Expert Systems with Applications 31 (2006) 542–550

www.elsevier.com/locate/eswa

Predicting the final prices of online auction items

Li Xuefeng *, Liu Lu, Wu Lihua, Zhang Zhao
School of Economics and Management, BeiHang University, Mailbox 10-42 XueYuan Road No. 37, Beijing 100083, China

Abstract
With the prevalence of the Internet and e-commerce, the online exchange market, especially the online auction market develops very fast. The
activities of online auction produce a large number of transaction data. If utilized properly, these data can be of great benefit to sellers, buyers and
website administrator. Typically, the final price prediction results may help sellers optimize the selling price of their items and auction attributes.
At the same time, part of the information asymmetry problems may be solved for buyers. Thus, transaction time can be shortened and cost can be
saved. In this paper, we collect large amounts of historical exchange data from Eachnet, an online auction website most famous in China and use
machine learning algorithms and traditional statistical methods to forecast the final prices of auction items. We propose an attribute construction
method to overcome the problem that auction bid list changes dynamically. Some experiments are performed and the prediction results are
discussed to verify the proposed solution.
q 2005 Elsevier Ltd. All rights reserved.

Keywords: Online auctions; Price prediction; Machine learning; Neural network

1. Introduction the exchange process. These data contain sufficient information

E-commerce, especially the online auction has experienced
great development for the recent years. Everyday, there are
thousands of users to do their businesses and exchanges on the
Internet. The online auction market has already become an
important business entity. According to estimation, online
exchanges will take 25 percent of all business exchange amount
by 2005. As the online auction website most successful in the
world, eBay was in possession of 135 million register users and
contained 1400 million auction items in its website by the end of
2004. Influenced by the flourish of global online auction market,
the online auction business in China burgeons fast too. Eachnet
(www.ebay.com.cn, Ebay–Eachnet) who once was the biggest
online auction website in China and was purchased by eBay in
2004 witnessed the progress of Chinese online auction market.
According to Iresearch China online auction research report
2004, While the market size will go up to 21 billion Yuan with an
annual average growth rate of 84 percent, the users of Chinese
online auction market will amount to 35 million in 2007 with an
annual average growth rate of 43 percent during the next 3 years.
Different from the traditional auction mode, online auction
can produce a large number of electronic data during
* Corresponding author. Tel.: C86 108 232 4787; fax: C86 108 231 5601.
E-mail address: leexuefeng@263.net (L. Xuefeng).
0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2005.09.077
on products and economic behaviors of actors. If utilized
properly, they will bring sellers, buyers and the website
administrator great benefits.
Online auctions are exchange mechanism for determining
prices of items in the Internet. They connect buyers and sellers
together in ways that were impossible previously. Although
online auctions provide many benefits over traditional
exchange methods, they also have several problems. One of
the main problems observed on online auctions is the winner’s
curse. The winner’s curse is that bidder pays a price higher than
the true value of the item. The cause for the occurrence of
winner’s curse is lack of the information on the part of the
bidder regarding the true vale of the item (Metha & Lee, 1999).
This shows that the information asymmetry, which is caused
either by the lack of expertise or lack of pricing information of
products, still exists in online auction.
Since a number of attributes, e.g. the feedback rating of a
seller, freight and the auction start and end time can vary over
time in the auction situation, it is a challenging task to use
traditional machine learning technique to predict the final
prices of the online auction items. This paper employs the
attributes of the seller and the auction item, the intrinsic
attributes of the auction itself, and the historical exchange data
to predict the final prices of the auction items. By learning, a
stable prediction model is acquired first. Then the prediction is
performed and the prediction results are provided.
Compared with previous work, the contributions of our
study can be summarized as follows: we provide a solution for
 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550
dealing with the time-dependent dynamics of bidding. By
attribute selection and construction, traditional machine
learning algorithms and statistical methods can be used to
predict the final price of the online auction items. We also
conduct a comparison between the prediction qualities of
regressions with neural networks and show that clustering of
data has a measurable effect on prices prediction result.
The rest of this paper is organized as follows. The Section 2
summarizes the related work to our research. In Section 3, we
describe the Ebay–Eachnet’s auction mechanism, and introduce
the collection and preprocessing of exchange data. Construction
and selection of attributes are also presented in detail. Section 4
reports our experimental work. It provides the details of data set,
evaluation metrics, procedure and results of different exper-
iments, as well as the discussion of the results. Section 5
provides some concluding remarks and directions for future
research.
2. Related work
In this section, we briefly present some of research literature
related to data analysis of online auctions and the final price or
the winning price prediction.
Considerable work that applied traditional techniques to the
online auction analysis has been made in the economics domain.
In the data mining and information systems field, there also
existed some researches on online auction. For example, Bapna
found that significant heterogeneity exists among the users of
electronic markets like Ebay and developed a stable taxonomy of
biding behavior in online auctions (Bapna et al., 2004). Bajari and
Hortacsu (2003) explored the determinants of bidder and seller
behavior. The winner’s curse is that bidder pays a price higher
than the true value of the item. Lucking-Reiley analyzed the effect
of various Ebay features on the final price of auctions
(Lucking-Reiley, Bryan, Prasad, & Reeves, 2000). They found
that seller’s feedback ratings have an important effect on his
auction prices, with negative comments having a much greater
effect than positive comments. However, these researches mainly
focused on the historical exchange data analysis. They only
described the past auction and did not involve the price prediction
(Shah, Joshi, Sureka & Wurman, 2003). As for the researches to
use the machine learning techniques to predict the final price of
the online auction items, fewer can be found.
There has been some work on price prediction of items in
online markets, e.g. airlines fares (Oren Etzioni, Rattapoom
Tuchinda, Craig A. Knoblock, & Alexander Yates, 2003).
They explored how to predict ticket purchase price through
mining airfare data to minimize ticket purchase price. In the AI
community, literature on time-dependent product price
predition was fewer. The trading agent competition (TAC)
(Wellman, Reeves, Lochner, & Vorobeychik, 2002; 2004) was
the only work that involved implicitly prices prediction in
auctions. TAC relied on a simulator of airline, hotel, and ticket
prices and the competitors built agents to bid on these. TAC
simulated flight prices using a stochastic process that followed
a random walk with an increasingly upward bias. In addition,
the TAC auction of airline tickets assumed that the supply of
543
airline tickets is unlimited. Several TAC competitors have
explored a range of methods for price prediction including
historical averaging, neural nets, and boosting. In the TAC, all
of work was performed with artificially generated data and did
not use any real auction data.
In this paper, we focus on solving this problem on how to
overcome the dynamics of bidding which changes drastically
over time. On the basis of data collection, construction and
selection of attributes, prices prediction is done.
3. Ebay–Eachnet’s auction mechanism, data collection
and preprocessing
3.1. Ebay–Eachnet’s auction mechanism
Ebay–Eachnet is the largest and most successful consumer–
consumer auction site in China. It has possessed of 11.6 million
register users at the end of the first quarter in 2005. The
exchange amounts exceeded 2500 million Yuan in 2004.
Similar to Ebay, Ebay–Eachnet runs a second price English
auction for a variety of consumer goods. The auction is ‘second
price’ because the winner pays the next highest bid, and they
are ‘English auctions’ because bids are ascending. The seller
can decide to have an auction last for 1, 3, 5, 7 or 10 days.
When a potential bidder views a current auction on EBay–
Eachnet, they see a description of the item for sale. This
description is up to the seller’s discretion, but is usually fairly
detailed and may include a picture of the item. The bidder also
can review the recent history of the seller and ratings by
previous customers. Most importantly, the bidder observes the
current high bid for the item. They also observe information on
other bidders for that auction including when they bid, but not
what their bid was. Lastly, all bidders know the exact time
remained before the close of the auction.
After observing this information, the bidder may decide to
bid on the item. The amount entered by the bidder is actually a
‘proxy’ bid. That is, EBay–Eachnet will take that bid and
automatically bid slightly above the next highest bidder up to
the amount entered. This system allows the bidder to leave the
auction, without having to worry about being outbidden, as the
proxy system will keep on bidding automatically. However, if a
bidder is outbidding, then Ebay–Eachnet can notify them by
email. The winner of the auction is the highest bidder at the end
of the auction. The winner pays an amount slightly above the
next highest bid.
In addition, Ebay–Eachnet also runs ‘buy it now’ and
‘English auctionCbuy it now’. In this paper, only English
auction is studied.
3.2. Data collection
A lot of information on Ebay–Eachnet auctions is publicly
available. Ebay–Eachnet posts on its website the complete bid
histories of closed auctions for at least one month after the ending
date. We used a specially designed data collection program to
gather the exchange data from Eachnet. The collection program
first constructs URL dynamically according to product ID. Then it
544 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550

Table 1
A valid auction record

Seller ID ST ET SC HI CS DC DW PW LB BT FP
5844742 2004-1-27 2004-2-3 5 2 1 1 1 2 3 56 ¥32
17:00:00 17:00:00

Table 2
The bid process of a valid auction

Bid number Bidder ID BC BA DE AC Bid time

1 17395264
2 17395264
3 17395264
4 13322974
5 17742581
6 17742581
7 17742581
8 17742581
9 17742581
10 17395264
3 ¥4.00 1 0 2004-2-1 15:35:00
3 ¥12.00 1 0 2004-2-1 15:36:00
3 ¥21.00 1 0 2004-2-1 15:36:00
2 ¥21.00 1 0 2004-1-30 15:33:00
2 ¥23.00 1 0 2004-2-1 16:08:00
2 ¥25.00 1 0 2004-2-1 16:08:03
2 ¥27.00 1 0 2004-2-1 16:09:00
2 ¥29.00 1 0 2004-2-1 19:14:00
2 ¥31.00 1 0 2004-2-1 20:38:00
3 ¥32.00 1 1 2004-2-2 23:43:00

sends query request and searches for the Web page that includes
keyword ‘exchanged’. Finally, the page is parsed to extract valid
data. The acquired valid data are stored in the relation database.
The resulting database covers 36366 valid auction records and
154158 valid bid records from Eachnet during January 27, 2004
to May 20, 2004. Each auction record consists of information of
the seller and the auction itself. Moreover, each bid record
consists of the information of the buyer and the whole bid process.
Tables 1 and 2 show a valid auction and its bid records. Table 3
shows the meanings of each field.
In this research, all Nokia mobile telephone data is extracted
with no differentiating model and new degree from
the collected dataset, which may verify clustering of data
having effect on prediction result. The extracted dataset covers
578 valid auction records and 3547 valid bid records from
January 27, 2004 to May 16, 2004.
3.3. Data preprocessing
Data preprocessing is a very important task to price
prediction. It involves the following steps.
1. Removing ‘buy it now’ auction records
There is a part of ‘buy it now’ auction data in our
dataset. Since we are only interested in English auction in
this study, so we remove this part of data and obtain a
resulting experiment dataset consisting of 535 English
auction records.
2. Dealing with the missing value
Since there is no the missing value in our dataset, this step is
skipped.
3. Data transformation
For improving learning speed and making data in the
comparable level, data transformation is done. The data
transformation technique used is Z-score as follows:
A


v0Z vKsA , where v is the original value, A and sA is the
average value and standard error of attribute A, respectively.
For instance, Fig. 1 shows the distribution of the final prices
before and after transformation.

Table 3
The meanings of each field in the Tables 1 and 2

Table 1 Table 2
Field name The meanings of field Field name The meanings of field
ST The start time of auction Bidder ID ID of bidder
ET The end time of auction BC The credibility of bidders
SC The credibility of sellers BA The bid amount
HI Whether there is invoice or not DE The demand
CS Whether there is customer service or not AC The acquirement
DC Who charges the shipping charges Bid time The bid time
SW The shipping way
PW The payment way
LB The level of limited buyer credit
BT The bowered times of this web page
FP The final price
 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550
Fig. 1. The distribution of the final prices before (left) and after transformation (right).
3.4. Construction and selection of attributes
3.4.1. Construction of attributes
To make the discussion more formal, consider an auction, k
on Ebay–Eachnet, 1%k%535 in our dataset. Let A be the set of
all auction records, AZ{ak}. Let B be the set of all bid records,
BZ{bk}. Obviously, there is one to many relationship between
ak and bk, see Fig. 2.
Since the bid record number of each auction is variable, for
effective learning, the preprocessing work must be carried out
so that there is only one unique bid process to correspond to
each auction. For using machine-learning algorithm to predict
the final price, the relationship between ak and bk must be
changed one to one. Let J be the set of attributes of bid records,
and bkj, j2J, denotes the cell value in the bid records related to
auction k. Four statistics of some attributes are computed so
that one to many relationships is changed to one to one. See
Table 4. DE and AC are omitted. Finally, 12 attributes are
constructed.
Thus, Tables 1 and 2 may be denoted by Table 5.
3.4.2. Selection of attributes
Let D be the prediction attributes set, S be the seller
attributes set, and A be the intrinsic attributes set of auction, C
is the constructed attributes set, then DZSgAgC.
Fig. 2. One to many relationship between akand bk.
545
SZ{SC}, SC denotes the creditability of the seller.
AZ{ST, ET, HI, CS, DC, SW, PW, LB, BT}, for the
meanings of each attribute, please see Table 3.
( )
MINBC; MAXBC; AVGBC; STDEVBC; MINBA; MAXBA;
CZ ;
AVGBA; STDEVBA; MINBT; MAXBT; DT; BIDCOUNT
for the meanings of each attribute, please see Table 4.
The selection of prediction attributes is one of the
important considerations for predicting the final price. In
this paper, to select the independent variables, stepwise
regression and discriminant analysis are used. Through the
analysis, the following some attributes are omitted, namely,
{ET, SW, PW, LB, MINBC, MINBT}. To our surprise, two
analysis methods both make attribute SC removed. However,
according to the research of Lucking-Reiley et al. (2000),
seller’s creditability has an important effect on his auction
prices, so SC is kept back. Let D 0 denote the final
independent variables set, then
D0 Z DKfET; SW; PW; LB; MINBC; MINBTg:
We define the dataset corresponding to D 0 as NOKIA.
According to the foregoing statements, we have a
formulization of prediction of the final price:
546 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550

Table 4
The computing of statistic and construction of attributes

Attribute name Statistic (constructed attributes) Description

BC MINBCZMin (bkj), jZBC Minimum of bidders’ creditability
MAXBCZMax (bkj), jZBC Maximum of bidders’ creditability
AVGBCZAVG (bkj), jZBC Mean of bidders’ creditability
STDEVBCZSTDEV (bkj), jZBC Standard error of bidders’ creditability
BA MINBAZMin (bkj), jZBA Minimum of bid amount
MAXBAZMax (bkj), jZBA Maximum of bid amount
AVGBAZAVG (bkj), jZBA Mean of bid amount
STDEVBAZSTDEV (bkj), jZBA Standard error of bid amount
Bid time MINBTZMin (bkj), jZbid time Minimum of bid time
MAXBTZMax (bkj), jZbid time Maximum of bid time
DTZMAXBT–MINBT Duration of auction (unit is second)
BIDCOUNTZCOUNT (bk) Bid counts

Table 5
A changed auction

Seller ID ST MINBC MAXBC AVGBC STDEVBC BIDCOUNT

5844742 2004-1-27 17:00:00 4 32 22.5 8.7464 10

4. The final price prediction
According to the attributes that were described in Section 3,
the task now is to predict the final price of an auction item. This
problem can be tackled with machine learning algorithms in
several ways. We define the problem in two ways to compare
the relative merits of each approach.
1. Continuous price prediction
We treat the price prediction task as a regression task and
use the training data to learn regression coefficients. We use
multivariate regression (MR) to predict a continuous price
value.
MR is a widely analysis technique owing to its ease of use
and intuitive theoretical basis (Mosteller & Tukey, 1977; Flury
& Riedwyl, 1988). MR involves fitting a parametric function
model to a set of data. In this sense, it is a form of inductive
supervised learning.
2. Discrete price prediction
The discrete price prediction actually is a multi-class
classification task. We discretize the final price into nine
intervals, as follows:
(0, 500], (500, 1000], (1000, 1500], (1500, 2000], (2000,
2500], (2500, 3000], (3000, 3500], (3500, 4000], (4000, 4500],
denoted by 1–9, respectively.
Thus, the price prediction can be regarded as a multi-class
classification problem. In this case, the output is a 500 range (?)
addition, we also present that clustering data has an important
effect on price prediction. Our experimental dataset is NOKIA,
described in previous section. It consists of 535 valid auction
records.
4.1. Evaluation metrics
4.1.1. Continuous price prediction
The basic metric we use to measure this performance is the
residual value oi, which is defined by oi Z yi Ky^i , where yi is
the actual function or dependent feature value for case i and y^i
is the estimate value generated by the regression model. This is
similar to the residual value calculated in classical parametric
regression but has the important difference since the examples
used are from out-of-sample data and not the examples used to
build the regression model. The average residual value over the
test data set is defined by
1X n
o
Z ðy Ky^i Þ:
n iZ1 i
The standard deviation of the residual values provides a
notion of the distribution of the residuals about the mean. As
we are interested only in relative performance of the model, the
standard deviations of residual values for a set of competing
models are normalized such that the maximum standard
deviation is one. The standard deviation of the residual values
is denoted by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 X n

instead of the specific price. We use logistic regression and so Z ðo Ko
i Þ:

neural network to implement the multi-class classification in
this paper.
In our regression and neural network experiments, the tools
used are, respectively, SPSS 13 and Matlab 6.5.
In this section, we present our experimental results of
continuous price prediction and discrete price prediction. In
nK1 iZ1 i
In addition, square of correlation coefficient R, namely
R-square or R2 is interesting. It is denoted by (SSR/SST), where
SSR is regression sum of squares, SST is total sum of squares.
Value of R2 is from zero to one and it shows that the proportion
is interpreted by the independent variables in all differentiation
 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550
Fig. 3. Histogram (left) and normal P–P plot of regression standardized residual (right).
of dependent variable through the regression relationship. In
practice, adjusted R square is used more often.
4.1.2. Discrete price prediction
Since discrete price prediction is a classification problem,
we use accuracy metric to judge the performance of our
experiments. For a set of test examples, accuracy metric is
defined as the number of correct predictions over the total
number of examples.
the number of correct predictions
accuracy Z !100%
the total number of examples
4.2. Experimental results
In all experiments, the dataset NOKIA was randomly split
into a training set comprising 70% of examples, and a test set
comprising 30% of examples by holdout method. Through
repeating 10 times, 10 training sets and corresponding test sets
were produced. Finally, 10 times of simulation were
performed.
4.2.1. Multivariate regression
The experimental results are shown in Figs. 3 and 4. In
Fig. 3, left side and right side denote the histogram and
Fig. 4. R-square (left) and standard deviation of the residual (right).
547
normal P–P plot of regression-standardized residual, respect-
ively, after certain training. As we can see from Fig. 3,
distribution of regression standardize residual is normal on
the whole. In Fig. 4, X-coordinate denotes the learning times
using 10 training sets and corresponding test sets,
Y-coordinate denotes adjusted R-square and standard devi-
ation of the residual in the left and right, respectively. The
average value of R-square is 0.7944, and the average value
of standard deviation of the residual is 0.3644. From the
results, model fit degree is not bad, but the residual is bigger.
Therefore, the result of continuous price prediction is not
good relatively. In fact, as we can see from Fig. 1, the
distribution of the final price is skewed, even after
transformation. This result might be caused by this fact.
4.2.2. Logistic regression and neural network
In the following experiments, logistic regression and BP
network method are used to classify the final price interval.
1. Logistic regression
Logistic regression is part of a category of statistical models
called generalized linear models. It allows one to predict a
discrete outcome, such as group membership, from a set of
variables that may be continuous, discrete, dichotomous, or a
mix of any of these.
548 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550
Fig. 5. The choice of the hidden nodes.
Fig. 6. Accuracy with BP and LR.
2. BP neural network
The back-propagation (BP) neural network, a feed-forward
multi-layer network based on back-propagation algorithm
developed by D. E. Rumelhart and J. L. McCelland in 1986,
has become one of the most widely used ANNs in practice. The
activation transfer function (ATF) of a BP network, usually, is
Fig. 7. The distributions of the final prices cluster 1 (left) and cluster 2 (right).
a differentiable sigmoid (S-shape) function, which helps to
apply the nonlinear mapping from inputs to outputs.
The structure design for BP network is two-fold. On the one
hand, the layer number of network must be determined. On the
other hand, the node number of each layer must be set.
Generally, the layer number of network is experientially set to
three layers, namely an input layer, a hidden layer and an
output layer. The choice of nodes number for each layer
depends on the practical case. According to the previous
section, here, the number of input node is 16, and the output
node is 1. Since there is no standard theory available, the trial-
and-error method is used in this paper for choosing the hidden
node number. By experiment comparison (illustrated in Fig. 5),
the resulting hidden nodes are set to 30. The transfer function
of the hidden layer and the output layer is tan-Sigmoid and
liner function, respectively.
The experimental results are shown in Fig. 6. As we can see
from Fig. 6, obviously, neural network is better than logistic
regression on classification accuracy. The average value of
accuracy of neural network and logistic regression is 91.29 and
76.46%, respectively. This result shows that neural network
outperforms the traditional statistical method in handling
highly skewed data. We know logistic regression requires a
large number of examples relate to the number of independent
variables. In our experiment, the number of examples is smaller
relatively.
4.3. Clustering effect on prediction result
In this section, firstly, clustering analysis is performed on
the data set NOKIA by k-means method. The number of cluster
is set to 2. The first cluster consists of 325 examples, and the
second cluster consists of 210 examples. The distribution of the
final prices of each cluster is shown in Fig. 7. From the result,
clustering effect is obviously overall. Secondly, discrete price
prediction is carried out for each cluster. The result is
illustrated in Fig. 8.
According to Fig. 8, accuracy of BP network increases
obviously, the average value of accuracy of cluster 1 and
 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550
Fig. 8. Accuracy with BP and LR.
cluster 2 is 94 and 95%, respectively. In contrary to BP
network, accuracy of logistic regression decreases, the average
value of accuracy of cluster 1 and 2 is only 68.4 and 63.4%
respectively.
4.4. Discussion
From experimental results, we have some important
observations.
The effect of continuous price prediction is not good by
contrast with discrete price prediction. This result verifies the
proved conclusion that multivariate regression was unfit to
analyze the skewed data. Further, we think that discrete price
prediction is more significant than continuous price prediction
in practice.
In the discrete price prediction, BP network holds the
promising prediction effect. Of course, the lack of examples
might lead to the experiment result of logistic regression not
good enough. Clustering of data has an important effect on the
prediction results. However, the results are significantly
different as to BP network and logistic regression. It makes
accuracy of the former increases, while that of the latter
decreases.
In fact, the attributes we selected are not all attributes we
can get. For example, the attributes such as whether the auction
item has a picture or not, whether the seller has auction
experience or not, and whether the seller has online shopping
experience or not did not be included. If more attributes are
used to predict, the prediction accuracy might be enhanced
further.
Since only one kind of data is used for prediction, whether
the prediction results can be generalized to other auction items
is unproved in this paper.
5. Summary and future research
By attributes construction, this paper overcomes the
problem that auction bid list changes dynamically. The
preprocessing enables the machine learning algorithms and
traditional statistical analysis methods to be applied to the
final prices prediction. The favorable results are obtained.
549
According to the experimental results, we believe that
machine-learning algorithms, such as BP network, outper-
form traditional statistical models.
In this paper, to avoid the complexity, we do not select some
text and picture attributes, which may affect the prediction
accuracy.
Our future study will be made from two directions. One is to
apply the final price prediction to recommender systems based
on online auction. In addition, we will extend the prediction
model to different product types.
Recommender systems (Resnick, & Varian, 1997;
Schafer, Konstan, & Riedl, 1999) have been widely used
in the B2C e-commerce context, such as Amazon website
and so on. However, their applications under the C2C
environment are fewer. The prediction to the final price of
the auction items enables the online auction website to
suggest the final price of the auction item for the seller.
This will help the seller optimize some auction attributes.
The suggested price may be regarded as bid reference for
the buyers. Then the exchange time might be shortened and
the final price can be maximized. At the same time, the
website also could get maximum profit. As a part of project,
we will place the emphases on the following four concrete
aims in the future.
1. Establish the interest models for the online auction
website users.
2. Study the influence importance of each attribute on the
final price.
3. Study how to enable the recommender systems to be
applied effectively to the online auction.
4. Establish the intelligent recommender systems based on
the online auction.
Acknowledgements
Funding for this research was supported by the National
Science Foundation of China under grants No.70371004 and
PhD program Foundation of Education Ministry of China
contract number 20040006023. We would like to thank
anonymous reviewers for their valuable comments.
550 L. Xuefeng et al. / Expert Systems with Applications 31 (2006) 542–550
References
Bajari, P., & Hortacsu, A. (2003). Winner’s curse, reserve prices, and
endogenous entry: Empirical insights from Ebay auctions. The Rand
Journal of Economics, 3(2), 329–355.
Bapna, R., Goes, P., Gupta, A., & Jin, Y. (2004). User heterogeneity and its
impact on electronic auction market design: An empirical explanation. MIS
Quarterly, 28(1).
Flury, B., & Riedwyl, H. (1988). Multivariate statistics a practical approach.
New York: Chapman & Hall.
Lucking-Reiley, D., Bryan, D., Prasad, N. & Reeves, D. (2000). Pennies from
Ebay: The determinants of price in online auctions. Technical report,
University of Arizona. http://www/vanderbilt/edi/econ/reiley/papers/Pen-
niesFromEBaiy.pdf.
Metha, K., & Lee, B. (1999). An empirical evidence of winner’s curse in
electronic auction. Proceeding of the 20th international conference on
information systems, Charlotte, NC.
Mosteller, F., & Tukey, J. W. (1977). Data analysis and regression. Menlo
Park, CA: Addison -Wesley.
Oren, Etzioni, Rattapoom, Tuchinda, Craig, A. Knoblock, & Alexander, Yates
(2003). To buy or not to buy: Mining airfare data to minimize ticket
purchase price KDD 2003 (pp. 119–128).
Resnick, P., & Varian, H. R. (1997). Recommender systems. Communications
of ACM, 40(3), C56–C58.
Schafer, J. B., Konstan, J. A., & Riedl, J. (1999). Recommender systems in
E-Commerce. Proceedings of the first ACM conference on electronic
commerce, Denver, CO (pp. 158–166).
Shah, H. S., Joshi, N. R., Sureka, A. & Wurman, P. R. (2003). Mining for bidding
strategies on ebay. In Lecture notes in artificial intelligence Springer.
Wellman, M. P., Reeves, D. M., Lochner, K. M., & Vorobeychik, Y. (2002).
Price prediction in a trading agent competition. Technical report,
University of Michigan.
Wellman, M. P., Reeves, D. M., Lochner, K. M., & Vorobeychik, Y. (2004).
Price prediction in a trading agent competition. Journal of Artificial
Intelligence Research, 21, 19–36.