This study tackles the problem of advertising content distribution in a liate networks. We model the a liate network as a new business-to-business relationship in which a master site tries to improve its pro ts by properly targeting the advertising information. In our speci c case, we will optimize discount coupons delivery for one of the biggest coupons site in the world. In our study we will employ adapted recommendation strategies based on collaborative ltering methods. The purpose of the present paper is to correlate the display of advertising information on a liate sites with the actual improvements in sales, as a direct method to obtain the user rating which does not exist.
This paper describes the research proposal we are working at and which is part of my PhD thesis. It speci cally targets electronic business by tackling a new and emerging businessto-business relationship. In electronic B2B, a provider has various means of attracting new customers, one of them being to supply and promote discount coupons. A proper platform for spreading out advertising information regarding discounts is required, and the information providers should possess intelligent tools to observe and evaluate the e ciency of the discounts information distribution. A liate programs represent possible solution to be adopted. In e-commerce, an a liate is a website which links back to an e-commerce site like Amazon.com. While in classical e-commerce, the provider can easily compute the e ciency and the e ectiveness of a particular a liate, because it can register all products sold through that a liate, in discount information advertising this operation is no longer possible. In discount advertising, the customer usually buys the product directly from the provider e-commerce platform and the provider has no mean to directly identify how the customer acquired the information about the discount. Thus, classical tools for personalized content delivery can not be directly applied for optimizing the advertising delivery information, due to the fact that the direct feedback loop is missing.
In this paper we will give a brief description of the research problem stressed in the paragraph above, stating out the challenges we are facing. The paper is organized as follows. Section 2 presents the research problem of my PhD thesis. In section 3 we present the state of the art, mainly concerning the intelligent techniques for content delivery and studies concerning the a liates problem. Section 4 describes the main challenges we face in our study, while section 5 presents the wow-coupons.com use case. We conclude the paper by presenting our future plans.
Our main research problem is to build an e-business system in the area of a liates marketing and content delivery. Speci cally, we try to develop a delivery system for a major coupons site. Coupons represent a well-established way of providing bene ts to customers for a speci c activity to improve their business.
Online coupons or online coupon codes are discounts and bargain deals for all major on line stores and shopping sites on the Internet. Online stores creates discount codes to select groups of customers. By using coupon codes some groups of people will enjoy discounts when shopping online, like free shipping. Discount coupons is a way to save with on line shopping and some codes will be automatically applied at checkout. Other codes must be copy pasted intro a special coupon text box before to con rm the order.
An independent research agency made a recent survey on
behalf of Pay By Touch on over 100 shoppers that were
questioned in May 2007. The survey proves [
In our study case we want to speci cally develop and implement a content delivery system for one of the top coupons site in USA (http://wow-coupons.com/). From now on, we will name this site as the master site. Our system will search, select and deliver coupons on behalf of 3'rd party sites that sell or recommend products, or for community sites that offer valuable user targeted content. Delivery will be targeted to a network of a liates, established in order to enhance information dissemination. Delivered content will assist the site visitors, the community members in their online shopping activities and will help them to save money when they buy products. Our main target is to optimize the system such us to deliver highly targeted discount coupons. For our target, we intend to adapt recommendation systems and collaborative ltering technologies. Another feature of our interest is to technically enable the coupons site to deliver the content in real-time, i.e. without the usage of o -line information exchange with the a liates. When a new coupon is added on the master site, the content delivered on the a liate sites should be seamless updated.
Collaborative ltering is of our interest because it is a way
to establish what items to display to web users who have
browsed some ads or made purchases in the past.
Collaborative ltering software compiles purchasing information on
customers to pool them into clusters and uses some cluster
members' purchasing patterns to predict the buying habits
of others in the same cluster. It does this in real time and,
for instance, puts an ad on the customer's screen while he
or she is making a purchase [
To directly apply collaborative ltering, we need to collect and represent di erent partner sites and people preferences. Preferences are typically based on item ratings (i.e. posteriori feedback) explicitly delivered by users. The system recommends products which were evaluated positively by another similar user or by a set of such users.
Our main challenge is that we cannot collect the user ratings or preferences because we deliver the advertising content (coupon codes). The coupon code for a particular product and promotion campaign is the same, regardless the site that publishes it. When a user applies a coupon by buying a product, we do not have a direct method to identify the a liate site that published the coupon to the user. Therefore, in our business setup, we do not have the feedback loop mechanism to allow us to use a classical collaborative ltering method.
Instead, we want to implement a module part of the master site system that will allow it build a pro le for each a liate. This module will track down and monitor how many times an a liate user viewed and clicked on every coupon displayed. On the other side, the master knows the number of products sold per every coupon delivered to all a liates. All these will aggregate in a statistical information that will replace the classical feedback from the partners and will provide the a liate pro le. The master can optimize and improve the coupons delivery algorithm for the entire a liates network.
In this section we described our research problem in the terms of our B2B relationships. But, the problem is more general, spanning over all business relationships where direct feedback or user rating is not available. By tackling the a liates problem, we intend to prove that intelligent techniques for content information delivery are more suited and can enhance the pro t of the advertisers, even if there are only statistical glues about the real pro le of the information users.
In this section we will investigate the state of the art in the elds covered by our research. We should emphasize that there is a lot of bibliography tackling recommender systems and collaborative ltering. Regarding the a liates, there are a lot of business studies concerning the a liates in e-commerce setups. In e-commerce setups, the feedback loop is closed because the master can identify through which channel a speci c purchase was performed. Therefore, besides a proper advertising, one of their biggest challenge is how to compensate or reward each a liate, accordingly to the pro ts they generated. In our study we are not interested in designing a proper reward scheme for a liates, we suppose that we have good mechanisms to maintain the network and make it growing.
Recommender systems are an important part of recent
ecommerce. They enable the increase of sales by suggesting
to users selected products on o er. The problem of how to
choose the most suitable items, possibly with respect to the
user's inclinations, is a challenging research problem that
has been investigated for many years [
Four fundamental approaches to recommendation can be
mentioned: demographic ltering, collaborative and
contentbased recommendation, and simpli ed statistical approaches
[
In demographic recommendation, users are classi ed based on their personal data, which was collected during the registration process, survey responses or other feedback methods. Each product is assigned to one or more classes with certain weights and the user is attracted to items from the class closest to their pro le. This is attribute based recommendation.
Collaborative recommendation is typically based on item ratings explicitly delivered by users. The system recommends products, which have been evaluated positively by another similar user or by a set of such users, whose ratings have the strongest correlation with the current user. This is user-touser correlation.
Content-based recommendation focuses on the similarity between products, usually taking into account their features like textual descriptions, hyperlinks, related ratings, or cooccurrence in the same purchased transactions or web user sessions. Items that are the closest to the most recently processed (viewed), are recommended regardless of user preferences. This is item-to-item correlation. Association rules and sequential patterns are the most interesting techniques used in recommendation based on item-to-item correlation. They are usually applied to data sets related to items such as purchases, ratings of TV programs, navigation paths rather than directly to item attributes.
In the statistical approach, the user is shown products based
on some statistical factors; usually popularity measures like
averages or summary ratings (the best rated), and numbers
of sold units (the best buy) [
The information overload problem a ects our everyday
experience while searching for knowledge on a topic. To
overcome this problem, we often rely on suggestions from others
who have more experience on the topic. However, in web
case where there are numerous suggestions, it is not easy to
detect the trustworthy ones. Shifting from individual to
collective suggestions, the process of recommendation becomes
controllable. This is attained with the introduction of
Collaborative Filtering (CF), which provides recommendations
based on the suggestions of users who have similar
preferences. Since CF is able to capture the particular preferences
of a user, it has become one of the most popular methods in
recommender systems [
A web site or other online service that receives extensive
tra c has the potential to analyze the resulting usage data
for the bene t of its user population. One of the most
common applications of such analysis is collaborative ltering.
A web site o ering items for sale or download can analyze
the aggregate decisions of the whole population, and then
make recommendations to individual users of further items
that they are likely to be interested in. The
recommendations made to a speci c user are thus based not just on his
or her own previous actions, but also on collaborative
information, the information collected from other users in the
system [
Collaborative ltering algorithms can be categorized as [
Both practical experience and related research have reported
that memory-based algorithms present excellent performance,
in terms of accuracy, for multivalue rating data. On the
other hand, model-based algorithms are e ciently handle
scalability to large data sets [
A liate marketing is a web-based marketing practice in which a business rewards one or more a liates for each visitor or customer brought about by the a liate's marketing e orts. A merchant, also known as an advertiser or retailer, is a web site or company that sells a product or service online, accepts payments and ful lls orders. A liates (also called publishers) place merchants' ads, text links, or product links on their web sites, shopping engines, blogs, etc. or include them in e-mail campaigns and search listings in exchange for commissions on leads or sales.
We expect to contribute in several main areas.
From the systemic point of view, we intend to de ne the overall picture comprising all business to business relationships. Figure 1 presents an overall sketch of our system. Our contribution will mainly reside in the WOW-Coupons.com master site where we will technically embed a recommender system.
From a systemic point of view, we intend to de ne the overall picture comprising all business to business relationships. Our contribution will mainly reside in the WOW-Coupons.com master site where we will technically embed a recommender system.
From a technical point of view, a challenge is how to pack
the content delivered to a liate sites so as the partners will
seamlessly accept the content and publish it into their
websites. The content should be delivered in such a way that
we will be able to register the number of views per coupon
and number of clicks per coupon. We envisage the usage of
server-controlled web development techniques that exchange
small amount of information behind the scene. AJAX
technologies [
From a algorithmic point of view, our research raises several open issues: to develop a statistical analysis tool in order to relate in a signi cant manner the collected information about content display (number of views and clicks) and the received payments from the earning reports. Statistical analysis will provide a way to collect the required feedback in the collaborative ltering methods to select a proper user pro le management scheme. This is highly related with adapting a recommendation system or collaborative ltering method to our setup. Up to now, we only investigated several alternatives, but we did not established which algorithm would be more suited for us. We expect the ICEC doctoral symposium to allow us to progress on this issue, by having discussions with highly esteemed experts to compare the intelligent content distribution to the actual functioning of the a liates system. Actually, there is not too much intelligence in the system, in the sense that the master delivers to each site the coupons related with the latest marketing campaign added in the master's database. In order to allow for a nonbiased comparison, we need to design a controlled experiment on wow-coupons.com, covering a de ned period of time and a stable part of the a liates network. Designing and running the experiments are a big challenge, because, in meantime, we need not to downward the business performance over the period of the experiment.
We started this research because we intend to improve the
way that WOW-Coupons.com delivers the coupons for the
a liate sites. The actual system delivers the last coupons
added on the master site, organized by store and by category.
Up to now, this delivery scheme proved to be winning, but
in nowadays this module needs some improvements because
we intend to enhance the user usage of time and also the
pro ts wow-coupons.com makes from the a liates network.
We will try to do a brief description of the site that will use
the recommender system that we want to provide below.
The overwhelming majority of consumers in the US collect
coupons to help them make the most of their money.
Consumers spend plenty during the holiday season, and with
tightening budgets, they are spending money where they see
the best deals. During the holidays, an impressive 71% of
survey respondents [
Printable retail coupons section will provide real coupons buyers can take to favorite major retailers and national franchise stores. Shoppers choose the discount, print it out, and go shopping without needing to sign up. The process eliminates clogged inboxes with endless promotional e-mails from dozens of mailing lists. Grocery printable vouchers section will provide the grocery coupons users need, when they need them. Print coupons to save on favorite brands at supermarkets and drugstores everywhere. Restaurant printable vouchers section will provide coupons that can be printed out and used to save money at restaurants across the nation. When eating out may seem like an extravagance, think again and enjoy dinner. Travel and entertainment printable vouchers section will provide discounts and vouchers for diverse traveling destinations whether in US, U.K. or internationally. Travelers can still a ord to take vacations or visit the family. The site has printable travel coupons and o ers for car rentals, accommodations, amusement parks, museums and lots more.
Just like printable vouchers, the best and biggest retailers and service providers o er online coupons and special discounts. The di erence is, these can be used only for purchases made online. On the site visitors can browse great o ers, go straight to a favorite store (arranged alphabetically or by date posted) to see the latest deals, or to use the navigation menu on the right side of the site. When an item is found, the user can follow directions in the description of each coupon to be sure that savings have been applied before paying for an order.
For implementing the architectural design for coupons
delivery, we employed the XML standard to provide an RSS
[
Our future plans intend to cover the above mentioned issues. We are only in the incipient stage of our research, because I have spent all the rst year in taking exams and now, we worked on de ning a proper problem and researching various technologies for this problem. Up to now, Wowcoupons got implemented without the optimization feature and to perform the optimization mentioned in this paper is a good plan for future research. We also think that this study will represent a good opportunity to advertise to the a liate marketing community the potential of intelligent techniques.
This work is supported by the Romanian Authority for Scienti c Research under project IDEI 573. We also acknowledge the support from Mrs. Elena Potoupa, CEO of WOW Things Inc., owner of wow-coupons.com.