Abstract: A method and system conduct an auction for advertising across multiple markets. A first market conducts a first auction for a first advertising impression. A first bid is selected as the auction winner and a ratio is computed as the first bid plus a spread to the second highest bid. The price to be paid is the lower of the first bid or the first bid multiplied by the ratio. The first bid, second bid, and spread are transmitted to a second market for a second advertising impression auction. The second market respects the rules of the first market's auction where such that if the first bid is selected as the winner, a ratio is computed as the first bid plus the spread to the next highest bid. The price to be paid for the second impression is the first bid multiplied by the new ratio.

Abstract: In one aspect, there is provided a method for polling, by a script at a browser, one or more servers by sending messages to the one or more servers. The one or more messages may solicit one or more bids for presenting an ad at the browser. The one or more messages may also enable access to information at the browser. The information may be relevant to the one or more servers making the one or more bids. The script may receive one or more responses from the one or more servers. The one or more responses include the one or more bids. The script may also evaluate the one or more bids, and select one of the bids based on the evaluation.

Abstract: A system and method for optimizing advertisements. To maximize revenue, a plurality of ad networks are tiered based on their pricing data, and in one embodiment, their cost per thousand impressions (CPM). Each tier includes a pricing data range. Periodically, the system may increase and decrease frequency caps for the ad networks to adjust the ad networks in the tiers. Frequency caps may be increased for an ad network when the CPM for the ad network is above the CPM range for the ad network's tier. The frequency caps may be decreased for the ad network when the CPM for the ad network is below the CPM range for the ad network's tier. For each ad network request received, the system traverses through the tiers of the plurality of ad networks for an ad network that is capable of serving an ad based on the ad network's frequency cap.

Abstract: A system reducing the number of default ads displayed by a user agent thereby maximizing revenue for webpage publishers. To maximize revenue, the system manages multiple ad networks so that webpage publishers can have access to advertisements from the multiple ad networks. Typically, when an ad network serves a default ad, the system queries the next ad network. If the next ad network serves a non-default ad, then the ad is served. Otherwise, the system selects another ad network until either a non-default ad is served or a specific period of time elapses or a maximum number of ad networks have been queried, in which case the default ad is served. Ad networks may be ranked by revenue generation and/or the probability of serving a default ad. Inquiries to ad networks may be biased according to such rankings.

Abstract: Embodiments are directed towards improving a user's experience and confidence in advertisements shown within web pages or other online media, in part, by providing additional information for a given advertisement to be displayed along with the given advertisement to encourage, or discourage, a user to click on or interact with the advertisement. Such information to be displayed about a given advertisement may include, without limitation, information about the advertisement, such as its security status (e.g. it has been found to be free from malware), information about the advertiser, including if the advertiser is determined to be trusted based on some criteria, customer satisfaction information, information about the goods and/or services being advertised, including, for example, if how they compare to a competitor, and information about the destination Web page or Web site that the advertisement leads to when a user clicks it.

Abstract: A system and method for advertisers to run multiple advertisement campaigns across several locations without having to configure each campaign on a per location basis is provided. The system may determine and assign budget, target radius, duration, pricing, times of day, etc., for each ad campaign and each location. These settings are not only configured at the time of initial setup, but may also be changed dynamically to adjust for actual results and to account for time-dependent variables. In being changed dynamically, the settings may be changed periodically or aperiodically (e.g., based on events or thresholds being crossed). The setup and updates of the budget, target radius, duration pricing, times of day, etc., may be performed using an optimization algorithm, in one embodiment, based on historical information and current campaign performance, respectively.

Abstract: A system and method for optimizing advertisements. To maximize revenue, a plurality of ad networks are tiered based on their pricing data, and more preferably, their cost per thousand impressions (CPM). Each tier includes a pricing data range. Periodically, the system increases and decreasing frequency caps for the plurality of ad networks to raise the plurality of ad networks to a higher tier. Frequency caps are increased for an ad network when the CPM for the ad network is above the CPM range for the ad network's tier. The frequency caps are decreased for the ad network when the CPM for the ad network is below the CPM range for the ad network's tier. For each ad network request received, the system traverses through the tiers of the plurality of ad networks for an ad network that is capable of serving an ad based on the ad network's frequency cap.

Abstract: A system and method for determining inappropriate content within images. A plurality of training images are used to teach the machine. The training images are converted into numerical data and stored along with its human judged label in a BigMatrix. Through the BigMatrix, a RandomForest is created to discern patterns among the training images and human-judged labels. To determine whether an image contains inappropriate content, the image is converted into numerical data. The numerical data is fed to the RandomForest generated from the plurality of training images and known content. The numerical data is fed down each tree within the RandomForest. When the numerical data is routed down through the branches of the trees and terminated at a leaf node, a vote for the leaf node is obtained. The overall response of the RandomForest is given by a majority rules vote for each tree within the RandomForest.