Since the emergence of advertisements, ads are seldom absent on any influential media, like flier, newspaper, radio, television, and there is no exception of the Internet. We can hardly find one website or portal with business power but without hanging advertisement banners on their board in this Internet times. There are 99 percent of all web sites offer standard banner advertisements (Buchwalter et al. 2001). The growth rate of the Internet ads market is still booming in 2006. The IAB Internet Advertising revenue report (2006 second-quarter and first six-month results) shows that the growth rate of the second quarter revenue based on its of first quarter in 2006 is 5.5% (4,061 versus 3,848 million) , and the growth rate of first 6-month revenue of 2006 based on its of 2005 is 36.7% (7,909 versus 5,787 millions). Although the unique price of Internet advertisement has not been yet as high as its of TV advertisement, the latter can be charged thousands of dollars to show on TV for couple seconds and this investment being effective and profitable cause the population of TV viewers is so large, but many properties on the Internet are worthy to utilize, for example, the lower cost and more convenience of connection to other countries and even other continents.
In 2006, England and Sweden have the online advertisements market beyond 10% of their total domestic market of advertisements, these two countries are the first two in the world with two-digit percent market share, and it may happen on Australia, Israel, Japan, Norway, South Korea, and Taiwan before 2008 predicted by eMarketer. The value of web ads may be illustrated by some decisions of the famous IT companies, Google acquiring Doubleclick for
$3.1 billion and Yahoo buying RightMedia for $680 million in April 2007, and Microsoft buying aQuantive for $6 billion in May 2007.
As the information technologies progress rapidly, the popularity of internet advertising
rises with many unique advantages which had not ever appeared on other traditional media.
Four main attractive characteristics of internet advertising:
1. Interaction: The advertisements on the Internet are not just showing and delivering information to browsers like ads on papers or TV. They can interact with their viewers step by step, like an extension of direct marketing. This feature can catch the viewer’s attention effectively and communicate with the targeted audience in depth.
2. Traceability: One of the most convenient functions of the Internet is hyperlinks, and the users of the Internet can reach the websites of the ad hosts easily and rapidly from the page they are viewing, and even finish the whole transaction process just on the net. This function makes the feedback and effectiveness of advertising traceable and measurable. It is a very useful tool for ads hosts to make decisions and to improve advertisement quality.
3. Segmentation: Website hosts can collect visitors’ information by online table directly or record their behavior on the Internet at a low cost. It is very useful for targeted promotion, which not only improves the effectiveness of advertising but also reduces the possibility of disturbing the people who are not happy to receive the ads.
4. Flexibility: Websites on the Internet can operate around the clock. Hosts can modify and update the ads easily anytime. This makes the promotion activities more flexible and powerful. One of the remarkable instances is the online auction.
As the importance of web advertisement increases, many researches focus on this subject in recent years. One part of researches is studying how to improve the effectiveness of web advertisement of each impression, and they are based on many different social science methodologies. Some researches founded the models and algorithms to select the attractive advertisements for the specified web guests by observing and recording their behavior and hobbies on the Internet. Some researches made effort to classify the types of advertisements and the types of web context (or content), and then put the relative advertisements on the matching web pages. Google, Yahoo!, AOL, and MSN the four big portals in American made
2
$12.5 billion in 2005, which is 53.7% of Internet ad spending in the year (EADP 2007). The phenomenon that revenues of web ads center on these extremely big portals is not only caused by their huge mass browsers but also the various services they provide which can hold the visitors longer time, be useful to website managers to record the browsers’ behavior, target their interests, and raise the effect of ads impressions. United States Patent No. 5948061,
“Method of Delivery, Targeting, and Measuring Advertising over Networks” belong to Double Click Inc. is one famous instance revealing the commercial value of spreading strategy of ads.
Another part of researches is studying how to select and allocate the advertisement orders to make best profits for advertisement agents or website owners. Internet advertising have to face the same challenge like ads on other media, how to fill in the ads to the contents which audience want to receive and not disturb them too much, and get the maximum revenues of media owners with the limited display space/time slots. This is the topic we study in this paper and we will review the literature below.
Before go back to our topic on the Internet, we may examine some researches which developed practical and applicable methods useful to schedule TV advertisements and were taken in practice by NBC. One special property of TV advertisements is that the first and last slots in a commercial break are more valuable because of their higher audience ratings than other slots, and the schedulers of TV ads have to guarantee the advertisers the proper percentage of their ads appearing in the hot period (i.e. the first or last slots in a break).
Besides, each ad being to be displayed must be classified to some specific type and two ads of competing hosts belonging the same type are not allowed to appear in the same break.
Bollapragada et al. (2004) presented a formal mathematical program formulation for this problem and developed a heuristic replacing the existing manual schedule process of NBC.
Another property of TV advertisements is that advertisers hope their ads could be displayed as evenly as possible during the feasible periods, by each identical ad as a unit. Bollapragada et
al. (2004) presented a formal formulation for this problem and developed heuristics for different size of problems to get the near-optimal solution within reasonable computing time.
The scheduling problem of Internet advertising was first formulated by Adler et al.
(2002), who presented a space-sharing model for this scheduling problem. They defined a fixed area for advertising on the webpage, and each ad needed put on has its specified geometry size and displayed frequency. One special constraint for the Internet ads scheduling problem is, we cannot put the same ads at the same time slot due to the real effectiveness consideration, for all the types of problems we will mention below. For the first type of problem in this paper, our schedule has to satisfy the frequency of each ad and cannot violate the space constraint at any moment, and our objective is to use the minimal time slots to finish all the tasks, and we can view this problem as a variant of bin packing problem. Another type of problem is that we set a fixed number of time slots (limited planning time horizon), and select the optimal subset of ads to maximize the revenues of ad agents, based on the assumption that the price of each ad showed once is in direct ratio to its display space. The authors also provide efficient algorithms in this paper to find optimal solutions to the former problem with specified restrictions and 2-approximation solutions for the latter problem. They also discuss how to make a decision to take the orders or not in online condition, and one algorithm was designed for it in the paper.
The maximum revenues problem in offline mode, which means all demands of ads must be implemented, was broadly discussed and studied, and many algorithms and approximation results were published. Amiri et al. (2003) used Lagrangean decomposition to solve the problem in relatively short computing time to get good results. Dawande et al. (2003) provided the first known algorithms with constant factor approximations to this problem, the simple one guarantees 1/4 and the complex one does 3/10 ratio of their performance to upper bound value presented in this paper. Freund et al. (2004) designed a algorithm guarantee 1/(3+ε)-approximate for general case and 1/(2+ε) for two special cases.
4
For the same objective function, to get the maximum revenues of ad agents, Menon et al.
(2004) introduced another model to the Internet ads scheduling problem. They relaxed the constraint of reaching all demands of accepted ads, and this model allowed the ad agents to make better revenues but just satisfy partial demands of the ads in schedule. They used Lagrangean decomposition and column generation to solve this problem and compare their performances, and the column generation got better results and took a shorter time. Amiri et al.
(2006) present a more flexible and practical model, the same relaxation on demands of ads as mentioned before, but a more precise pricing scheme for the real world trading. The ad customers can list out several levels of demands been implemented, and each display of ad need to pay to ad agents by its levels, the level with more counts of display, the higher individual price of each display. Authors of this paper deploy a Lagrangean decomposition-based solution procedure to deal with problems of this new model, and this approach can perform very well for reasonably large problems.
Except the maximum revenues objective, some researches were trying to solve the problem call “MINSPACE”, defined by Dawande et al. (2003), we set a fixed number of time slots for use, and we have to schedule all the ads with their demands exactly, and try to minimize the maximum slot fullness. The solution of this problem can help the ad agents decide what size the banner is being efficient and suitable for their customers’ demands.
Dawande et al. (2003) provide a 2-approximation algorithm for this problem. Dawande et al.
(2005) present an online version of the MINSPACE problem, and provided an algorithm with the performance bound of (2-1/N), and they provided two off-line algorithms with performance ratios (1+1/ 2 ) and 3/2 , respectively. The N stands for the maximum capacity of one slot.
In this thesis, we present a new model that schedules the off-line orders contents with static pricing scheme but limit the feasible time window for each ad order (i.e. release date and due date), with the same objective of maximizing revenues of ad agents. The rest of this
thesis is organized as follows. We present a new model called Scheduling of Banner Advertisements with Time Windows in Chapter 2. In Chapter 3, we introduce our heuristic to solve this problem and implement the heuristic with a small size data set. In Chapter 4, we design one upper bound for our objective, and we will implement the upper bound with the same data set. Chapter 5 is our computational experiments and analysis of the results. Chapter 6 is the conclusion and remarks.
6