US 20050235030 A1
The present invention is a system, method, and computer program product for tracking and measuring digital content that is distributed on a computer network such as the Internet. The system collects online advertisement data, analyzes the data, and uses the data to calculate measurements of the prevalence of those advertisements. The system processes raw traffic data by cleansing and summarizing the traffic data prior to storing the processed data in a database. An advertisement sampling system uses site selection and definition criteria and a probe map to retrieve Web pages from the Internet, extract advertisements from those Web pages, classify each advertisement, and store the data in a database. A statistical summarization system accesses the processed raw traffic data and the advertisement data in the database to calculate advertising prevalence statistics including the advertising frequency, impressions, and spending.
1. A system for estimating the prevalence of digital content on the World-Wide-Web, comprising:
an estimating device for estimating the global traffic to a plurality of Web sites to provide traffic data;
a sampling device for statistically sampling the contents of said plurality of Web sites to provide sampling data;
a storage device for storing said traffic data and said sampling data; and an accessing device for accessing said traffic data and said sampling data stored in said storage device.
The present application claims priority from, and incorporates by reference, the provisional application for letters patent, No. 60/175,665, filed in the United States Patent and Trademark Office on Jan. 12, 2000, and provisional application for letters patent, No. 60/231,195, filed in the United States Patent and Trademark Office on Sep. 7, 2000.
The present invention relates generally to a system, method, and computer program product for tracking and measuring digital content that is distributed on a computer network such as the Internet. More particularly, the present invention relates to a system, method, and computer program product that collects online advertisement data, analyzes the data, and uses the data to calculate measurements of the prevalence of those advertisements.
The increase in the popularity of the Internet and the World-Wide-Web (“Web”) is due, in part, to the interactive technologies that a Web page can employ. These interactive technologies directly affect the Web as an advertising medium because the technologies introduced new advertising formats such as fixed icon sponsorship advertisements, rotating banners and buttons, and interstitial advertisements (i.e., online advertisements that interrupts the user's work and takes over a significant percentage of the screen display). Even though the creation of the advertisement is different, the affect on the viewer is similar to traditional advertising. For example, a banner advertisement or logo icon on a Web page creates an impression of the product for the viewer that is equivalent to a traditional billboard advertisement that promotes a product by presenting the brand name or slogan. Similarly, a sponsor's logo on a Web page creates an impression of the sponsor for the viewer that is equivalent to seeing a sponsor logo on the scoreboard at a college basketball game.
The rapid and volatile growth of the Internet over the last several years has created a high demand for quality statistics quantifying its magnitude and rate of expansion. Several traditional measurement methodologies produce useful statistics about the Internet and its users, but the complexity of the Internet has left some of these methodologies unable to answer many important questions.
Online advertising is one area where traditional methodologies do not lend well to measurement. Each day, thousands upon thousands of electronic advertisements appear and then disappear from millions of Web pages. The transitory nature of online advertising activity warrants a novel methodology to accurately measure advertising activity.
Existing advertisement tracking and measurement systems automate the collection of Web pages, but fail to automate the collection of the online advertisements. Since the content of an online advertisement changes or rotates over time, accurate reconstruction of the frequency of specific advertisements requires continuous sampling of relevant Web pages in the correct proportions. Furthermore, due to the sheer size of the Web, sampling algorithms must be finely tuned to optimize the allocation of resources (i.e., network bandwidth, database storage, processor time, etc.) and simultaneously enable maximum Internet coverage. The existing advertisement tracking and measurement systems fail to meet these needs because they are not optimized for resource allocation and do not continuously sample relevant Web pages in the correct proportion.
In view of the deficiencies of the existing systems described above, there is a need for an advertisement tracking and measurement system that uses resources more intelligently, is friendlier to the Web sites that it visits, is scalable, and produces accurate measurements. The invention disclosed herein addresses this need.
The present invention is a system, method, and computer program product for tracking and measuring digital content that is distributed on a computer network such as the Internet. The system collects online advertisement data, analyzes the data, and uses the data to calculate measurements of the prevalence of those advertisements.
In the preferred embodiment, traffic data from a variety of sources and complimentary methodologies fuels the traffic analysis system, an intelligent agent (i.e., software that interact with, learn from, and adapt to an environment). The traffic analysis system processes raw traffic data by cleansing and summarizing the traffic data prior to storing the processed data in a database. When the statistical summarization system calculates the advertising frequency, impressions, and spending, it relies upon the processed data from the traffic analysis system.
The advertisement sampling system, also known as the “prober” or “Cloudprober”, use a robust methodology that continually seek out the most significant and influential Web sites to probe (i.e., monitor). Moreover, the site selection and definition performed by the present invention dictates the Web pages that comprise each Web site to ensure that complete, singularly branded entities are reported as such. The advertisement sampling system uses intelligent agent technology to retrieve Web pages at various frequencies to obtain a representative sample. This allows the Cloudprober to accurately assess how frequently each advertisement appears in the traffic data. After the Cloudprober fetches a Web page, the advertisement sampling system extracts the advertisements from the Web page. In the preferred embodiment, the advertisement extractor, also known as the “extractor”, invokes an automatic advertisement detection (“AAD”) process, a heuristic extraction process, to automatically extract all of the advertisements from the Web page.
Following extraction of the advertisements from the Web page, the advertisement sampling system invokes a classification engine to analyze the advertisement fragments. The classifier processes each fragment to determine a classification for the fragment and then stores the fragment and classification data in a database. The result of the analyses and processing performed by the advertisement sampling system is a rich catalog of advertising activity that can be easily queried by a client.
The present invention uses a Web front end and user interface to access and update the data in the database. The Web front end provides a client, or user, of the present invention with a query interface to the database populated by the traffic analysis, advertisement sampling, and the statistical summarization systems. The user interface is a graphical user interface that includes a separate component for system account management, site administration, taxonomy administration, advertising content classification, and rate card collection. The user interface allows an account manager and operator to maintain and administer the present invention. The user interface also allows a media editor to review the data in the database to verify the accuracy and integrity of the vast amount of data collected by the present invention. This data integrity process routinely investigates unusual or outlying data points to calibrate the system and adapt it to an ever-changing environment.
The accompanying figures best illustrate the details of the present invention, both as to its structure and operation. Like reference numbers and designations in these figures refer to like elements.
As shown in
The traffic sampling system 120 is a program that monitors and records Web activity on the Internet 100. The traffic sampling system 120 is an intermediary repository of traffic data between a Web surfer (not shown) on the Internet 100 and a Web server 112. The Web server 112 shown in
The present invention can use any commercially available traffic sampling system that provides functionality similar to the Media Metrix audience measurement product. Other possible mechanisms to obtain a traffic data sample include:
1. “Proxy Cache Sampling” gathers data such as user clickstream data, and Web page requests from a global distributed hierarchy of proxy cache servers. This data passes through an intermediate mechanism that provides pre-fetch and caching services for Web objects. As of May 1999, traffic statistics calculated by the present invention represent the distillation of raw data from nine first-tier and approximately 400 second-tier caches in the United States, as well as an additional 1100 worldwide.
2. “Client-Side Panel Collection” retrieves sample data from each panelist via a client-side mechanism and transfers that data to a collection repository. The client-side mechanism may monitor the browser location bar, use browser, a client-side proxy, or TCP/IP stack hooks.
3. A “Transcoder” is a proxy that rewrites HTML, usually for the purpose of adding elements for generation of advertisement revenue or page headers/footers. Free Internet service providers (“ISPs”) typically use this technique.
4. Any content distribution mechanism that replicates Web page or site content in a manner meant to ease network congestion or improve user experience.
5. Any content filtering mechanism that evaluates requests for URLs and takes actions to allow or disallow such requests.
6. From server logs maintained by Internet service providers (“ISPs”) or individual Web sites.
The traffic analysis system 210 receives raw traffic data from the traffic sampling system 120. The traffic analysis system 210 cleanses the raw traffic data by removing information from the traffic data that may identify a particular user on the Internet 100 and then stores the anonymous data in the database 200. The traffic analysis system 210 estimates the global traffic to every significant Web site on the Internet 100. This present invention uses this data not only for computing the number of advertising impressions given an estimate of the frequency of rotation on that page, but also in the probe mapping system 320. In one embodiment, the traffic analysis system 210 receives traffic data from a cache site on the Internet 100. The goal is to accurately measure the number of page views by individual users, and therefore the number of advertising impressions.
The advertisement sampling system 220 uses the anonymous traffic data to determine which URLs to include in the sample retrieved from the Web server 112. The advertisement sampling system 220 contacts the Web server 112 through the Internet 100 to retrieve a URL 114, 116, 118 and extract the advertisements therein along with the accompanying characteristics that describe the advertisements. The success rate for retrieval of creatives is high. Analysis indicates that the present invention captures over 95% of creatives served. The advertisement sampling system 220 stores these advertisement characteristics in the database 200. The advertisement sampling system 220, for example, the Cloudprober, Online Media Network Intelligent Agent Collection (“OMNIAC”), or the Cloudprober, repeatedly probes prominent Web sites, extracts advertisements from each Web page returned by the probe, and classifies the advertisements in each Web page by type, technology and advertiser.
The traffic analysis system 210 and the advertisement sampling system also present the data retrieved from the Internet 100 to the statistical summarization system 230 for periodic processing. The statistical summarization system 230 calculates the advertising frequency, impressions, and spending on per site per week basis.
The graphical user interface for the present invention includes the user interface 240 and Web front end 250. The account manager 260, operator 262, and media editor 264 access the user interface 240 to administer access by the client 140 to the Web front end 250 (e.g., account and password management), define sites and probe instructions, and manage the advertising taxonomy, content classification, and rate card collection for the advertising prevalence system 130. The Web front end 250 is the Web browser interface that a client 140 uses to retrieve the advertisement measurement results from the database 200 as generated by the traffic analysis system 210, advertisement sampling system 220, and the statistical summarization system 230.
The traffic analysis system 210 includes an anonymity system 310 and traffic summarization process 312.
The anonymity system 310 cleanses the data received from the traffic sampling system 120 by removing information that identifies a particular user on the Internet. The data is rendered anonymous by passing all user information (e.g., originating internet protocol (“IP”) number or cookies) through a cryptographically secure one-way hash function; this assures the utmost privacy for Web users without devaluing the resulting data. The anonymity system 310 presents the cleansed data to the traffic summarization system 312 which in turn stores the aggregated URL count information in database 200.
The traffic summarization process 312 receives cleansed data from the anonymity system 310. The anonymous traffic data is summarized to yield traffic totals by week or month for individual URLs, domains, and Web sites. The traffic summarization process 312 scales the data by weighting factors to extrapolate total global traffic from the sample.
The advertisement sampling system 220 in
The probe mapping system 320 generates a probe map, i.e., the URLs 114, 116, 118 that the advertisement sampling system 220 will visit. This probe map assists the advertisement sampling system 220 with the measurement of the rotation of advertisements on individual Web sites. The preferred embodiment of the present invention continuously fetches various Web pages in the probe map. In an alternative embodiment, the present invention visits each URL in the probe map approximately every 6 minutes. Another embodiment can vary the fetching rate by considering several factors including the amount of traffic that visits the Web site as a whole and the individual Web page in question, the number of advertisements historically seen on the Web page, and the similarlity of the historically observed ad rotation to other sampled pages.
The Web page retrieval system 322 uses this probe map generated by the probe mapping system 320 to determine which Web pages it needs to sample and the frequency of the sampling. For each URL in the probe map generated by the probe mapping system 320, the Web page retrieval system 322 fetches a Web page, extracts each advertisement from the Web page, and stores the advertisement's attributes in the database 200. The data retrieved from each URL in the probe map is used to calculate the frequency with which each advertisement is shown on a particular Web site
The advertisement extractor 326 extracts the online advertisements from the result of the simulation performed by the Web browser emulation environment 324. The advertisement extractor 326 identifies features of the advertising content (i.e., “fragments”) extracted from the Web pages returned by the probe mapping system 320 that are of particular interest. Advertisements are the most interesting dynamic feature to extract, however, an alternative embodiment of the present invention may use the extraction technology to collect any type of digital content including promotions, surveys, and news stories. The advertisement extractor 326 can use various advertisement extraction methods, including rule-based extraction, heuristic extraction, and comparison extraction.
Rule-based extraction relies upon a media editor 264 to use the user interface 240 to create rules. The user interface 240 stores the rules in the database 200 and the advertisement extractor 326 applies the rules to each Web page that the Web page retrieval system 322 retrieves. The effect of running a rule is to identify and extract an HTML fragment from the Web page (i.e., the part of the page containing the advertisement). The advertisement extractor 326 first converts the HTML representation of the fetched Web page into a well-formed XML representation. Following this conversion, the rules are applied to the parse tree of the XML representation of the Web page.
Heuristic extraction relies upon the similarity of advertisements at the HTML or XML source code level because the advertisements are typically inserted by an advertisement server when the Web page is generated in response to the Web browser emulation environment 324 request to display the Web page. Heuristic extraction analyzes the source code for clues (e.g., references to the names of known advertisement servers) and extracts fragments that surround those clues. The advantage of this method is that the extraction is automatic and the media editor need not create the rules.
Comparison extraction repeatedly fetches the same Web page. This extraction method compares the different versions of the Web page to determine whether the content varies from version to version. The portion of the Web page that varies with some degree of frequency is usually an advertisement and is extracted.
The structural classifier 328 parses each advertisement and stores the structural components in the database 200 and passes those components to the statistical summarization system 230. Each advertisement fragment extracted by the advertisement extractor 326 is analyzed by the structural classifier 328. The process performed by the structural classifier 328 comprises duplicate fragment elimination, structural fragment analysis, duplicate advertisement detection.
The structural classifier 328 performs duplicate fragment elimination by comparing the current advertisement fragment to other fragments in the database 200. Two advertisement fragments are duplicates if the fragments are identical (e.g., each fragment has the exact same HTML content). If the structural classifier 328 determines that the current fragment is a duplicate of a fragment in the database, the advertisement sampling system 220 logs another observation of the fragment and continues processing fragments.
The structural classifier 328 performs structural fragment analysis on the XML representation of the Web page by determining the “physical type” of the fragment (i.e., the HTML source code used to construct the advertisement). Physical types that present invention recognizes include banner, form, single link, and embedded content. Banner advertisement fragments include a single HTML link having one or two enclosed images and no FORM or IFRAME tag. Form advertisement fragments include a single HTML form having no IFRAME tag. Single link advertisement fragments include a link with textual, but no IMG, FORM, or IFRAME tags. Embedded content advertisement fragments reference an external entity using an IFRAME tag. After performing this analysis, the structural classifier 328 updates the advertisement fragment in the database. For a banner advertisement fragment, the structural classifier 328 stores the link and image URL's in the database 200. A form advertisement fragment requires the creation of a URL by simulating a user submission that sets each HTML control to its default value. The structural classifier 328 stores this URL and the “form signature” (i.e., a string that uniquely describes the content of all controls in the form) in the database 200. For a single text advertisement fragment, the structural classifier 328 stores the URL for the link and all text contained within the link in the database 200. For embedded content advertisement fragments, the structural classifier 328 stores the URL associated with the external reference in the database 200. This URL is loaded by the system, and the referenced document is loaded. Once the loaded document has been structurally analyzed, the original fragment inherits any attributes that result from analysis of the new fragment.
The structural classifier 328 performs duplicate advertisement detection on each advertisement fragment that has a known physical type because these fragments represent advertisements. Each unique advertisement has information, including which site definitions are associated with the fragment, stored in the database 200. The structural classifier 328 determination of uniqueness depends on different criteria for each type of fragment. The first step for every type of definition is to resolve all URLs associated with the record. URLs that refer to images are loaded, and duplicate images are noted. HTML link URLs, also known as “click URLs”, are followed each time a new ad is created. The final destination for a click URL, after following all HTTP redirects, is noted. This is also done for simulated link submission URLs associated with form definitions. Once all URLs have been resolved, the structural classifier 328 determines whether the advertisement is unique. Banner advertisement fragments are considered unique if they have the same number of images, if the images are identical, and if the destination URL is identical. Form advertisement fragments are considered unique if they have the same signature, and the same destination URL. Single link advertisement fragments are considered unique if they have the same textual content and the same destination URL.
The statistical summarization system 230 calculates the advertisement statistics for each unique advertisement in the database 200. The present invention calculates, for each Web site, the advertising impressions (i.e., the number of times a human being views an advertisement). The present invention calculates the advertising impressions, I, using the formula I=T×R, where T is the traffic going to the site, and R is the rotation of advertisements on that site. The present invention also calculates the spending, S, using the formula S═I×RC, where I is the advertising impressions for a Web site, and RC is the rate code for the Web site. Most advertising buys are complicated deals with volume purchasing discounts so our numbers do not necessarily represent the actual cost of the total buy.
The Web front end 250 is a graphical user interface that provides a client 140 with a query interface to the database 200 populated by the traffic analysis system 210, advertisement sampling system 220, and the statistical summarization system 230. The client 140 can use the Web front end 250 to create, store, edit and download graphical and tabular reports for one or more industry categories depending on the level of service the client 140 selects.
The user interface 240 in
The account manager 260 uses the system account management 340 module of the user interface 240 to simplify the administration of the Web front end 250. The account manager 260 uses the system account management 340 module to create and delete user accounts, manage user account passwords, and check on the overall health of the Web front end 250.
The operator 262 uses the site administration 342 module of the user interface 240 to simplify the administration of the site definitions. Analysts from the Internet Advertising Bureau estimate that over 90% of all Web advertising dollars are spent on the top fifty Web sites. Site selection begins by choosing the top 100 advertising by considering data from Media Metrix, Neilsen/Net Ratings, and the proxy traffic data in the database 200. These lists are periodically updated to demote Web sites with low traffic levels and promote new sites with high traffic levels. The present invention also includes Web sites that provide significant content in key industries. A site chosen for inclusion in the site definitions must have the structure of the site analyzed to remove sections that do not serve advertisements, originate from foreign countries, or are part of a frame set. Sites that originate from a foreign country, such as yahoo.co.jp, sell advertising in the host country, and therefore are not applicable to the measurements calculated by the present invention. Web sites that use an HTML frameset are treated very carefully to only apply rotation rates to the traffic from the sections of the frameset that contain the advertisement. These combined exclusions are key to making accurate estimates of advertising impressions. The present invention also tags sections that cannot be measured directly, due to registration requirements (e.g., mail pages). Since Web sites change frequency, this structural analysis is repeated periodically. Eventually the analysis stage will automatically flag altered sites to allow even more timely updates.
The media editor 264 uses the taxonomy administration 344, advertising content classification 346, and rate card collection 348 modules of the user interface 240. The taxonomy administration 344 module simplifies the creation and maintenance of the attributes assigned to advertisements during content classification including the advertisements industry, company, and products. The taxonomy names each attribute and specifies its type, ancestry and segment membership. For example, a company Honda, might be parented by the Automotive industry and belong to the industry segment Automotive Manufactures. The advertising content classification 346 component assists the media editor 264 with performing the content classification.
The structural classifier 328 performs automated advertisable assignment to determine what the advertisement is advertising. This process include assigning “advertiseables” (i.e., attributes describing each “thing” that the advertisement is advertising) to each advertisement fragment. In another embodiment of the present invention, the advertisement sampling system 220 uses an extensible set of heuristics to assign advertisables to each advertisement. In the preferred embodiment, however, the only automatic method employed is location classification. Location classification relies on the destination URL in order to assign a set of advertisables to an advertisement. A media editor 264 uses the user interface 240 to maintain the set of classified locations. For example, the first time a media editor observes an advertisement in which the click-thru URL is www.honda.com, he can enter this URL as pertaining to the advertiser “Honda Motors”. Any subsequent advertisement that includes the same click-thru URL will also be recognized as a Honda advertisement. A classified location comprises a host, URL path prefix, and set of advertisables. Location classification assigns a classified location advertisables to an advertisement if the host in the destination URL matches the host of the classified location and the path prefix in the classified location matches the beginning of the path in the destination URL.
The structural classifier 328 performs human advertisable assignment and verification as a quality check of the advertisable data. This phase is the most human intensive. A media editor 264 uses a graphical user interface module in the user interface 240 to display each advertisement, verifies automatic advertisable assignments, and assigns any other appropriate advertisables that appear appropriate after inspection of the advertisement and the destination of the advertisement. The location classification database is also typically maintained at this time.
The media editor 264 uses the rate card collection 348 module to enter the contact and rate card information for a Web site identified by the traffic analysis system 210, as well as, designated advertisers. Rate card entry includes the applicable quarter (e.g., Q4 2000), advertisement dimensions in pixels, fee structure (e.g., CPM, flat fee, or per click), cost schedule for buys of various quantities and duration. The media editor also records the URL address of the online media kit and whether rates are published therein. Contact information for a Web site or advertiser includes the homepage, name, phone and facsimile numbers, email address, and street address.
The first step in the process is to normalize the results from the traffic analysis system 210. The traffic analysis system 210 provides the traffic received by each Web page in the traffic data sample.
The next in the calculation process is to compute the Scaled Observations for each advertisement on each Web site 410, 420 by summing the product of the advertisement views from
Also in the alternative embodiment of
The alternative embodiment of
For instance, in
The sampling records 510 partition of database 200 comprises database tables that are logically segmented into traffic data 512, advertisement view logging 514, and advertising structure 516 areas.
The traffic data 512 area contains data processed by the traffic sampling system 120, anonymity system 310, and statistical summarization system 230. The data stored in this schema includes a “munged” URL, and the count of traffic each URL receives per traffic source over a period of time. A “munged” URL is an ordinary URL with the protocol field removed and the order of the dotted components in the hostname reversed. For example, the present invention transforms an ordinary URL, such as http://www.somesite.com/food, into a munged URL by removing the protocol field (i.e., “http://”) and reversing the order of the dotted components in the hostname (i.e., “www.somesite.com”). The resulting munged URL in this example is “com.somesite.www/food”. The present invention uses this proprietary URL format to greatly enhance the traffic data analysis process. The traffic sampling system 120 populates the traffic data 512 area in database 200. The probe mapping system 320 accesses the data in the traffic data 512 area to assist the Web page retrieval system 322 and the statistical summarization system 230 with the calculation of the advertising impression and spending statistics.
The advertisement view logging 514 area logs the time, URL, and advertisement identifier for each advertisement encountered on the Internet 100. This area also logs each time the system does not detect an advertisement in a Web page that previously included the advertisement. In addition, the system logs each time the system detects a potential advertisement, but fails to recognize the advertisement during structural classification. The structural classifier 328 and the Web page retrieval system 322 of the advertisement sampling system 220 populate the advertisement view logging 514 area in database 200. The statistical summarization system 230 accesses the data in the advertisement view logging 514 area to determine the frequency that each advertisement appears on each site.
The advertisement structure 516 area contains data that characterizes each unique advertisement located by the system. This data includes the content of the advertisement, advertisement type (e.g., image, HTML form, Flash, etc.), the destination URL linked to the advertisement, and several items used during content classification and diagnostics, including where the advertisement was first seen, and which advertisement definition originally produced the advertisement. The structural classifier 328 component of the advertisement sampling system 220 populates the advertisement structure 516 area in database 200. The user interface 240 accesses the data in the advertisement structure 516 area to display each advertisement to the media editor 264 during classification editing. The Web front end 250 also accesses the data in the advertisement structure 516 area to display the advertisements to the client 140.
The probing definitions 520 partition of database 200 comprises database tables that are logically segmented into site definition 522, probe map 524, and advertisement extraction rule definition 526 areas.
The site definition 522 area carves the portion of the Internet 100 that the system probes into regions. The primary region definition is a “site”, a cohesive entity the system needs to analyze, sample, and summarize. The system defines each site in terms of both inclusive and exclusive munged URL prefixes. A “munged URL prefix” is a munged URL that represents the region of all munged URLs for which it is a prefix. An “inclusive munged URL prefix” specifies that a URL is part of some entity. An “exclusive munged URL prefix” specifies that a URL is not part of some entity, overriding portions of the entity included by an inclusive prefix. To illustrate, the following is list of munged URLs that may result from the processing of a set of URLs in a traffic sample.
The probe map 524 area contains a weight for each URL in each site that the system is measuring. This weight determines the likelihood that the system will choose a URL for each probe. The system generates the weights by running complex iterative algorithms against the traffic data and the probing records in the database 200. An analysis of the traffic data can discern which URLs have been visited, how often users have visited those URLs. The result of the analysis guarantees that the system performs advertisement sampling of these URLs in similar proportions, given certain constraints such as a maximum number of probes to allocate to any single URL. The data in the sampling records 510 partition of the database 200 is useful for determining which URLs are in need of special handling due to past behavior (e.g., a URL is sampled less infrequently if the system has never detected an advertisement in the URL). The probe mapping system 320 populates the probe map 524 areas in the database 200. The probe mapping system 320 accesses the data in the probe map 524 area to allocate the probes. The statistical summarization system 230 accesses the data in the probe map 524 area to determine which URLs should have their rotations scaled to counter the effect of probe map constraint enforcement.
The advertisement extraction rule definition 526 area describes Extensible Markup Language (“XML”) tags, typically representing a normalized HTML document, that indicate those portions of the content that the system considers to be advertisements. The system defines an extraction rule in terms of “XML structure” and “XML features”. “XML structure” refers to the positioning of various XML nodes relative to others XML nodes. For example, an anchor (“A”) node containing an image (“IMG”) node is likely an advertisement. After using this structural detection process to match the advertisement content, the system examines the features of the content to determine if the content is an advertisement. To continue the previous example, if the image node contains a link (“href”) feature that contains the sub-string “adserver”, it is very likely an advertisement. Features may match based on a simple sub-string, as in the example, or a more complicated regular expression. Another form of extraction rule may point to a specific node in an XML structure using some form of XML path specification, such as a “Xpointer”. The media editor 264 populates the advertisement extraction rule definition 526 area in the database 200. The advertisement extractor 326 of the advertisement sampling system 220 accesses the data in the advertisement extraction rule definition 326 area to determine which portions of each probed page represent an advertisement.
The advertising support data 530 partition of database 200 comprises database tables that are logically segmented into advertisable taxonomy 532, advertising information 534, rate card 536, and extended advertisable information 538 areas.
The advertisable taxonomy 532 area contains a hierarchical taxonomy of advertisables, attributes that describe what the advertisement is advertising. This taxonomy includes industries, companies, products, Web sites, Web sub-sites, messages, etc. Each node in the hierarchy has a type that specifies what kind of entity it represents and a parent node. For example, the hierarchy may specify that products live within companies, which in turn live within industries. The media editor 264 populates the advertisable taxonomy 532 area in the database 200. The user interface 240 accesses the data in the advertisable taxonomy 532 area to generate statistical data and record where companies, industries, etc. tend to advertise. The Web front end 250 also accesses the data in the advertisable taxonomy 532 area to display this information to the client 140.
The advertising information 534 area contains the data that describe what each unique advertisement recorded by the system advertises. This tables in this area associate advertisables with advertisements. For example, the system may associate a company type of advertisable with a specific advertisement to indicate that the advertisement is advertising the company. The system uses the following methods to associate an advertisable with an advertisement:
1. A “direct classification” assigns an advertisable directly to the advertisement. For example, a media editor 264 creates a direct classification by specifying that a particular advertisement advertises the “Honda” advertisable.
2. A “location classification” assigns an advertisable to a location prefix that the system uses to match the destination of the advertisement. For example, a media editor 264 creates a location classification by specifying that the location “com.honda” indicates an advertisement for Honda. An advertisement that points to “com.honda.wwv/cars”, therefore, associates the advertisement with Honda.
3. An “ancestral classification” assigns an ancestor of the advertisable to an advertisement. For example, if a direct classification assigns Honda to an advertisement, the “automotive” industry advertisable is a predecessor of Honda. Ancestral classification uses this relationship to associate automotive to the advertisement.
The media editor 264 populates the advertising information 534 area in the database 200. The user interface 240 accesses the data in the advertising information 534 area to generate statistical data.
The rate card 536 area contains data describing the cost of advertisements on a Web site. These costs include monetary values for each specific shape, size, or length of run that advertisers on the Internet 100 use to determine the cost of advertisement purchases. The system stores rate card data for each Web site that the system probes. The media editor 264 populates the rate card 536 area in the database 200. The user interface 240 accesses the data in the rate card 536 area to generate statistical data.
The extended advertisable information 538 area contains additional information about specific types of advertisables not readily captured in the taxonomy hierarchy. Specifically, this includes additional information related to Web sites and companies, such as company contact information, Web site, and media kit URLs. This information extends the usefulness of the system by providing additional information to the client 140 about probed entities. For example, a client 140 may follow a hyperlink to company contact information directly from a system report. The media editor 264 populates the extended advertisable information 538 area in the database 200. The Web front end 250 accesses the data in the extended advertisable information 538 area to deliver value-added information to a client 140.
The advertising summary 540 partition of database 200 comprises database tables that are logically segmented into advertising statistics 542, data integrity 544, and advertising information summary 546 areas.
The advertising statistics 542 area describes how often an advertisement appears on each Web site. The system calculates and stores the following statistics in this area.
1. The proportion of page views that display an advertisement to the total number of page view. The system determines this statistic by analyzing the probing records.
2. The number of impressions that an advertisement received. The system determines this statistic by measuring traffic levels for the Web site using the site definition and traffic data, and multiplying that measurement by the proportion of page view calculated above.
3. The amount of spending that an advertisement received. The system determines this statistic by applying the rate card information to the number of impressions that the advertisement receives calculated above.
The statistical summarization system 230 populates the advertising statistics area 542 in the database 200. The Web front end 250 accesses the data in the advertising statistics 542 area to report spending, impressions, and advertising rotation to the client 140.
The data integrity 544 area contains in-depth information about statistical outliers and other potential anomalies resulting from trend and time slice analyses. This automated monitoring and analysis guarantees that the system will contain accurate analysis data. In addition, the system uses real world advertising information, as an input to the system, to verify the accuracy of the analysis data. The data integrity analysis system, performed by the statistical summarization system 230, populates the data integrity 544 area in the database 200. The operator 262 accesses the data integrity 544 area to detect potential errors and monitor general system health.
The advertising information summary 546 area summarizes advertising information in a format that is compact and easy to distribute. The system extracts the data in this area from the advertising support data 530 partition. While the data is not as descriptive as the data in the advertising support data 530 partition, it provides the ability to quickly perform a precise query. The advertising support data 530 partition associates each advertisement with a company, product, or industry. If the system associates multiple advertisables of the same type with an advertisement, a single advertisable is chosen to summary those associates using an assignment priority system, as follows:
1. Advertisables associated with an advertisement using direct classification receive the highest possible priority, “M”.
2. Advertisables associated with an advertisement using location classification receive priority equal to the string length of the location prefix to which they are assigned, therefore, a long location prefix string will receive a higher priority than a short location prefix string.
3. Advertisables associated with an advertisement using ancestral classification receive the priority of the assigned ancestor.
4. The advertisement receives the highest priority advertisable in each type.
5. When two ancestors having the same type and priority are assigned to an advertisement, a conflict occurs and must be corrected by the media editor 264.
The statistical summarization system 230 populates the advertising information summary 546 area in the database 200. The Web front end 250 accesses the advertising information summary 546 area to generate reports for the client 140.
The following description discusses one embodiment of the database structure illustrated in
The database objects comprising the “core schema” are most frequently used by various components of the OMNIAC system. Code bases that rely on this schema include implementation of the back end processes that pull advertisements from the Web. Additionally, database schemas utilized by other components associated with OMNIAC are composed of some or all of the tables in the core schema. The core schema is conceptually composed of four sub-schemas including advertising, advertisements, probing, and sites. The advertising sub-schema holds information about “advertiseable” entities along with which entities each advertisement is advertising. The advertisements sub-schema describes the advertisements that the system has located and analyzed. The probing sub-schema defines “when”, “where”, and “how” for the probing process. The sites sub-schema describes Web sites, including structural site definitions and rate card information.
Of the four sub-schemas, Advertising serves the most general purpose and is therefore the most frequently referenced. The primary table in this sub-schema is ADVERTISABLE, which defines advertisables. Many of the conceptual entities in OMNIAC's universe are advertisables: industries, companies, products, services and Web sites are all defined here. The type field, referencing the ADVERTISABLE_TYPE table, differentiates between different types of advertisables, and the parent field organizes records hierarchically, establishing such relationships as industry-contains-company and company-produces-product.
In addition to the inherent grouping implied by the parent-child relationship defined in ADVERTISABLE, ADVERTISABLE_GROUP_MEMBER is used to further group advertisables. Examples of groups defined in this way include automotive classes, travel industry segments, and types of computer hardware.
Other tables in the Advertising sub-schema serve to define what is advertised by each advertisement. ADVERTISES is used to associate advertisables directly with advertisements. LOCATION_ADVERTISES, CLASSIFIED_LOCATION and LOCATION_MATCHES also indirectly associate advertisables with advertisements via the advertisement's destination location.
“Advertisements” referred to above are references to records in AD, the primary table in the Advertisements sub-schema. The Advertisements sub-schema serves to define each advertisement in OMNIAC's universe. Every unique advertisement has a record in AD, along with one or more records in AD_DEFINITION. Advertisement definitions are unique XML fragments OMNIAC has retrieved from the Web. Ads are unique advertisements defined by sets of advertisement definitions determined to be equivalent during automated classification.
Other tables in Advertisements contain advertisement attributes, referenced by AD and AD_DEFINITION. AD_TECHNOLOGY describes known Web technologies used to render advertisements, while TEXT describes textual content for certain advertisements. FUZZY_WEB_LOCATION contains fuzzy locations found in advertisements. Afuzzy location is a URL that needs to be processed by the system, such as an anchor or image. Once OMNIAC has loaded a fuzzy location, a reference is made to MIME_CONTENT if the URL references an image, or DEST_WEB_LOCATION if the URL references another HTML page.
Moving on, the Probing sub-schema controls the behavior of OMNIAC's probing and advertisement extraction components. The primary purpose of this schema is to define target sets. A target set is a conceptual construct that instructs OMNIAC to fetch a set of pages at certain intervals, extracting advertisements using a set of rules called extraction rules. Each target set is defined by a row in TARGET_SET.
The frequencies, locations, and extraction rules that make up each target set are defined in STROBE, AD_WEB_LOCATION, and EXTRACTION_RULE, respectively. The many-to-many relationships between rows in these tables are defined in TS_RUNS_AT, TS_PROBES, and TS_APPLIES.
The fourth and final sub-schema is Sites, which simply records information about Web sites. Each site or subsite defined in the advertisable hierarchy has a corresponding record in SITE_INFO, along with a number of rows in SITE_DOMAIN and SITE_MONTHLY_DATA. SITE_DOMAIN describes the physical structure of a site in terms of inclusive and exclusive URL stems. SITE_MONTHLY_DATA records advertising rate cards, third party traffic estimates, and cache statistics for each site on a monthly basis.
The analysis schema is an extension to the core schema that includes a number of additional tables populated by OMNIAC's analysis module. The analysis module is the unit in charge of processing information held in the core schema, producing a trim dataset that accurately describes advertising activity.
Like the core schema, the analysis schema is composed of four conceptual sub-schemas composed of tables implementing common functionality. These sub-schemas include advertising decomposition, advertisement view summarization, slot statistics, and site statisitics. The advertising decomposition sub-schema holds information about each advertisement in the system, including attributes and what the advertisement is advertising. The advertisement view summarization sub-schema summarizes advertisement views, recording how many times each advertisement was seen in each slot over the course of a day. The slot statistics sub-schema describes advertisement rotation for each slot during each time period. The site statistics sub-schema describes site information, including advertisement rotation for each time period.
The primary table in the Advertising Decomposition sub-schema is AD_INFO, which contains de-normalized records describing advertisement attributes. AD_INFO records are keyed off of ID's in the AD table; an AD_INFO record exists for each AD record that has been completely classified and represents a valid advertisement. AD_INFO is populated by the analysis module from the advertising relationships described in the core schema tables ADVERTISES and LOCATION_ADVERTISES.
Fields in AD_INFO that specify what is advertised by an advertisement are: CATEGORY (industry), ORGANIZATION (company), ORGANIZATION_GROUP (industry segment), ORGANIZATION_OVERGROUP, COMMODITY (product/service), COMMODITY_GROUP (product/service segment), COMMODITY_OVERGROUP, and MESSAGE.
AD_INFO also includes fields describing a number of non-advertising attributes. FORMAT, referencing AD_SLOT_TYPE.ID, specifies the form factor of an advertisement. TECHNOLOGY, referencing AD_TECHNOLOGY2.ID, specifies the technology used to implement the advertisement. DEFINITION, IMAGE, and DESTINATION specify the AD_DEFINITION, IMAGE, and DEST_WEB_LOCATION records associated with the advertisement. These three fields mirror fields in the AD table.
The Advertising Decomposition schema contains a few tables in addition to AD_INFO. ADV_IMPLICATION is a cache of advertisable implications derived from the hierarchy in ADVERTISABLE. This is used to speed operation of the analysis module. AD_INFO_FLATTENED is a more readily queried version of AD_INFO containing advertisement/advertisable pairs for each of the fields in AD_INFO that reference ADVERTISABLE. Finally, AD_TECHNOLOGY2 describes advertisement technologies understood by the analysis module that are presentable to the user in the front end.
The Advertisement View Summarization sub-schema covers the single table PLACEMENT_SUMMARY. PLACEMENT_SUMMARY is keyed off of day, advertisement, and slot, and contains, in the CNT field, the number of times an advertisement was seen in a slot on a particular day.
The analysis module populates PLACEMENT_SUMMARY by aggregating hits recorded in the APD n tables, one of which exists for each day, n being the ID of the day in question. These tables are created and populated by the back-end as advertisement hits flow into the system.
The third sub-schema in the Analysis schema is Slot Statistics. This sub-schema describes advertisement behavior in the context of slots in addition to information about the slots themselves. A slot is a location on the Web in which advertisements rotate, currently defined in terms of the location ID (a reference to AD_WEB_LOCATION.ID) and extraction rule ID (a reference to EXTRACTION_RULE.ID).
The primary table in the Slot Statistics is SLOT_AD_VIEWS, which records the total views and relative frequency for each advertisement in each slot during each time period. The primary key of this table is composed of the fields PERIOD_TYPE, PERIOD, LOCATION_ID, RULE_ID and AD_ID. Two fields exist outside of the primary key: CNT holds the total number of advertisement views, and FREQUENCY holds the relative frequency.
Also in this sub-schema is SLOT_SUMMARY, which records general slot information outside the context of individual advertisements. Accordingly, this table is keyed off the PERIOD_TYPE, —PERIOD, LOCATION_ID and RULE_ID fields. The CNT field records total advertisement views in the slot; this field is divided into the SLOT_AD_VIEWS.CNT to determine relative frequency. Also in SLOT_SUMMARY is a SLOT_TYPE field that specifies the type of advertisement seen most frequently in the slot, and SITE_ID, which specifies which site the slot resides within.
The final table in the Slot Statistics sub-schema is SLOT_TYPE_COUNT. This table is used to determine which value to use in SLOT_SUMMARY.SLOT_TYPE. The number of times each advertisement format was seen is recorded, and the slot type that receives the most views is stuck into SLOT_SUMMARY.SLOT_TYPE.
The object model for the memory 610 of the advertising prevalence system 130 employs a three-tier architecture that includes the presentation tier 620, infrastructure objects partition 630, and business logic tier 640. The object model further divides the business logic tier 640 into two partitions, the application service objects partition 650 and data objects partition 660.
The presentation tier 620 retains the programs that manage the graphical user interface to the advertising prevalence system 130 for the client 140, account manager 260, operator 262, and media editor 264. In
The infrastructure objects partition 630 retains the programs that perform administrative and system functions on behalf of the business logic tier 640. The infrastructure objects partition 630 includes the operating system 636, and an object oriented software program component for the database management system (“DBMS”) interface 632, system administrator interface 634, and Java runtime platform 638.
The business logic tier 640 retains the programs that perform the substance of the present invention. The business logic tier 640 in
When the traffic sampling system 120 finishes collecting information about a URL 114, 116, 118, it sends a message to the TCP/IP interface 622 that invokes a method to create a visit object 642 and stores information about the connection in the visit object 642 state. Visit object 642, in turn, invokes a method in the traffic analysis application 652 to process the information retrieved by the traffic sampling system 120. The traffic analysis application 652 stores the processed data from the anonymity system 310 and probe mapping system 320 in the traffic analysis data 662 state and the database 200.
After the traffic analysis application 652 processes a URL 114, 116, 118 identified by the traffic sampling system 120, the visit object 642 invokes a method in the advertising sampling application 654 to retrieve the URL 114, 116, 118 from the Web site 110. The advertising sampling application 654 processes the retrieved Web page by extracting embedded advertisements and classifying those advertisements. The advertising sampling application 654 stores the data retrieved by the Web page retrieval system 322 and processed by the Web browser emulation environment 324, advertisement extractor 326, and the structural classifier 328 in the advertising sampling data 664 state and the database 200.
After the traffic analysis application 652 and the advertisement sampling system 654 process the URL 114, 116, 118 identified by the traffic sampling system 120, the visit object 642 invokes a method in the statistical summarization application 656 to compute summary statistics for the data. The statistical summarization application 656 computes the advertising impression, spending, and valuation statistics for each advertisement embedded in URL 114, 116, 118. The statistical summarization application 656 stores the statistical data in the statistical summarization data 666 state and the database 200. Even though
Although embodiments disclosed in the present invention describe a fully functioning system, it is to be understood that other embodiments exist which are equivalent to the embodiments disclosed herein. Since numerous modifications and variations will occur to those who review the instant application, the present invention is not limited to the exact construction and operation illustrated and described herein. Accordingly, all suitable modifications and equivalents which may be resorted to are intended to fall within the scope of the claims.