-commerce-oriented internet platforms.
The next section describes the data and its analysis. In the data analysis part, the profile of user search
and specific features of Taobao’s search advertising strategies are presented. The paper concludes with a
discussion on the managerial implications and avenues for further exploration.
III. Data and Analysis
In sponsored search advertising, a customer searches a term using a search engine and the search results (links)
including those sponsored by advertisers are returned. Customers’ clicking on links sponsored by advertisers
results in payments to search engines. Thus, it is in the interest of search engines to maximize the click-through
rate (CTR), defined as the ratio of the number of clicks an advertisement received and the number of times the
advertisement was shown. Advertisement features include bid phrases (bid words, keywords), the title of the
advertisement, the text of the advertisement, the URL of the landing page, the landing page, and the advertiser
information such as account, campaign, advertisement group, and the advertisement. Query features include the
search keywords and related expansion. Context features include display and geographic locations, time, search
history, and other customer data. Many of these data features appear in the two data sets from Taobao described
below.
The data from Taobao’s paid search ranking (pay-for-performance) system are used for this study. Two data sets
were sampled at one author’s request by data scientists at Taobao between 2 pm and 2:05 pm on May 16, 2013,
recording online customers’ behavior during this time period. The short 5-minute sampling period was chosen
so that the data file size is manageable. These data files include Chinese characters taking up several gigabytes
of storage space. The variables captured in the two data files, the page view log file and the click log file, are
described in detail below.
Page View Log file
Table 1 describes the important variables in this data file. This file recorded the information related to the
advertisements displayed to the customer after the customer’s search using keyword(s) from 2 pm to 2:05 pm on
May 15, 2013. The file contains 997,407 observations and 65 variables. About one third of the variables are ID-
type variables referring to cities, provinces, keywords merchants placed bids on, customer queries, advertising
customers, advertisement groups, etc. Variables such as adhighestprice, adprice, adrelativeposition, adscoretag,
and adsnumperpage are explained in Table 1, and are similar to the online advertisement features described in
Graepel et al. (2010). Three variables, pvtime, city, and province show the time and location of the customer
when viewing this page. These belong to the context features also described in Graepel et al. (2010). Some of
these variables are used to track advertising customers, specific advertisements, or traffic related to the websites.
The variable rawquery_keyword contains the search keywords and its related expansion. The variable bidword
contains words that are a part of the variable rawquery_keyword. Both variables belong to the query features
described in Graepel et al. (2010). Additional explanation of select variables and the information they capture
are given in Table 1.
Frequencies of the variable adsnumperpage are reported in Table 2, ordered from the most frequent to
the least frequent, after sorting the data by customers (acookie) and the keywords (rawquery_keyword) these
customers searched, removing duplicate values. Sorting by these two variables and removing the duplicate
values were necessary so that each search result webpage corresponds to only one value for the variable

adsnumperpage. The most common number of advertisements per page is 13 (37%) and the next one is 3
(18.3%). The range observed is 59. The number of advertisements commonly displayed shows a wide range
from 3 to 13. This certainly is on account of the user, the type of product searched for that session, and the
length of the search session. It also shows a high degree of customization of advertisements to the user. The
number of advertisements shown in response to user search will likely depend, among other factors, on the type
of product searched, whether the user uses a query word, and the time spent in a search session. Exploration of
these relationships is a topic for future research. Knowledge of these relationships will reveal the specific
strategies employed by Taobao on the platform.
The variable adprice contains the prices of products displayed in the sponsored advertisements on the
right and at the bottom of the page, for each customer identified by the variable acookie. The variable
adhighestprice captures the highest list price or regular price of the advertised product from which the price is
marked down to adprice. For most advertisements, adhighestprice is crossed out and adprice, i.e., the
discounted price, is displayed next to it. The adhighestprice ranges from 5 to 9,900 Chinese Yuan, with 164 as
the average and 122 Yuan as the median price. The histograms of both variables, adprice and adhighesprice,
are shown in Figures 1 and 2. A few observations may be made about these histograms. The long-tailed
distribution observed in these price histograms demonstrates that the advertisements shown are for products
from the low-end to the high-end of the price spectrum. Avenues of further inquiry include whether the price
dispersion for advertised products within a category differs between product categories, whether the dispersion
depends on the user using a query key word for search, and whether the dispersion and the length of search
session are related.
Fig. 3 displays the histogram of adscoretag, recording the number of clicks on the advertisement. The
range of this variable value is from 0 to 999, with a mean of 373 and median of 311. The first quartile is 207 and
the third quartile is 481. The histogram in Figure 3 shows that most advertisements resulted in fewer than 500
clicks. The histogram shows a long-tailed distribution proving that some types of advertisements stand out. An
avenue for further research is to study the characteristics of such advertisements and figure out the what factors
might have caused a spike in the number of clicks.
Fig. 4 displays the histogram of the percentage by which adprice is marked down from adhighestprice,
i.e., the discount rate based on adprice and adhighestprice. There is no discount for about half of the
advertisements, and 25% of the advertisements have a markdown of at least 5%. An important insight would be
to know whether the amount of price discounts had any effect on the click through rates (CTR) and whether the
amount of discount varied across product categories.
Tables 3 and 4 list the top 10 cities and provinces where customers are located. Almost half of the
customers are in the coastal provinces (Guangdong, Zhejiang, Jiangsu, Shandong, Beijing, Shanghai), cities that
are relatively more developed compared to the rest of China. Most of the top 10 cities in Table 4 are also along
the coast. Some customers are from foreign countries. Further analysis could explore if user search behavior
and Taobao’s advertising strategy show marked difference across geographical regions. If user behavior is
dependent on geographical status then that information could be made part of the overall customization of
advertising strategy.
Click Log File
The second data file contains click logs between 2 pm and 11:41 pm on May 15, 2013. There are 312,660
observations and 90 variables. In Table 5 the variables relevant to this study are described.
The variable fromdomainname records the origin or URL from which the advertisement originated.
The frequency of this variable is shown in Table 6. Over 43% of the clicks originated from Taobao’s main
search engine s.taobao.com. The next frequent domain is etao.com, a vertical shopping search engine in China
once powered by Microsoft’s Bing search engine. The third domain is list.taobao.com, which is market list by
category of products. The rest of the domains are other specialized shopping websites run by Taobao. Clearly,

Taobao’s main search engine brings the most number of clicks (43.81% of all clicks). The observed pattern of
advertisement origin shows that most of the advertisements originate from Taobao owned or controlled
affiliates. There is apparently a high degree of cross selling through advertisements across the platforms as a
result. The strategy, it seems, is to use the various websites under the Taobao control umbrella to channel user
clicks to the main e-commerce platform of Taobao.
Frequencies of the number of advertisements per page (adsperpage) are listed in Table 7 ordered from
the most frequent to the least frequent. Close to half (44.31%) of the clicked advertisements are part of 13-
advertisement combination per page (8 advertisements on the right and 5 advertisements at the bottom). Over
12% of the clicked advertisements are part of 24-advertisements combination per page. Other values of
adsperpage correspond to far fewer number of clicks. It is interesting to note that almost half (44.31%) of the
users who clicked an advertisement were shown the standard 13 advertisement combination per page. It appears
that the platform is not making a more customized approach to user characteristics that would in turn result in a
wider range of the number of advertisements than that is observed.
Tables 8 and 9 list the top cities and provinces the customers who clicked the advertisements are
located. As is in the page view log file, most customers are located in coastal provinces and cities. As noted
earlier, further analysis could explore if user search behavior and Taobao’s advertising strategy show marked
difference across geographical regions. If user behavior is dependent on geographical status then that
information could be made part of the overall customization of advertising strategy.
Table 10 lists the frequencies of the relative position (the variable relativeposition) of the clicked
advertisements on a page. This variable assumes integer values between 1 and 75, values corresponding to
relative positions of the advertisement compared to other advertisements on the webpage, as designated by
Taobao on its search result webpage. Based on the frequencies in Table 10, it is clear that the relative position
of 1 (the top position on the right) corresponds to the most prominent or desirable position on the webpage as it
corresponds to the greatest number of clicks. There is clear correlation between the position of advertisement
and the frequency of clicks. As the relative position increases, the corresponding number of clicks decreases,
clearly a result of less prominent positions on the web page. This result demonstrates that specific positions of
advertisements on the webpage are instrumental in influencing the number of clicks, a matter of great interest
for advertisers who want to maximize click-through revenues.
IV. Conclusion
In conclusion, the data analysis presented in this paper is one of the first in its kind to reveal the search based
adverting strategy of Taobao, based on actual transactional data from Taobao. China has emerged as the world’s
largest e-commerce market, and the analytical sophistication employed in monetizing customer search in the e-
commerce realm will only increase and this paper contributes in some measure to our understanding of this
important market. We have discussed the potential avenues to explore additional insights based on the analysis
existing data as well as with future data from Taobao. This paper presents a rather preliminary analysis of field
data from the Taobao platform, which by itself is an important contribution because of the paucity of
transactional data from actual e-commerce platforms made accessible to researchers and to the public. The
purpose is to highlight the potential research avenues that could be explored, which we feel would be immensely
relevant to emerging market businesses engaged in some form or other in following the Taobao’s business
model.

V. Figures and Tables
Figure 1. Histogram of the variable adprice, the price of the merchandise displayed in the advertisements.
Figure 2. Histogram of the variable adhighestprice
Figure 3. Histogram of the variable adscoretag, the number of clicks received by the advertisements.

Figure 4. Histogram of the percentage by which adprice is marked down from adhighestprice.
Table 1. Description of variables in the Page View Log file
Variable Name Description of Variable
acookie A variable identifying the customer on the website.
adhighestprice The highest price of the advertised product from which the price is marked
down.
adprice The price of the advertised product (the marked down price from
adhighestprice).
adrelativepostion The position of the ad on the page ranging from 1 to 75 with 1 indicating the
best position.
adscoretag The number of times an ad has been clicked.
adsnumperpage The number of advertisements per page.
bidword The keyword a merchant placed the highest bid on.
city The city in which the customer is located.
province The province in which the customer is located.
pvtime Time when page view took place; used to calculate time until click.
rawquery_keyword The keyword a customer searches for on Taobao.com. The words for this
variable contain the bidword. A search for these keywords activates
advertisements for display. If the customer clicks on any of these
advertisements, Taobao gets paid.
sessionid The ID of each session of page view. Sponsored advertisements displayed on
the right after a search are considered as part of one distinct session ID while
sponsored advertisements displayed at the bottom from this search are
considered part of another distinct session ID. It can be used to link to the ID
in the click log files (see Section 2.2) variable, pvid. This variable was
generated randomly with 32 alphanumeric characters in length.

Table 2. Frequency of the number of advertisements per page (adsnumperpage), ordered from the most frequent
to the least frequent.
adsnumperpage Frequency Percent Cumulative Frequency Cumulative Percent
13 13827 37.06 13827 37.06
3 6834 18.32 20661 55.38
16 4099 10.99 24760 66.36
5 3130 8.39 27890 74.75
6 1902 5.10 29792 79.85
24 1771 4.75 31563 84.60
32 1320 3.54 32883 88.13
20 1045 2.80 33928 90.94
27 727 1.95 34655 92.88
1 685 1.84 35340 94.72
12 645 1.73 35985 96.45
7 451 1.21 36436 97.66
4 179 0.48 36615 98.14
15 148 0.40 36763 98.53
8 143 0.38 36906 98.92
40 137 0.37 37043 99.28
10 91 0.24 37134 99.53
30 78 0.21 37212 99.74
2 35 0.09 37247 99.83
50 23 0.06 37270 99.89
60 18 0.05 37288 99.94
9 11 0.03 37299 99.97
11 11 0.03 37310 100.00
13 13827 37.06 13827 37.06
Table 3. Frequencies of the province origin of the customers, top 10 provinces.
province Frequency Percent Cumulative Frequency Cumulative Percent
Guangdong 3042 11.3 3042 11.3
Zhejiang 2604 9.67 5646 20.97
Jiangshu 2424 9 8070 29.98
Shandong 1697 6.3 9767 36.28
Beijing 1665 6.19 11432 42.47
Shanghai 1313 4.88 12745 47.35
Sicuan 1049 3.9 13794 51.24
Henan 990 3.68 14784 54.92
Hebei 989 3.67 15773 58.59
Fujian 968 3.6 16741 62.19

Table 4. Frequencies of the city origin of the customers, top 10 cities.
city Frequency Percent Cumulative Frequency
Cumulative
Percent
Beijing 1665 6.19 1665 6.19
Shanghai 1313 4.88 2978 11.06
Guangzhou 798 2.96 3776 14.03
Hangzhou 751 2.79 4527 16.82
Shenzhen 715 2.66 5242 19.47
Foreign Countries 624 2.32 5866 21.79
Chengdu 549 2.04 6415 23.83
Shuzhou 548 2.04 6963 25.87
Chongqing 500 1.86 7463 27.72
Tianjin 487 1.81 7950 29.53
Table 5. Description of variables in the click log file and a sample of fifteen observations (the values under city
and keyword are English translation of Chinese characters).
Variable Name Description of Variable
pvid An identification (ID) variable similar to sessionid in the page view log file. It
can be used to link to the ID in the click file’s variable, sessionid (see Table 1).
This variable was generated randomly with 32 alphanumeric characters in
length. An exact match indicates the customer clicked the ad after viewing the
ad on the page.
clicktime Time when ad was clicked. It can be used to calculate time until click.
clickprice The price of the clicked product.
clickpriceorigin The original price of the clicked product.
highestprice The highest price of the advertised product from which the price is marked
down.
fromdomainname Indicate the part of taobao.com the click originated from. There are a number of
sites of taobao.com, e.g., taobao’s search engine or tmall (online store fronts at
taobao.com).
clickcity The city in which the clicking customer is located.
clickprovince The province in which the clicking customer is located.
keyword The keywords a customer searches for on the site.
customerid ID assigned to merchandise advertisers
clickcookie A variable identifying customers who clicked on the advertisements.
pvtime Time when page view took place; used to calculate time until click.

Table 6. Frequencies of the origin or URL (fromdomainname) of the advertisements, domains with at least one
percent of the total advertisements.
fromdomainname Frequency Percent
Cumulative
Frequency
Cumulative
Percent
s.taobao.com 133306 43.81 133306 43.81
tao.etao.com 51171 16.82 184477 60.63
list.taobao.com 18020 5.92 202497 66.56
re.taobao.com 17212 5.66 219709 72.21
trade.taobao.com 14435 4.74 234144 76.96
s8.taobao.com 14160 4.65 248304 81.61
www.taobao.com 6149 2.02 254453 83.63
list.tmall.com 5870 1.93 260323 85.56
s8.m.taobao.com 4977 1.64 265300 87.2
search.taobao.com 4838 1.59 270138 88.79
favorite.taobao.com 4558 1.5 274696 90.29
rate.taobao.com 3874 1.27 278570 91.56
nvzhuang.taobao.com 3814 1.25 282384 92.81
s.m.taobao.com 3814 1.25 286198 94.07
Table 7. Frequency of the number of advertisements per page (adsperpage) in the click log file, ordered from the
most frequent to the least frequent.
adsperpage Frequency Percent
Cumulative
Frequency
Cumulative
Percent
13 138535 44.31 138535 44.31
24 38406 12.28 176941 56.59
3 21426 6.85 198367 63.44
16 18254 5.84 216621 69.28
27 13457 4.3 230078 73.59
5 10440 3.34 240518 76.93
40 9230 2.95 249748 79.88
20 7352 2.35 257100 82.23
12 7249 2.32 264349 84.55
7 7195 2.3 271544 86.85
6 6996 2.24 278540 89.09
32 6575 2.1 285115 91.19
10 5316 1.7 290431 92.89
1 5250 1.68 295681 94.57
30 3798 1.21 299479 95.78
4 2376 0.76 301855 96.54
15 2260 0.72 304115 97.27
9 2129 0.68 306244 97.95
18 1837 0.59 308081 98.54
8 1745 0.56 309826 99.09
11 1303 0.42 311129 99.51
N 433 0.14 311562 99.65
60 365 0.12 311927 99.77
14 354 0.11 312281 99.88
2 243 0.08 312524 99.96
50 75 0.02 312599 99.98
55 51 0.02 312650 100
28 6 0 312656 100
100 3 0 312659 100
23 1 0 312660 100

Table 8. Frequencies of the province origin of the customers, top 13 provinces.
clickprovince Frequency Percent
Cumulative
Frequency
Cumulative
Percent
Guangdong 32052 10.25 32052 10.25
Jiangshu 30204 9.66 62256 19.91
Zhejiang 30149 9.64 92405 29.55
Shandong 21800 6.97 114205 36.53
Beijing 16279 5.21 130484 41.73
Shanghai 14086 4.51 144570 46.24
Henan 13122 4.2 157692 50.44
Sicuan 12510 4 170202 54.44
Hebei 12456 3.98 182658 58.42
Liaoning 11659 3.73 194317 62.15
Hubei 11148 3.57 205465 65.72
Fujian 10899 3.49 216364 69.2
Anhui 9913 3.17 226277 72.37
Table 9. Frequencies of the city origin of the customers, top 10 cities.
clickcity Frequency Percent
Cumulative
Frequency
Cumulative
Percent
Beijing 16279 5.21 16279 5.21
Shanghai 14086 4.51 30365 9.71
Hangzhou 8359 2.67 38724 12.39
Shenzhen 7496 2.4 46220 14.78
Guangzhou 7159 2.29 53379 17.07
Shuzhou 6445 2.06 59824 19.13
Tianjin 6227 1.99 66051 21.13
Chengdu 5852 1.87 71903 23
Chongqing 5657 1.81 77560 24.81
Nanjing 4927 1.58 82487 26.38
Table 10. Frequencies of the relative positions (relativeposition) of the clicked advertisements on a page, top 16
positions.
relativeposition Frequency Percent Cumulative Frequency Cumulative Percent
1 42904 13.72 42904 13.72
2 38422 12.29 81326 26.01
3 34769 11.12 116095 37.13
4 22854 7.31 138949 44.44
5 19331 6.18 158280 50.62
6 17033 5.45 175313 56.07
7 15403 4.93 190716 61
10 15187 4.86 205903 65.86
11 15048 4.81 220951 70.67
9 14413 4.61 235364 75.28
8 13715 4.39 249079 79.66
12 13294 4.25 262373 83.92
13 11620 3.72 273993 87.63
14 4235 1.35 278228 88.99
15 3911 1.25 282139 90.24
16 3663 1.17 285802 91.41

References
[1] Graepel, T., Candela, J., Borchert, T., Herbrich, R., Web-Scale Bayesian Click-Through Rate
Prediction for Sponsored Search Advertising in Microsoft’s Bing Search Engine. In: Proceedings of the
27th
International Conference on Machine Learning, 2010.
[2] Svensen, M., Xu, Q., Stern, D., Hanks, S., Bishop, C., “Broad vs Narrow: Modelling Strategies for
Online Behavioural Targeting,” in Proceedings of the Fifth International Workshop on Data
Mining and Audience Intelligence for Advertising (ADKDD), San Diego, USA, ACM Press, 2011.
[3] Richards, J., and Li, Min., The Chinese E-commerce Search Advertising Business: A Case Study of
Taobao. Contemporary Management research, 14(2), 2018, 121-142.
[4] The Alibaba Phenomenon; E-commerce in china. (2013, Mar 23). The Economist, 406, 15.
[5] Wei, D., Geng, P., Ying, L., & Shuaipeng, L. (2014, May 31 2014-June 2 2014). A prediction study on
e-commerce sales based on structure time series model and web search data. Paper presented at the
26th Chinese Control and Decision Conference (2014 CCDC).

Profile of User Search Behavior and Advertising on Alibaba’s Taobao Platform

More Related Content

What's hot (20)

Similar to Profile of User Search Behavior and Advertising on Alibaba’s Taobao Platform (20)

More from International Journal of Business Marketing and Management (IJBMM) (20)

Recently uploaded (20)

Profile of User Search Behavior and Advertising on Alibaba’s Taobao Platform