WO2007145775A2 - Extraction de mots-clés et production de publicité ciblée - Google Patents

Extraction de mots-clés et production de publicité ciblée Download PDF

Info

Publication number
WO2007145775A2
WO2007145775A2 PCT/US2007/011992 US2007011992W WO2007145775A2 WO 2007145775 A2 WO2007145775 A2 WO 2007145775A2 US 2007011992 W US2007011992 W US 2007011992W WO 2007145775 A2 WO2007145775 A2 WO 2007145775A2
Authority
WO
WIPO (PCT)
Prior art keywords
keyword
keywords
category
content source
keyword extraction
Prior art date
Application number
PCT/US2007/011992
Other languages
English (en)
Other versions
WO2007145775A3 (fr
Inventor
Alec Reitter
Barbara Chang
Ken Sun
Raghav Gupta
Alvaro Bolivar
Alan Lewis
Original Assignee
Ebay Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US11/646,012 external-priority patent/US8001105B2/en
Priority claimed from US11/646,039 external-priority patent/US8209320B2/en
Priority claimed from US11/645,946 external-priority patent/US7831586B2/en
Application filed by Ebay Inc. filed Critical Ebay Inc.
Publication of WO2007145775A2 publication Critical patent/WO2007145775A2/fr
Publication of WO2007145775A3 publication Critical patent/WO2007145775A3/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/951—Indexing; Web crawling techniques

Definitions

  • This disclosure relates to methods and systems supporting online advertising and online transactions by a user. More particularly, the present disclosure relates to application programming interfaces for keyword extraction and contextual advertisement generation.
  • An increasingly popular way of delivering Internet advertisements is to tie the presentation of advertisements to particular user behaviors and/or user profiles, and/or user demographics.
  • user behaviors include user access to a particular web page, user selection (also called mouse-clicking or clicking) of a particular location on a web page, user entry of a particular search string or keyword, and the like.
  • advertisers or vendors pay to have their advertisements presented in response to certain kinds of events—that is, their advertisements are presented when particular user behaviors warrant such presentation. If a particular advertisement (ad) leads to some user action, an advertiser may receive remuneration for the ad.
  • Some conventional web-based merchants use affiliate programs.
  • the merchant In an affiliate program, the merchant itself must track purchase transactions and reward 3 rd party affiliates when purchase transactions are completed. This transaction tracking and rewarding process imposes a significant administrative burden on the merchant.
  • the tracking/reward functionality must be replicated for each merchant that chooses to use such a system.
  • Current technology does not provide a solution for off-loading this tracking/reward functionality to a 3 rd party without risking an increase in fraudulent transactions and a decrease in the time-efficiency of processing purchase transactions.
  • United States Patent No. 5,948,061 discloses methods and apparatuses for targeting the delivery of advertisements over a network such as the Internet. Statistics are compiled on individual users and networks and the use of the advertisements is tracked to permit targeting of the advertisements of individual users. In response to requests from affiliated sites, an advertising server transmits to people accessing the page of a site an appropriate one of the advertisements based upon profiling of users and networks. BRIEF DESCRIPTION QF THE DRAWINGS
  • Figure 1 illustrates a high-level architecture of the keyword extractor system in an example embodiment.
  • Figure 2 illustrates the basic logic used in the Editor Kit front end of an example embodiment.
  • Figure 4 illustrates a graph showing association of category ID to query term in an example embodiment.
  • Figure 5 illustrates a graph showing association of bid amounts to query term.
  • Figure 6 illustrates a graph showing association of BIN amount to query term.
  • Figure 7 illustrates a graph showing association of product reference ID to query term.
  • Figure 8 illustrates a graph showing association of Bid amounts to ad query term.
  • Figure 9 illustrates a graph showing association of BIN amount to query term.
  • FIG 10 A user interface hi an example embodiment is presented in Figure 10 as a generic mockup showing impressions generated from one ad unit.
  • An example of the contextual ad is illustrated in Figure 11.
  • Figure 12 illustrates an example of a real-time preview in an example embodiment.
  • Figure 13 illustrates an example of an EK to KE interface interaction for various embodiments.
  • Figures 14-17 illustrate a flowchart to diagram the slot selection process in an example embodiment.
  • Graphic and text ads are available with a fixed color scheme as pictured in Figure 18 in an example embodiment.
  • Examples of customizable ads are pictured in Figure 19 in an example embodiment.
  • Examples of the text ads are pictured in Figure 20 in an example embodiment.
  • FIG. 21 A visual example of the tab-based flash ad in an example embodiment is shown in Figure 21.
  • Figure 22 illustrates fixed color theme examples in an example embodiment.
  • Figure 23 illustrates an Ad Components Legend in an example embodiment.
  • FIG. 24 An example of a Host Tools page is illustrated in Figure 24 for an example embodiment.
  • FIG. 25 An example of a "Create Your Own Ad page" is illustrated in Figure 25 for an example embodiment.
  • An example of choosing manual content is shown in Figure 26 for an example embodiment.
  • Figures 27 and 28 illustrate category selection in an example embodiment.
  • Figure 29 illustrates the selected category display mock-up for an example embodiment.
  • Figure 30 illustrates the selected advanced options mock-up for an example embodiment.
  • Figure 31 illustrates the error message mock-up for an example embodiment.
  • Figure 32 illustrates the preview area mock-up for an example embodiment.
  • Figure 33 illustrates the custom color mock-up for an example embodiment.
  • Figure 34 illustrates the color palette mock-up for an example embodiment.
  • Figure 35 illustrates an example of a color/size drop down in an example embodiment.
  • Figure 36 illustrates an example of a custom ad title for an example embodiment.
  • Figure 37 illustrates an example of the Create Your Own Ad page for an example embodiment.
  • Figure 38 illustrates a graph showing association of product reference ID to query term in an example embodiment.
  • Figures 39-40 illustrate a step-by-step process in an example embodiment of how the Host AdContext component selects item listings to display based on the Keyword Extractor recommendations.
  • Figure 41 is a block diagram of a network system on which an embodiment may operate.
  • Figures 42 and 43 are block diagrams of an example computer system on which an embodiment may operate.
  • a computer-implemented system and method for application programming interfaces for keyword extraction and contextual advertisement generation are disclosed.
  • numerous specific details are set forth. However, it is understood that embodiments may be practiced without these specific details. Tn other instances, well-known processes, structures and techniques have not been shown in detail in order not to obscure the clarity of this description.
  • a computer-implemented system and method for keyword extraction and contextual advertisement generation includes a keyword extraction engine operable to extract keywords from various sources based on user interaction with networked content. Further, the system includes a contextual advertisement generator to produce advertising or information content correlated with content with which a user is or has interacted.
  • a keyword extraction engine operable to extract keywords from various sources based on user interaction with networked content.
  • a contextual advertisement generator to produce advertising or information content correlated with content with which a user is or has interacted.
  • Functionality for various embodiments includes components for: 1) building the core components to analyze content and perform keyword extraction for a Host platform; 2) building the additional components to enable the Editor Kit to integrate with the keyword extraction system and provide contextual advertisements; 3) enabling contextual advertising capabilities for the Editor Kit by providing ranked and scored suggestions for: searches (keywords + category constraint), categories, products and catalog properties; 4) using the keyword extraction system in multiple system features both on and off the Host where unstructured text content is read by users.
  • the Host may lack the capability to analyze large volumes of unstructured text and determine if that text contains any keywords that would be of value to the user reading the text with relation to the Host. Creating this capability will enable several short-term and long-term product opportunities for on-Host product features, the Host related sites and off-Host third-party deals.
  • the first use case of this capability is contextual advertising with the use of the Editor Kit. This will help content oriented affiliates who publish diverse content frequently and across large numbers of pages and sites.
  • functionality is included to analyze content from any URL via HTTP; extract searches (keyword + category constraint), categories, products and catalog properties that are relevant to the content; rank the relevant results according to measures of popularity, supply and other performance measures obtained from tracking aggregate user behavior on the site; and use feedback to improve the rankings and results over time.
  • the following use cases illustrate how the keyword extractor service of an example embodiment is used in different system features.
  • the service is designed to make it extensible across all the use cases.
  • Editor Kit Contextual affiliates place an Editor Kit (EK) sniplet on one Advertiser or more web pages. When the page is viewed the sniplet is executed, informs the EK servers of the URL and other parameters. The EK server invokes the keyword extractor service and passed in the
  • the affiliate web page is fetched and analyzed.
  • the EK server uses the keywords to create ad placements on the affiliate page.
  • Meta Content Generator Meta keyword and description tags are generated automatically based on the content of certain pages
  • affiliates Through an API call, affiliates could perform keyword extraction on their own content and build innovative applications on top of this service.
  • the Keyword Extraction (KE) service provides at least two levels of service in an example embodiment. One is a near real-time capability for time critical applications. The second is a delayed analysis capability with a less stringent expectation for the return of results.
  • the Keyword Extractor is a system/service that analyzes HTML content and extracts relevant keywords from that content by using a variety of information. This information includes frequency of user queries, listings availability and catalog data.
  • Figure 1 illustrates a high-level architecture of the keyword extractor system in an example embodiment.
  • consumer applications There will be many different types of consumer applications that will call the- keyword extractor service.
  • the main ways consumer applications interact with the service include: submitting URLs to the service; receiving data back from the service; and emitting impression and clickthrough tracking events (only the Host consumer applications).
  • the consumer application could either be an internal Host product feature or an API (application programming interface) call being used by affiliates to access the service.
  • the following parameters can be provided by the consumer application when requesting results from the keyword extractor.
  • URL n/a This is a URL that specifies the location of the content that the consumer application would like to have keywords extracted for.
  • the only protocol that will be supported is HTTP.
  • HTTPS may also be supported.
  • the URL should be for an HTML document and not some other file format (i.e. JPG, PNG, AVI, etc.)
  • Host Site 0 or US This identifies which site and therefore set of ID keywords and metrics that should be used to perform the keyword extraction and ranking against. This will also restrict the results (i.e. keywords, listings, products) that are returned. Only one site ID allowed.
  • Category n/a Allows an optional category ID to be specified in ID Hints order to provide more accurate results by defining the starting category (or categories) for the category determination process. Multiple category IDs will be accepted.
  • a category ID can be provided at any level in the category tree.
  • Various embodiments are beneficial and advantageous in the capability for an affiliate to pass a hint with the API call in the form of one or more host item category IDs that cause the system to refine its search and return only keywords and categories within those user provided categories specified in the affiliate API call.
  • This component is the primary interface to the keyword extractor service. If the particular URL-algorithm combination that the consumer application is requesting has already been processed and exists in the cache then the data is returned. If the URL-algorithm combination has not been processed and cached yet then a check to determine whether it's still being processed or this URL-algorithm combination has never been seen before. Depending on the result either a status code is returned to the consumer application or the URL-algorithm required is published as a batch execution event in order to get the URL fetched and analyzed.
  • Figure 2 illustrates the basic logic used in the Editor Kit front end of an example embodiment.
  • the interface is a method available for consumer applications to call the service and obtain results. Many different types of consumer applications will call the service and may generate a high load of requests. In order to invoke the service, several parameters may be provided, some are required while others are optional.
  • External consumer applications e.g. affiliate applications
  • Algorithm "Default The consumer application can optionally specify the name of an extractor/ranker algorithm (i.e. particular extraction process) to use. This will be used to either force the service to use a specific algorithm or to allow A/B testing between different algorithms to determine which one is more effective.
  • an extractor/ranker algorithm i.e. particular extraction process
  • Algorithm 1.0 Different variants of the same algorithm may exist. version This will be used to either force the service to use a specific version or to allow A/B testing between different versions to determine which one is more effective.
  • Interval URL is re-fetched and analyzed again to determine if the content has been changed. This is just the initial value and the system will change this automatically on subsequent fetches. Zero indicates that the URL should not be re- fetched.
  • Assets Return Allows the consumer application to specify which set all assets of assets to return in the dataset.
  • the URL is brand new (actually if the URL-Site ID-Category ID Hint- Algorithm-Algorithm Version combination is new; since this combination defines the primary key) then the URL needs to be fetched and analyzed. If the same URL is submitted again but any of the other parameters in the combination is different, then a new fetch and analyze is required since these parameters will influence the result.
  • various embodiments can initiate a re-fetch of the content of a particular page based on the observed changes in the keywords extracted for that page. In this manner, the keywords extracted for a page can act as a proxy for the content of the page. Thus, as observed changes in the keywords extracted for a page occur, changes to the content of the page can be inferred and a re-fetch of the page can be automatically initiated.
  • URL Uniqueness If URLs are submitted with session specific parameters embedded in the query string (i.e. session ID), then the system will treat each of these URLs as unique (even if they display essentially the same content). The system may not attempt to identify these types of URLs.
  • This database contains the information that is generated by the Extractor Service for each URL and returns the data to the Editor Kit Server FE (Front End) when a particular URL is requested again in order to eliminate the latency associated with the fetch-and-extraction process.
  • the cache needs to be periodically flushed (e.g. every 7 days) of URLs that have not been accessed during the previous time period in order to save space.
  • the time period for performing the check and flushing the cache should be made configurable.
  • the cache needs to be extensible enough to hold different result data types in the future (i.e. Listings, Reviews, Guides, Kijiji postings) or the system should be able to support multiple types of caches, depending on the use case.
  • the cache should support all languages, including double-byte ones.
  • This component contains queues for unfetched and fetched URLs as well as associated metadata for those URLs.
  • the Fetch/Extract Consumer component is responsible for consuming URL.FETCH events via BES. For each event consumed the component will fetch the HTML content and any associated external CSS files from the target URL. Once the HTML and CSS have been fetched from a URL, the information is passed to the Extractor Service for processing.
  • Pages Fetched The fetcher will save content on only single pages identified by the URL in the URL.FETCH BES event. The fetcher will not identify additional URLs (i.e. links) on the page in order to crawl deeper into the site.
  • Page Re-Fetch Any URL submitted needs to be periodically re-fetched (unless the re-fetch time was set to zero) in order to determine if the content on the page has changed and caused a different set of keywords to be created.
  • Each page has an optimal re- fetch time that is algorithmically determined by the system. The goal is to refresh the cache contents at or near the same frequency as the URLs publishing frequency.
  • the traditional method of determining whether to re- fetch a page would be to schedule a periodic batch job that would check the re-fetch times on all URLs to determine if any of them had expired. Those URLs whose re-fetch times had expired would be queued up for fetching and re-analysis.
  • the alternative is to examine re-fetch times of those URLs that are requested as a result of an impression event. If a URL is served and its re-fetch time has expired or is close to expiring (i.e. within 30 minutes, configurable without a train roll) then an event to re-fetch and analyze the URL is published.
  • the Fetcher should not execute any scripting commands on the page, but the script code should be saved along with everything else in the page source content.
  • the HTML-to-Text Parser will be responsible for separating out useable content from non-content such as scripts.
  • Multimedia Objects that are referred to on the page and not text
  • Graphics oriented such as Flash, Java applets, graphics or
  • Java Applets multimedia assets will not be fetched or saved.
  • CSS file should be read as well.
  • the CSS file passed to the HTML-to-Text Parser along with the HTML. If the CSS is inline to the HTML in the head or element level commands then just save it as part of the page source.
  • the Fetcher should only fetch the content of IFRAMEs whose src is in the same domain as the parent document that is being fetched. If the src is in a different domain then do not fetch the content of the foreign IFRAME.
  • the Fetcher should not see SSI directives as they should've been processed by the server and the Fetcher would only see the results of the SSI directive. In any case, if there are any types of SSI directives, they should be ignored by the Fetcher.
  • Each URL has a re-fetch interval or time that is calculated after each time the page contents are fetched, extracted and analyzed.
  • the purpose of calculating the re-fetch interval is to determine the optimal frequency at which a URL changes or updates its content, on average, so that the results that are cached for the URL are as up-to-date as possible.
  • the system should continuously monitor and adjust as the publishing patterns of the URL change. To start with, all URLs receive a default re- fetch interval (i.e. 24 hours).
  • a fingerprint needs to be generated for each extract and analysis. In this case, the fingerprint will be generated from the top n Searches (keyword and category) for a given URL.
  • a set comparison needs to be made between the Searches in the previous fingerprint and the current one. If any of the individual Searches has changed (keyword or category) or the sets have a different number of Searches in them then the fingerprint has changed; ranking or scores are not relevant. If the sets are equivalent then the fingerprints and therefore the URL contents can be considered the same.
  • Keyword Store If the Keyword Store is refreshed with new data, this may cause a certain set of pages to change their fingerprints the next time they are refreshed causing a decrease in their re-fetch interval.
  • the fetcher When fetching pages from sites, the fetcher should identify itself as a robot belonging to the Host.
  • the string should be "HostBot/1.0" or something similar. This will be the string that appears in web server logs.
  • Exclusion tag This string should be "ebaybot" or something similar. This will be used by webmasters in robots.txt files to prevent the Fetcher from accessing specific areas of their website. Fetching Exceptions
  • the fetcher will likely encounter many situations (i.e. 4xx status codes) where it cannot properly fetch a page. In general, the fetcher should attempt to re- fetch the page when it encounters problems. Appropriate error messaging should be provided to the consumer application (i.e. EK Server FE) so that the consumer application can take the appropriate action.
  • EK Server FE EK Server FE
  • This component is responsible for taking the page data (HTML and CSS) from either the Fetch/Extract Consumer or Editor Kit Preview FE components and extracting and ranking keywords from that data with the help of the HTML-to-Text Parser and the Text-to-Keyword Extractor/Ranker components.
  • the results should be assembled into a dataset that is easily parsed (i.e. XML) by the consumer application (EK Server FE or EK Preview FE).
  • the dataset should be easily extended to hold new types of data (i.e. Listings, Reviews, Guides, Kijiji postings).
  • Each data type should have its own section in the data.
  • Each section should its results sorted by rank in ascending order.
  • No keywords can be extracted from the content on the URL (i.e. minimal content on the page or content about a subject that has no overlap with the Host buyer and seller interests).
  • a default set of results should be provided.
  • the default results will be the top n ranked keywords within one or more categories. If no category hints were provided, then the category root will be used.
  • situation #2 when no keywords can be extracted from the content, the system should remember this so that the page is not re-fetched and a useless extraction performed each time the URL gets an impression.
  • This requirement is for taking the raw HTML and CSS, if it exists, and generate clean text for easier processing by the Keyword Extractor/Ranker SIBE service.
  • the parsing will also analyze the content and provide weightings for certain pieces of text that it thinks are more important. These weightings will influence the final weighting of the keywords.
  • the main goal of the parser is to extract the content and remove all the extraneous elements that are needed to format a page.
  • all tags for structural and formatting should be removed, unless specifically mentioned in the table below, and all the content in the body of the document should be kept. Note that some of the tags and CSS directives will be used to generate weightings to influence keyword ranking. Additionally, meta information tags can be used to determine keywords and context of a page.
  • the page being parsed may implement content indicator tags. These are the Host proprietary tags that indicate which ranges of text are actual content (as opposed to UI (user interface) navigation elements). These will allow better accuracy on pages that contain a lot of irrelevant content on the pages.
  • the tags are optional and may not be embedded in all of the pages that the system fetches.
  • a page may have multiple pairs of content indicator tags embedded in it. If the one or more pairs of content indicator tags are present, then only the text in those sections should be parsed. Any content outside of those tags will be thrown away, unless there are no tags present at all — then the entire page should be analyzed.
  • the content indicator tags should not be nested or overlapping in any manner.
  • the content indicator tags should be easily parsed and should not cause the page they are embedded in to render differently. Possible implementations are using a DIV tag with a custom class or using a comment tag with a specific string. Comments are probably preferred because they will not cause browsers to place a line break before and after the DIV element.
  • the weighting scheme will apply a set of rules that are based on the formatting and metadata of a page.
  • the rules will assign a weighting to the areas of text and metadata that meet the requirements of a rule. Text that does not meet any of the rules will be classified as neutral weight (zero). Note that this does not mean that neutral weighted text has zero weighting.
  • Adding new rules and changing weightings for existing rules should be easily configured without a train roll.
  • the weighting of the text should not destroy any of the context in the document (i.e. phrases should not be broken up). This is because formatting boundaries are not guaranteed to coincide with token boundaries; therefore weighting chunks of text should not create artificial token boundaries.
  • There will be only one set of weighting rules for all sites (having per-site rules is too much of a maintenance problem). Rules can apply to various elements on a page but will mostly apply to meta tags, HTML tags and CSS rules.
  • the text should be pre-processed in order to normalize all variations and put the text into a consistent format. Some scrubbing procedures may not apply to all languages.
  • the scrubbing process should be identical to the one used for scrubbing queries on the site in order to ensure that both processes are generating the same output.
  • the Text-to-Keyword Extractor/Ranker component consumes the output (derived from the actual content) of the HTML-to-Text Parser and analyzes it with the aid of the data in the Keyword Store.
  • the output is a set of assets that is contextually related to the analysis and metadata for each asset.
  • An asset represents anything that could be surfaced to the user reading the original content. This document will only focus on specific assets that exist on the Host (.com and country sites) but the system should be designed to be extensible enough so that other assets could be added, whether from the Host (i.e. Reviews and Guides) or come from affiliated sites such Kijiji (i.e. postings).
  • Spell-check, Stemming and Transliteration All of the tokens should be run through three operations: spell-check, stemming and transliteration. Each token should be spell-checked and replaced with the top spell-check suggestion if there is an error in the original token. Perform stemming and transliteration on all the tokens in order to normalize them. After these three processes are performed on the tokens then the metrics should be combined according to pre-defined rules.
  • the general framework of the ranking system in an example embodiment is to use multiple inputs in the form of histograms and combine the data from the histograms in an equation that will produce a numerical value for each token.
  • the framework should be flexible enough to add additional algorithms in the future and also change the equation that combines the data from the histograms into producing a numerical value for each token. This is needed to support A/B testing of different algorithms.
  • New algorithms can be created either by: 1) configuring the weightings of various histogram inputs into an existing algorithm. This should be done by without requiring a new train roll. Different weightings for the same basic algorithm should be available simultaneously for A/B testing; or 2) creating a brand new algorithm that computes the ranking in a brand new manner. This can be done with a new train roll.
  • rankings will be produced for the following:
  • Ranking keywords will use both relevance data and financial data to rank each [keyword + category]. Both sets of information may be combined in order to create a single value for each [keyword + category] that indicates the relative value of the [keyword + category] compared to the other [keyword + category] items in the same result set.
  • Keyword Scoring Step 1 The step of categorization and tokenization of the content is based on site- wide supply and demand data as well as URL specific data all of which are processed by the pseudocode set forth below. Note that site-wide bid/BIN data will not be used in this step.
  • Each [keyword + category] will have a relevance score computed for it as part of the categorization and tokenization step.
  • Keyword Scoring Step 3 The relevance score for each [keyword + category] is transformed into a probability score based on the relative difference between [keyword + category] elements for the page.
  • Keyword Scoring Step 4 Given a set [keyword-)- category] items from the relevance scoring steps, the system should use the combination of "Host ID x URL x [keyword + category]" to look up the historical financial performance of that combination (recency- weighted average of bid, BIN and registration activity). If the historical financial performance is not available then an estimate should be used by looking up the alternate aggregate financial information for the keyword. The financial score will be in the currency of the Site ID used and represent the expected value of bid/BIN and registration activity created per one thousand impressions of a given "site ID x URL x [keyword + category]" combination.
  • Keyword Scoring Step 5 The expected financial value for each [keyword 4- category] is transformed into a probability similar to relevance probability in step 3, where the expected value of the [keyword + category] combination is considered in relation to the combined expected value of all the other [keyword + category] combinations for the page.
  • the relevance probability and financial probability for each [keyword + category] item are mathematically combined to produce a single probability score for each [keyword + category] item. Once the single probability score is calculated for each of the [keyword + category] items in the set, the results can be stored in the URL to keyword cache for a front-end application to access.
  • Keyword Store contains the various histograms that act as inputs to the ranking algorithm.
  • Each country site on the Host should have its own set of histogram data to provide accurate results for content-driven ads that will be driving users to that specific site.
  • This component should also be made extensible so we can add sets of metrics from additional platforms to the system (i.e. Rent.com or Kijiji sites) when we extend keyword extraction to those platforms and suggest assets from those systems. Different platforms will likely have different sets of histograms.
  • the histograms in this section are those that will specifically run the keyword extractor system for the Host platform.
  • the data for the following histograms do not necessarily need to be implemented as separate data structures. They are separated logically in the diagram for clarity.
  • the histogram data should be refreshed on a regular basis in order to accurately track trends and provide accurate recommendations.
  • FIG. 4 illustrates a graph showing association of category ID to query term in an example embodiment.
  • the category ID of the listing is associated with query term and the count of view item actions for that particular query term-category ID combination is incremented by 1. If a user views multiple listings after a search (i.e. repeatedly clicking back in their browser and clicking on different listings in the search results), then all of those actions should be counted. View item counts at child categories should be aggregated up the category tree to parent categories.
  • Supply Histogram Supply Frequency This histogram will associate with each keyword (i.e. search query) the number of listings available on the site on a per category basis. This histogram should be built by determining the number of listings (all formats) whose listing title and description match the given keyword available across all categories (i.e. execute a search). This data should be broken out on a per category basis and the number of listings at child categories should be aggregated up the category tree to parent categories.
  • This histogram will associate with each keyword the frequency with which users bid or BIN on listings related to those keywords on a per category basis.
  • Figure 5 illustrates a graph showing association of bid amounts to query term. For each listing that a user bids on after performing a search then the category ID of the listing, as well as the bid amounts (absolute and differential), are associated with the query term and the count of bid actions for that particular query term-category ID combination is incremented by I. The event should be tracked regardless of how it was generated (i.e. manual bid or proxy bid). Bid counts and amounts at child categories should be aggregated up the category tree to parent categories.
  • FIG. 6 illustrates a graph showing association of BIN amount to query term. For each listing that a user BINs on after performing a search then the category ID of the listing, as well as the BIN amount (absolute), are associated with the query term and the count of BIN actions for that particular query term-category ID combination is incremented by 1. BIN counts and amounts at child categories should be aggregated up the category tree to parent categories.
  • FIG. 7 illustrates a graph showing association of product reference ID to query term. For each listing that a user views after a search, the product reference ID of the listing, if available, is associated with query term and the count of view item actions for that particular query term-product ID combination is incremented by 1. If a user views multiple listings after a search (i.e. repeatedly clicking back in their browser and clicking on different listings in the search results) then all of those actions are counted, if a product reference ID is available. If no product reference ID is associated with the listing then no event is registered and not stored in the histogram.
  • Product reference ID counts at child categories should be aggregated up the category tree to parent categories. Only the top n products within each category need to be stored. The rest of the data accumulated by the query index does not need to be held in the Keyword Store since, for Keyword Extractor purposes, we will only be interested in suggesting the top few products associated with a keyword and/or the products that show the greatest acceleration in search to view item clicks.
  • the clickthrough rate can be calculated for any asset that is made available in ads via the Contextual Editor Kit.
  • CTR clickthrough rate
  • the system can be influenced to rank assets with better CTRs higher and produce better results. Note that there can be multiple CTRs for a given keyword due to the various types of impressions and clickthrough events.
  • Ad CTR (# of click events of all types / # ad unit impressions) This tracks the clickthrough rate of the listing section that is
  • Listing CTR surfaced for a specific search (keyword + category).
  • Listing CTR (# of listing clickthroughs / # listing impressions) This tracks the clickthrough rate of the product suggestion that is surfaced for a specific keyword.
  • Product CTR (# of product clickthroughs / # of product impressions)
  • Category CTR (# of category clickthroughs / # of category impressions)
  • Search Term CTR (# of search term clickthroughs / # of search term impressions)
  • Search Box CTR (# of search box clickthroughs / # of search box impressions) Editor Kit Bid Histogram
  • FIG. 8 illustrates a graph showing association of Bid amounts to ad query term.
  • the data should be attributed back to the keyword impression (not any intermediate keywords they searched on). This histogram is similar to the one described above, but differs in one key data point.
  • the data should be attributed back to the keyword impression that was made in the ad placement on an affiliate site (not any intermediate keywords a user searched on during their session) while the category ID and dollar amounts are derived from the particular listing.
  • the various other feedback tracking dimensions described herein need to be captured as well.
  • Figure 9 illustrates a graph showing association of BIN amount to query term.
  • the system needs to analyze feedback events in aggregate.
  • the financial value driven by a particular site ID x URL x [keyword + category] combination will be tracked and aggregated on a regular basis.
  • the aggregate data is then used to refine the financial expectation and probability score.
  • Various embodiments can accumulate data over the course of regular intervals. Calculate Weekly Financial Value
  • This component will periodically update the histograms stored in the Keyword Store from various data sources.
  • the Keyword Store should be refreshed on a regular basis.
  • a blacklist should be used to filter keywords, based on precise match, prior to the keywords entering the Keyword Store. This will prevent these keywords from being recommended (assuming the Keyword Store gets rebuilt and not incrementally updated). There will be an additional blacklist check in the Editor Kit Server Front- End in order to catch keywords that are added to the blacklist between refreshes of the Keyword Store.
  • the query index will use data generated by various embodiments to find queries entered by users searching on the Host and aggregate metrics about the user's actions after they obtain their search results and associate those aggregate metrics to the query terms.
  • the user's search and related activity must all occur within the same user session.
  • the query index should support all languages used by on-platform sites.
  • the index should be refreshed periodically (i.e. ideally daily refresh, minimum weekly refresh) in order to accurately track changes in aggregate user actions.
  • the system will be extensible to incorporate additional sources of data to be used in the keyword extracting and ranking algorithms.
  • Keyword Data Mart KWDM
  • the metrics will be used in the token scoring algorithm to influence the final ranking of keywords towards those keywords that are expected to generate more revenue. If the Keyword RPC metric is not available for a keyword during the ranking process, then the ranking algorithm should use the other keyword metrics available.
  • the first class is feedback events. These events will allow the system to improve its performance, over time, by incorporating actual user activity into the system.
  • the second class is performance tracking events. These are for internal reporting purposes in order to evaluate the system's performance.
  • the tracking system should be extensible enough to add different events for different use cases or to track new performance measures in the future.
  • the feedback events will be aggregated on a regular basis by a batch job in order for easier consumption by the Keyword Store Builder.
  • the events should be tracked only for users that are shown an impression and clickthrough on one of those impressions and navigate the site. Events should be tracked at 100%. Note that some of the tracking information will only be available at impression time but will need to "follow the user" as downstream tracking events are generated. The following are the individual measures that should be tracked when events occur.
  • # Impressions The number of times that an asset is displayed to a user.
  • # Clickthroughs The number of times that a user clicks on an asset.
  • # View items The number of times that a user, who has previously viewed an asset and clicked through and, has also viewed a listing.
  • # Bids and $ Bids The number of times that a user, who has previously viewed an asset and .clicked through and, has also placed a bid (either manually or by proxy).
  • Asset type Indicates the type of asset that the even is Y associated with.
  • Site ID Identifies the Host site that the event is associated with or was generated on.
  • Application + Identifies the particular consumer application that Y version generated the event.
  • URL Identifies the URL that the event is associated Y with.
  • Domain Identifies the domain that the event is associated Y with.
  • N Search term Identifies the search term that the event is Y (keyword + associated with. This is the original search term category ID) that was surfaced in the ad placement. If a user clicks through on the ad and then proceeds to search using a different term and then views some items because of that, then the original search term would still be used, not the new user query term.
  • Impression Identifies the category that the event is associated Y Category ID with. This is the original category ID that was surfaced in the ad placement. If a user clicks through on the ad and then proceeds to browse to items in a different category, then the original category ID for the ad impression would still be used, not the new category ID.
  • Ad Type Identifies the type of ad (i.e. flash, text + image, N text only) that created the impression that the event is associated with
  • Ad Format Identifies the specific ad format (size/layout) that N created the impression that the event is associated with
  • Ad Color Theme Identifies which particular color theme out of the N fixed set available was used.
  • Custom Display For the custom ad format, the information N Options displayed is controlled by the affiliate (i.e. image, title, price, # bids, time left). Tracking this will enable reporting on what the most popular combinations are.
  • the user hint / category hint is provided by the publisher of the web page in order to refine the search results that are provided.
  • Display Rank Identifies the display rank of the asset that user N clicked on that the event is associated with.
  • the display rank is on a per-asset basis (i.e. if two types of assets are displayed then there will be at least two display ranks of 1 ).
  • Keyword Rank The actual ranking of the keyword associated N with the asset being displayed.
  • Algorithm + Identifies the extractor and ranking algorithm that N version were used to generate the results
  • Bid/BIN category Identifies the category that the bid/BIN event N ID occurred in (allows for analysis to determine what factors drive bid/BIN activity)
  • the objective of the feedback tracking is to determine which assets (i.e. searches, listings, products, categories, etc.) are actually performing the best and to use that data to influence the ranking of various assets.
  • assets i.e. searches, listings, products, categories, etc.
  • Impression Events If ad unit is requested, the system should register one or more of the following impression events (the number and type of impressions is dependent upon what results are displayed).
  • a user interface in an example embodiment is presented in Figure 10 as a generic mockup showing impressions generated from one ad unit.
  • Ad Unit This is the parent ad unit that may Always once each Impression contain one or more sub-sections (and time the ad is served. therefore cause additional impression type events). This impression event will registered once everytime an Editor Kit sniplet is viewed on a page.
  • the number of listings displayed does displayed from two not affect the number of impressions different searches registered. then two impressions events should be registered.
  • search term search term (keyword + category) Once for each unique Impression is displayed then a search term search term (keyword impression should be registered. + category).
  • Impressions may need to be differentiated from each other in some embodiments. For example, there are multiple types of search on the Host (core search, product based search, three variations of Stores search).
  • clickthrough Description Type Whenever a user viewing an impression actually takes and action and either clicks through or executes a search then the clickthrough event should be registered. All clickthrough events should be counted. There will be no filtering of repeat clicks from the same user as there will be no issues with click fraud.
  • the clickthrough to the corresponding DCP should be Clickthrough registered.
  • Search Box The clickthrough to the corresponding SRP should be Clickthrough registered. This should be registered as a different type of event than a search term clickthrough.
  • Event Type Description Bids Whenever a user bids on a listing then the following information should be registered:
  • the bid event itself (so the total number of bids can be aggregated)
  • the differential amount of the bid The absolute amount of the bid
  • the proxy bid information should be tracked even if the user's session ends
  • the BIN event itself (so the total number of BINs can be aggregated)
  • the absolute amount of the BIN (ignore snipping costs)
  • Performance tracking events generate data that will allow us to analyze various aspects of system performance.
  • Host logo either graphic or text
  • a Host logo impression logo is displayed. should be registered.
  • Performance tracking data should be accessible so that ad-hoc queries can be run in order to determine how the Contextual Editor Kit ad placements are performing. Since the analysis will be mostly ad-hoc the types of queries and aggregations will not well-known in advance. Batch Job to Aggregate Tracking Data
  • This batch job takes the detailed level tracking metrics for the feedback system and aggregates them into a useful form that can be used to complete the feedback loop.
  • the batch job should aggregate data on a weekly basis.
  • the primary purpose is to separate the data out at a keyword-URL level so when the aggregated data is folded back into the ranking algorithm that keywords on URLs that have higher clickthrough rates or higher onsite activity will have those keywords rank higher as time progresses.
  • the feedback data should only be used if a statistically adequate number of data points (i.e. impressions) for the given keyword- ⁇ URL
  • the feedback data should be aggregated according to the dimensions described herein. Each week of aggregate data should be maintained so that historical trends can be taken into account during the ranking of tokens.
  • the aggregated keyword level data should be made easily accessible. It may be advantageous to pull the entire dataset on occasion to analyze the data for various correlations to paid search performance.
  • Keyword Extractor currently suggests keyword+category combinations only if the keyword appears on a page that it processes. KE cannot suggest keywords that do not appear on the page even if they are highly relevant to the context of the page.
  • KE can gain the capability to suggest new and relevant keywords that do not appear in the original content of a page.
  • the current Related Searches feature works by aggregating data about consecutive searches performed by users. This data allows the website to know that users who search for, say "harry potter”, are also likely to search for "harry potter wand” and “harry potter dvd” afterwards. These follow-on searches are automatically suggested whenever a person types in a search on a particular host.
  • KE combines Related Searches data with the keyword data extracted from a page. This could be done in a variety of circumstances, such as when page content is sparse, the page content is of the type that does not yield an adequate number of keyword matches or when related searches may yield better relevance or have a higher expected return. For instance, the top ranked keyword+category combination extracted from a page would be used to look up all the related searches for it. Then the related searches returned would be scored according to their relevance and expected return (although, since the related searches do not appear on the page they will not have any context data which will need to be compensated for during scoring; this could involved the use of Related Searches clickthrough rate data). The keyword+category combinations and related searches can then be combined and rank ordered as usual. The combined data can then be passed to various applications for use, such as AdContext.
  • affiliate a person or company who drives traffic to the Host via advertising placements. affiliates are compensated by a revenue share and ACRU commissions.
  • Host AdContext an affiliate ad that will utilize the Keyword Extractor (KE) backend to provide contextual ad content.
  • KE Keyword Extractor
  • Various embodiments utilize the Keyword Extractor (KE) service for two front-end applications.
  • KE Keyword Extractor
  • EK Editor Kit
  • Syndicated content allows developers to access contextual data from the
  • the Keyword Extractor service can be used to automatically generate contextual EK listings.
  • the Editor Kit has been augmented to provide the Contextual Advertisement Generator.
  • the EK is an ad used by affiliates to direct traffic from their web site to the Host. • affiliates earn revenue share and ACRU commissions from the traffic they drive to the Host.
  • the EK dynamically displays listings from keyword search terms specified by the affiliate at set-up.
  • Keyword Extractor Various embodiments have a direct dependency on the Keyword Extractor functionality described above. Refer to the Keyword Extractor description herein for details on the following issues. • Consumer application - use cases for the Keyword Extractor including the
  • an EK to KE interface interaction may occur.
  • This interaction may include a real-time preview, as detailed in Figure 12 for an example embodiment. Performance requirements for the real-time preview are noted in the KE description above.
  • the dataset returned to the EK is then used to generate relevant listings as outlined above.
  • the Javascript code is then embedded into a web page's source code.
  • the ad is then considered published.
  • the ad will load once the web page is rendered, and the EK to KE interface interaction as shown in Figure 13 may occur.
  • the URL used for a published ad is the web page that the ad is placed.
  • the URL submitted by the affiliate at creation should not be used after the preview.
  • Both the category ID(s) and site ID are the parameters specified by the affiliate at creation.
  • the KE service will return the top overall keywords if there is a Fetch error or if no keywords are found. The top keywords are described in the KE description section above.
  • Data Warehouse Reporting A subset of the data tracked by various embodiments herein will be tracked by the Data Warehouse (DW). This data will be used to evaluate the performance of the Contextual Advertiser. The events and dimensions listed below have been defined in the KE description above. The following events may be tracked in the DW. 1 ) # of Impressions 2) # ofClickthroughs
  • the DW should track the above events by each of the following dimensions. 1 ) Asset Type T) Site ID
  • Commission Junction CJ
  • Mediaplex Mediaplex Tracking pixels for both Commission Junction (CJ) and Mediaplex should be added to all ads. Commission Junction (CJ) and Mediaplex are well-known advertising-related services. These pixels will trigger CJ and Mediaplex to record an impression each time an ad is rendered.
  • PID# should be replaced with the PID provided by the affiliate when the ad was created.
  • AID# should be replaced with the corresponding # for each site
  • ROTATION_ID should be replaced with the corresponding # for each site
  • Random Number Generator (at least 4 digits)
  • the Cache Buster value must be identical on all ad tag components for each ad placement. Proper cache busting ensures the correct clickthrough URLs are delivered, and ensures correct banner weighting/rotation. Debug Mode
  • Keyword data Keyword, rank, score, display rank, etc.
  • Category data Category ID, category name, score, etc.
  • any other metadata related to the URL and captured by Keyword Extractor i.e. last refetch time, next refetch time, etc..
  • the debug information should be fo ⁇ natted to be easily readable within the XML output
  • Partner contingency functionality can be supported as follows, o
  • a PC wire off option can be added to the KE integration with the CA. o This will preclude the CA from relying completely on the Keyword
  • Keyword Extractor to function. o
  • the CA will only allow the manual method of choosing keywords.
  • AU data from the Keyword Extractor will be character encoded in UTF-8 in an example embodiment.
  • the Editor Kit must do the necessary conversion to match the EK content with the character encoding of the affiliate's URL.
  • the keywords used by the EK should be scrubbed of the blacklisted keywords from the sources listed below.
  • the excluded keywords should be based on a precise match.
  • the most current keywords on the SIBE blacklist should be excluded from the keywords used for the EK.
  • affiliate-specified Exclusions affiliates may specify a keyword exclusion list when creating their EK.
  • the generated search listings should use the current Editor Kit functionality. Below are the current defaults used by the EK (mirrors search front-end default). • Primarily searches Host core, both auctions and buy it now listings. • Store listings are used as a back-up if no core listings are available.
  • EK-specific default Only listings with time left greater than some time limit (e.g. two hours) should be included. Make this parameter configurable to a shorter time without the need for a train roll.
  • the keywords and categories from the KE service will be used to populate the EK with relevant search listings. Use the slot selection process below to select which keywords generate each listing.
  • Each [keyword + category] combination has a value assigned to it. This value indicates the probability that this keyword can be selected for a given ad slot.
  • a game For each ad slot, starting with the first slot, a game must be played to determine which [keyword + category] from the entire set will win the ad slot. Once a [keyword + category] combination wins a slot, it is removed from consideration, for the current instance of the ad, and the game is played for the next slot and the remaining [keyword + category] combinations left in the set.
  • V(KW n ) R(KW n ) is the range of the n [keyword + category]
  • Figures 14-17 illustrate a flowchart to diagram the slot selection process in an example embodiment. If the ad layout is the tab-based Flash ad, the same selection process should be used to generate the keywords.
  • EK ads there are three types of EK ads as follows. • Graphic and text
  • graphic and Text Ads incorporate the following content in an example embodiment.
  • Viral marketing link Graphic and text ads are available with a fixed color scheme as pictured in Figure 18 in an example embodiment.
  • Graphic and text ads are also available as customizable ads. These ad designs do not include the horizontal stripe at the bottom. Examples of customizable ads are pictured in Figure 19 in an example embodiment.
  • Text only ads will not include any graphics including listing images, logos or the horizontal stripe.
  • the default color scheme for the text-only ads may be as follows.
  • Flash 8 can be used to build the Flash ads. This version of Flash should allow .gifs to appear in the Flash ad.
  • Tab-based Flash Ad There may only be one type of Flash ad created for a particular embodiment.
  • Flash ads will always use the "Top Keywords" as the fall back for null search results. Most- watched items will not be an available fall back option for null search results. Below is a table of the behavior for the different sizes
  • Flash ad It should be determined if the user who is viewing the Flash ad has Macromedia Flash 8 installed on their machine. If a user does not have Flash 8 installed, they will not be able to see the ad correctly. A contextual graphic and text ad should be displayed as the back-up for the Flash ad.
  • One embodiment may do a silent downgrade (detailed below) to a text and image ad when Flash isn't installed on the user's machine.
  • the size selection has expanded to accommodate the new sizes defined by the Interactive Advertising Bureau (IAB).
  • IAB Interactive Advertising Bureau
  • affiliates can customize the size of a graphic/text or text-only ad. Flash ads are only available at the fixed sizes.
  • the custom size ad uses the existing functionality that the EK create page currently allows. The following parameters are configurable in an example embodiment.
  • Single listings are stacked vertically in the ad layout.
  • the ad will flex to the specified width. Keywords and categories can't be added to the custom size layout.
  • FIG. 22 illustrates fixed color theme examples 5 in an example embodiment.
  • affiliates can customize the colors of a graphic/text or text-only ad.
  • the configurable color options in an example embodiment include the following. 10 • Background color
  • Colors can be specified by hex number or by choosing a shade on the color wheel.
  • Figure 23 illustrates an Ad Components Legend in an example embodiment.
  • AU links should open up the landing page in a new window.
  • the product image for a listing should be resized according to the ad size specifications. No product images should appear in the text-only ads. Product images should link to the View Item Page (VIP) for that featured listing.
  • VIP View Item Page
  • the listing title should display at a maximum of 3 lines. Any titles that are longer than 3 lines will be truncated with an ellipses defined at the end. Shorter listing titles can appear as just 1 or 2 lines.
  • the title text should link to the View Item Page (VIP) for that featured listing.
  • VIP View Item Page
  • Price Display For Buy It Now items the price should be shown as "Price $x.xx”.
  • Time Left should display according to the following specifications.
  • Search Box Searches made in the search box will use the same default parameters used for searches from the Host homepage. Parameters include the following.
  • Core searches are the default. Store searches only appear if no core results available.
  • Sorting should default to listings ending the soonest displaying first.
  • the system includes affiliate tracking for the EK search box.
  • a search is initiated when the user clicks the Go button.
  • a search can also be initiated by hitting the ⁇ enter> key.
  • the search box may be pre-populated with "Search Host". When the user clicks on the search box, this text should disappear and a blank box with a text cursor should remain.
  • Those ads that include a categories section should display categories in the following order of priority. • Categories specified by the affiliate when creating the EK.
  • FIG. 24 An example of a Host Tools page is illustrated in Figure 24 for an example embodiment. Create Your Own Ad page
  • FIG. 25 An example of a "Create Your Own Ad page" is illustrated in Figure 25 for an example embodiment.
  • the user interface (UI) mock-ups in this document are for a visual example only.
  • Host Header as noted in the figures will remain as it does on the current page. Included elements — Host logo, site navigation, search, user greeting, breadcrumb trail, Sun logo.
  • Content selection determines if the new Editor Kit will use the Keyword Extractor backend (Automated) or the manually-selected keywords (Manual).
  • Automated Automated is the default selection in the Content Selection drop-down.
  • Keyword Extractor will fetch keywords from this URL for the real-time preview.
  • Keywords can be input as separate keywords. o These keywords will be comma-delimited.
  • the backend processing will rank the keywords in ascending order, based on the order the affiliate entered them in the keyword field.
  • Each keyword is then filtered by the category filters provided by the affiliate.
  • a user types in harry potter, eragorn. The user then selects Books > Children's Books and DVDs as two category filters. The backend processing will play the game with the following ranked keyword +category combinations.
  • affiliates may choose categories to filter their ad content. These selected categories serve two functions.
  • a pop-up layer may be used for affiliates to select categories.
  • This HTML pop-up layer is similar to an AJAX pop-up layer used to choose categories.
  • Figures 27 and 28 illustrate category selection in an example embodiment.
  • All selected categories may appear on the main page after the AJAX window has been saved and closed. These selected categories should include the breadcrumb paths. See the selected category display mock-up illustrated in Figure 29 for an example embodiment.
  • a pop-up layer is used for affiliates to select advanced options. See the Tools in the "Create Your Own Ad” page section.
  • a filter can be added to allow for only charity listings to be displayed. This filter will be triggered when a checkbox is selected. The charity checkbox should not be selected by default.
  • a charity ID can be specified to narrow the displayed listings to a specific charity. There is currently an API that allows for charity- specific listings to be returned when a charity ID is supplied.
  • the charity ID filter will not be configurable in the 'Create Your Own Ad' page. The default will be to include all charities.
  • AJAX window has been saved and closed. See the mock-up illustrated in Figure 30 for an example embodiment.
  • the preview button triggers the generation of ads in the preview area. As stated earlier, if "Manual” is selected, the current EK functionality is used. The KE service is not called. The Refresh button must be clicked in order to update the preview ads with any user changes.
  • the reset button will clear all fields populated by the affiliates. This includes any options selected in the Advanced Options and Category Selector popup windows. All fields should revert back to their default state.
  • the reset button should trigger a warning message pop-up. An affiliate must click on OK before the reset is completed.
  • the reloaded pages with error messaging should retain any selections and inputs the affiliate has already made. See the mock-up illustrated in Figure 31 for an example embodiment.
  • the preview area is where the ads are displayed with real-time content. affiliates can select color and size in this area. They can copy the code directly from this section into their source code. See the mock-up illustrated in Figure 32 for an example embodiment.
  • a drop-down of available color themes will display at the top of the Preview section.
  • the color scheme drop-down will not appear for Flash ads.
  • the default selection will be the Host Stripe. Custom Color/Font will be the last option in the
  • the color picker appears when a user is selecting a color in the custom color or custom ad sections.
  • a user can perform the following actions.
  • the popup layer opens near where the mouse is located.
  • Font When the custom color theme is chosen, a "Font" drop-down should also appear. The font options should appear in a drop-down.
  • a drop-down of all available sizes will display.
  • the default size in the drop- down will be "All”.
  • the last listed size in the drop-down will be "Custom Ad”.
  • the page will scroll down to that specific size. "Custom Ad” will not appear in the size drop-down for Flash ads.
  • Figure 35 illustrates an example of a color/size drop down in an example embodiment.
  • Custom Size (show only with custom ad)
  • Custom Title (show only with custom ad)
  • Show Column Limitations by size The table below outlines which "show columns" are available by size.
  • the first column indicates the minimum width needed to fit the X'd show columns on the right. For example, if a user chooses the width of their ad to 340 or above but below 455, they are limited to 45 characters in their title, a gallery image, title and current price only.
  • the checkboxes for the show columns are used to remove any show columns that the user does not want to appear in the ad. The remaining ad content will expand — no empty columns should appear. .
  • Custom Ad When Custom Ad is selected, only an ad template will appear in the preview section. This template should reflect the real-time changes that are being made in the custom fields above.
  • This preview template is identical to the existing functionality of the current EK ad. None of the new ad formats can be used when the custom ad is chosen. The updated Javascript code should be displayed next to the ad template.
  • the custom ad title is user text that can be added to a custom ad. This is existing functionality that the EK currently offers.
  • the title text is added below the Host logo in the ad (see screenshot below).
  • the default text color is black (#000000). 90 characters maximum are allowed.
  • the title text may wrap depending on the size of the ad. Further customization is allowed.
  • An example of custom ad title is shown in Figure 36 for an example embodiment.
  • o Width Please enter a width between 120 - 800 pixels.
  • o # of items The # of items should be between 1 - 200 items.
  • o Min Height Minimum height cannot exceed 20000 pixels.
  • the corresponding Javascript for each ad should appear next to the ad preview.
  • the Copy Code button will serve as a "hot key” to automatically highlight the code and trigger the copy function (ctrl-c).
  • the URL included in the JS should have the tracking information embedded.
  • An example of the Create Your Own Ad page is shown in Figure 37 in an example embodiment.
  • Asynchronous JavaScript and XML 5 or AJAX can be used to enable features to update the page without refreshing the entire page.
  • Features can leverage AJAX and retrieve new data from the server asynchronously without stalling the affiliate's interaction with the page. The result is a more responsive interface, since the amount of data interchanged between the web browser and web server is vastly reduced.
  • Example application When the seller selects a meta category on the Browse
  • the system will post the seller's selection to the sever and fetch the list of L2 categories for that the selected meta without reloading the page.
  • the information exchanged between the client and the server will be kept to the minimal to minimize load time.
  • Dialogs will be movable if the user clicks and drags the title bar area.
  • a sniffer When a user lands on the "Create Your CA" page, a sniffer should detect if the user has turned off JS on their machine. If the user does not have JS enabled, a warning window should pop up directing the user to enable JS in their browser options.
  • Keyword Extractor service There is an API call that will allow affiliates to access the Keyword Extractor service dataset.
  • the keywords that the API returns should be scrubbed against the Host Blacklist.
  • a configurable parameter should be added to limit the number of keywords returned. The default will be set at no limit. This limit should be configurable internally without a train roll. It should not be configurable by external developers using the API.
  • the API will interface with the Keyword Extractor service. The real-time preview should not be accessible via the API.
  • Host AdContext may only display content from the site ID of the Host site in which the ad was created.
  • the ad may not take into account the site ID of the user who is viewing the ad. For example, if a German user views an American website, the Host AdContext should display a .DE items and links.
  • the location of a user viewing a Host AdContext advertisement generator (eAC) created advertisement should be identified and used to generate more targeted content.
  • an IP address can specify which country, region or city that a user is located.
  • a mapping of location to relevant site ID can either be done via a pre-defined lookup table or whatever automated method that already exists (e.g. from cross border trade logic)
  • Geo-targeting should be added to the Advanced Options section of the Create flow. Multiple site IDs can be supported for geo-targeting. The default site ID will be set to the site ID of the create flow (e.g. US create flow will have a default of 0). The default should be configurable by the affiliate. Geo-targeting must be explicitly enabled by the affiliate. Tracking geo-targeted impressions, clicks, conversion metrics, etc. should also be captured by site ID and by tracking provider.
  • the KE should use the viewer's location to determine which site ID to use for generating keyword/category recommendations. For example, if a viewer is identified as being in France, the Keyword Extractor will use Host.fr's supply, demand and conversion data to determine the most relevant keywords. The first time a viewer from a new site ID visits the affiliate page, top keywords from the new site ID are returned. The affiliate page for the new site ID is added to the queue to be scanned for keywords. This is how the KE currently functions when a user visits an affiliate URL for the first time. Geo-targeting is only turned on for site IDs that have been set up by the affiliate in the Create flow.
  • affiliates can create ads featuring products with the Product Kit (non- contextual). This feature will add product data to the Host AdContext. There should be different levels of how product data is displayed in the ads. The affiliate should have the flexibility to choose which format they prefer, such as the following.
  • a new product-specific histogram can be added to the current KE functionality.
  • a site product Reference ID histogram will associate with each keyword the frequency with which users click through to an item which is associated with specific product on a per category basis.
  • Figure 38 illustrates a graph showing association of product reference ID to query term in an example embodiment. For each listing that a user views after a search, the product reference ID of the listing, if available, is associated with query term and the count of view item actions for that particular query term-product ID combination is incremented by 1. If a user views multiple listings after a search (i.e. repeatedly clicking back in their browser and clicking on different listings in the search results) then all of those actions are counted, if a product reference ID is available.
  • the Host AdContext may only allow an affiliate to choose either Keyword Extractor recommendations or a static list of keywords provided by the affiliate. This feature will allow a hybrid of both sources to drive ad content.
  • An affiliate can use the automated Keyword Extractor to drive content, in addition to providing keywords. These keywords can be used in the following ways.
  • the system can detect if multiple eAC units are displaying on one page.
  • the system can ensure that the items displayed across multiple ads are not duplicated.
  • the system can allow the user to differentiate between different placements, either on the same page or on different pages in their reporting.
  • the feedback loop will need to separately track the impressions, clicks and conversion metrics of Express domains and keywords.
  • RTM provides valuable data on user preferences and previous purchases. This information can further target Host ads to offer compelling recommendations to the user.
  • Various embodiments described herein add personalization to ads in order to increase click-through and also personalizing Host landing pages to increase conversion. Host users visiting non-Host sites can be identified (e.g. cookie sniffing) and matched with their Host user profile.
  • Various embodiments provide the following basic features.
  • Host AdContext uses keyword and category recommendations from the Keyword Extractor (KE) to decide what ad content to display. These recommendations are contextual to the page that the ad is placed.
  • KE Keyword Extractor
  • KE recommendations can be enhanced using data to incorporate user data as an additional weighting factor when choosing ad content.
  • This data can include the following. • Previous buying habits
  • the Leap Pad + baby combination will be weighted heavier in the listing selection process.
  • a user's Host preferences can also be used to further refine the items featured in the ad (e.g. BIN-only, location, etc.)
  • Keywords can be expanded by related searches logic.
  • Categories can be expanded by related categories logic.
  • Cluster Recommendations Cluster profiles can be built for users with matching attributes - segments.
  • An embodiment can build a customized search results landing page with the following features.
  • Figures 39-40 illustrate a step-by-step process in an example embodiment of how the Host AdContext component selects item listings to display based on the Keyword Extractor recommendations.
  • POST-SEARCH ACTIVITY ANALYSIS Detection of an end-user interaction with a Web page, generating keywords based on the end-user interaction with the Web page, and performing a search using the keywords is described.
  • a program running on a client computer may monitor the end-user's browsing activities on the Internet.
  • the end-user clicks on a link information associated with the link, such as the link's label, may be parsed and the Post-Search Activity analysis used to generate one or more keywords.
  • the keywords may be used to perform a search to generate search results which would be the feedback responsive to the keywords.
  • the post-search activity analysis maybe combined with inventory/listing available on the site.
  • a program monitors a -user's web browsing activity.
  • the client program may be configured to trigger delivery of search results to the client computer when the user goes to a website in a category of websites.
  • the search results may be responsive to one or more keywords derived from addressing information, such as the uniform resource locator (URL) of a website visited by the user.
  • URL uniform resource locator
  • a keyword extracting device which extracts keywords by, a) monitoring the content change using frequency of spidering activity, b) using a pattern processing means and c) creating a fingerprint of content to monitor frequency of change is described.
  • Keyword extractor Detection of an end-user interaction with a Web page, generating keywords based on the end-user interaction with the Web page, and performing a search using the keywords is described.
  • This is a system wherein said keyword extractor ' comprises plural and selectable extraction algorithms customized for different situations/content.
  • the keyword extractor can monitor events (e.g., disaster vs. positive event) and selectively determine when to contextualize.
  • the API FOR USER PROVIDED URL Detection of an end-user interaction with a Web page, generating keywords based on the end-user interaction with the Web page, and performing a search using the keywords is described.
  • the search results may be responsive to one or more keywords derived from addressing information as opposed to full text, such as the uniform resource locator (URL) of a website visited by the user, hi general, there is no direct end-user interaction with a web page URL.
  • the API's described herein allows a 3rd party programmer to utilize the KE service directly and receive the unformatted data to use as s/he sees fit.
  • a URL is provided to the URL, but there may or may not be an end-user visiting that URL at that point in time the API is invoked.
  • the system wherein said keyword extractor leverages on an on site feature called related searches.
  • the related searches may be search results presented by a program including content that relates to a keyword employed by the user in the request for information. Because the keyword represents the user's interest in a topic, the chance of the user being interested in the related search content is increased.
  • Contextual information to be delivered to the end user is selected using the keyword information.
  • the contextual information corresponds to ad information which is provided by a campaign provider.
  • the contextual advertiser uses keyword recommendations and categories from keyword extractor to customize the ad unit to have new ads that are flash, text and rich media advertisements. By using scores to statistically pick advertisement slots for keywords this invention gives them exposure, builds traffic and uses them optimally.
  • a system and method for processing a request using a contextual product toolbar comprises the steps of a) sensing context on page and return keyword to be inserted into toolbar search box b) indicating insertion to user such that a pre-populated search box is arrived at, c) including call to action with highlighting (e.g...., mouse-over, smart tag) to highlight contextual keywords on page, and d) retrieving items and populating toolbar.
  • This functionality would either utilize an API call or directly call the KE service.
  • a system and method for ad placements in different media and forums like blogs, Emails, RSS and Wikis is described where the contextual information corresponds to ad information provided by a campaign provider.
  • a system and method for generating cross asset/site suggestions which can include a product review or a product guide for a keyword or a score for a keyword is described.
  • An important feature of this invention is the ad which suggests items that are monetized best based on the feedback, which is contextually sensitive and is not limited to listings.
  • a system comprising: a plurality of Web pages; a set of automatically generated suggested tags, each of which is associated with an indication of either the content of a Web page, or its relevance to certain search engine queries, or both; suggested tags associated with each of the plurality of Web pages, wherein the tag has been associated with the Web page according to the preference of a user is described.
  • PERSONALIZATION OF KEYWORD SUGGESTIONS Detection of an end-user interaction with a Web page, generating keywords based on the end-user interaction with the Web page, and performing a search using the keywords is described.
  • the systems and methods include a mechanism for providing interest and demographic data that may be applied to filter the Web page at the provider side resulting in personalization of keyword suggestions.
  • the extracted keyword values are applied to filter content for delivery to a requesting Web client.
  • a diagram illustrates a network environment in which various example embodiments may operate.
  • a server computer system 100 is coupled to a wide-area network 110.
  • Wide-area network 110 includes the Internet, or other proprietary networks, which are well known to those of ordinary skill in the art.
  • Wide-area network 110 may include conventional network backbones, long-haul telephone lines, Internet service providers, various levels of network routers, and other conventional means for routing data between computers.
  • server 100 may communicate through wide-area network 110 to a plurality of client computer systems 120, 130, 140 connected through wide-area network 110 in various ways.
  • client 140 is connected directly to wide-area network 110 through direct or dial-up telephone or other network transmission line.
  • clients 130 may be connected through wide-area network 110 using a modem pool 114.
  • a conventional modem pool 114 allows a plurality of client systems to connect with a smaller set of modems in modem pool 114 for connection through wide-area network 110.
  • wide-area network 110 is connected to a gateway computer 112.
  • Gateway computer 112 is used to route data to clients 120 through a local area network (LAN) 116.
  • LAN local area network
  • server computer J 00 can communicate with client computers 150 using conventional means.
  • a server computer 100 may operate as a web server if the Internet's World-Wide Web (WWW) is used for wide area network 110.
  • WWW World-Wide Web
  • web server 100 may communicate across the World-Wide Web with clients 150.
  • clients 150 use a client application program known as a web browser such as the Internet ExplorerTM published by Microsoft Corporation of Redmond, Washington, the user interface of America On- LineTM, or the web browser or HTML renderer of any other supplier.
  • clients 150 may access image, graphical, and textual data provided by web server 100 or they may run Web application software.
  • Figures 42 and 43 show an example of a computer system 200 illustrating an exemplary client 150 or server 100 computer system in which the features of an example embodiment may be implemented.
  • Computer system 200 is comprised of a bus or other communications means 214 and 216 for communicating information, and a processing means such as processor 220 coupled with bus 214 for processing information.
  • Computer system 200 further comprises a random access memory (RAM) or other dynamic storage device 222 (commonly referred to as main memory), coupled to bus 214 for storing information and instructions to be executed by processor 220.
  • Main memory 222 also may be used for storing temporary variables or other intermediate information during execution of instructions by processor 220.
  • Computer system 200 also comprises a read only memory (ROM) and /or other static storage device 224 coupled to bus 214 for storing static information and instructions for processor 220.
  • ROM read only memory
  • An optional data storage device 228 such as a magnetic disk or optical disk and its corresponding drive may also be coupled to computer system 200 for storing information and instructions.
  • Computer system 200 can also be coupled via bus 216 to a display device 204, such as a cathode ray tube (CRT) or a liquid crystal display (LCD), for displaying information to a computer user. For example, image, textual, video, or graphical depictions of information may be presented to the user on display device 204.
  • CTR cathode ray tube
  • LCD liquid crystal display
  • an alphanumeric input device 208 is coupled to bus 216 for communicating information and/or command selections to processor 220.
  • cursor control device 206 such as a conventional mouse, trackball, or other type of cursor direction keys for communicating direction information and command selection to processor 220 and for controlling cursor movement on display 204.
  • the client 150 can be implemented as a network computer or thin client device.
  • Client 150 may also be a laptop or palm-top computing device, such as the Palm PilotTM.
  • Client 150 could also be implemented in a robust cellular telephone, where such devices are currently being used with Internet micro- browsers.
  • Such a network computer or thin client device does not necessarily include all of the devices and features of the above-described exemplary computer system; however, the functionality of an example embodiment or a subset thereof may nevertheless be implemented with such devices.
  • a communication device 226 is also coupled to bus 216 for accessing remote computers or servers, such as web server 100, or other servers via the
  • the communication device 226 may include a modem, a network interface card, or other well-known interface devices, such as those used for interfacing with Ethernet, Token-ring, or other types of networks.
  • the computer system 200 may be coupled to a number of servers 100 via a conventional network infrastructure such as the infrastructure illustrated in Figure 41 and described above.
  • the system of an example embodiment includes software, information processing hardware, and various processing steps, which will be described below.
  • the features and process steps of example embodiments may be embodied in articles of manufacture as machine or computer executable instructions.
  • the instructions can be used to cause a general purpose or special purpose processor, which is programmed with the instructions to perform the steps of an example embodiment.
  • the features or steps may be performed by specific hardware components that contain hard- wired logic for performing the steps, or by any combination of programmed computer components and custom hardware components. While embodiments are described with reference to the Internet, the method and apparatus described herein is equally applicable to other network infrastructures or other data communications systems.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Transfer Between Computers (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

La présente invention concerne un système et un procédé mis en œuvre par ordinateur qui permet l'extraction de mots-clés et la production de publicité contextuelle. Dans un mode de réalisation le système comprend un service d'extraction de mots-clés pour : recevoir d'une application de consommateur une demande d'activation de service d'extraction de mots-clés via une API (Interface de programmation d'applications), ladite demande comprenant une identité d'une source de contenu, ainsi qu'une identification d'un processus d'extraction particulier qui sera utilisée par le service d'extraction de mots-clés sur la source de contenu identifié ; déterminer si le service d'extraction de mots-clés a déjà traité la source de contenu identifiée et conserver les mots-clés extraits dans un stockage de données ; extraire des mots-clés de la source de contenu identifiée en utilisant le processus d'extraction particulier identifié dans la demande ; et mettre à disposition des mots-clés extraits à l'application de consommateur.
PCT/US2007/011992 2006-06-09 2007-05-21 Extraction de mots-clés et production de publicité ciblée WO2007145775A2 (fr)

Applications Claiming Priority (8)

Application Number Priority Date Filing Date Title
US80438706P 2006-06-09 2006-06-09
US60/804,387 2006-06-09
US11/646,012 US8001105B2 (en) 2006-06-09 2006-12-27 System and method for keyword extraction and contextual advertisement generation
US11/646,039 US8209320B2 (en) 2006-06-09 2006-12-27 System and method for keyword extraction
US11/645,946 US7831586B2 (en) 2006-06-09 2006-12-27 System and method for application programming interfaces for keyword extraction and contextual advertisement generation
US11/646,012 2006-12-27
US11/646,039 2006-12-27
US11/645,946 2006-12-27

Publications (2)

Publication Number Publication Date
WO2007145775A2 true WO2007145775A2 (fr) 2007-12-21
WO2007145775A3 WO2007145775A3 (fr) 2008-09-18

Family

ID=38832267

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2007/011992 WO2007145775A2 (fr) 2006-06-09 2007-05-21 Extraction de mots-clés et production de publicité ciblée

Country Status (1)

Country Link
WO (1) WO2007145775A2 (fr)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7831586B2 (en) 2006-06-09 2010-11-09 Ebay Inc. System and method for application programming interfaces for keyword extraction and contextual advertisement generation
EP2324436A2 (fr) * 2008-07-15 2011-05-25 Google, Inc. Contextes géographique et de mots-clés dans des applications intégrées
US8001105B2 (en) 2006-06-09 2011-08-16 Ebay Inc. System and method for keyword extraction and contextual advertisement generation
US8209320B2 (en) 2006-06-09 2012-06-26 Ebay Inc. System and method for keyword extraction
CN111581510A (zh) * 2020-05-07 2020-08-25 腾讯科技（深圳）有限公司 分享内容处理方法、装置、计算机设备和存储介质
US11341531B2 (en) 2007-07-09 2022-05-24 Ebay Inc. System and method for contextual advertising and merchandizing based on user configurable preferences

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020128925A1 (en) * 2000-12-11 2002-09-12 Patrick Angeles system and method for detecting and reporting online activity using real-time content-based network monitoring
US20030100131A1 (en) * 2001-11-28 2003-05-29 Winbond Electronics Corp. Dispatching method of manufacturing integrated circuit
US20040143465A1 (en) * 2002-12-26 2004-07-22 Kouji Imanishi Consolidated air cargo management system
US20040249816A1 (en) * 2003-06-03 2004-12-09 Pioneer Corporation Information communication apparatus, information communication method, and information recorded medium
US20050004903A1 (en) * 2002-03-15 2005-01-06 Fujitsu Limited Regional information retrieving method and regional information retrieval apparatus
US20060020596A1 (en) * 2004-06-02 2006-01-26 Yahoo! Inc. Content-management system for user behavior targeting

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020128925A1 (en) * 2000-12-11 2002-09-12 Patrick Angeles system and method for detecting and reporting online activity using real-time content-based network monitoring
US20030100131A1 (en) * 2001-11-28 2003-05-29 Winbond Electronics Corp. Dispatching method of manufacturing integrated circuit
US20050004903A1 (en) * 2002-03-15 2005-01-06 Fujitsu Limited Regional information retrieving method and regional information retrieval apparatus
US20040143465A1 (en) * 2002-12-26 2004-07-22 Kouji Imanishi Consolidated air cargo management system
US20040249816A1 (en) * 2003-06-03 2004-12-09 Pioneer Corporation Information communication apparatus, information communication method, and information recorded medium
US20060020596A1 (en) * 2004-06-02 2006-01-26 Yahoo! Inc. Content-management system for user behavior targeting

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7831586B2 (en) 2006-06-09 2010-11-09 Ebay Inc. System and method for application programming interfaces for keyword extraction and contextual advertisement generation
US8001105B2 (en) 2006-06-09 2011-08-16 Ebay Inc. System and method for keyword extraction and contextual advertisement generation
US8200662B2 (en) 2006-06-09 2012-06-12 Ebay Inc. System and method for application programming interfaces for keyword extraction and contextual advertisement generation
US8209320B2 (en) 2006-06-09 2012-06-26 Ebay Inc. System and method for keyword extraction
US10621240B2 (en) 2006-06-09 2020-04-14 Ebay Inc. System and method for keyword extraction
US11341531B2 (en) 2007-07-09 2022-05-24 Ebay Inc. System and method for contextual advertising and merchandizing based on user configurable preferences
EP2324436A2 (fr) * 2008-07-15 2011-05-25 Google, Inc. Contextes géographique et de mots-clés dans des applications intégrées
EP2324436A4 (fr) * 2008-07-15 2013-09-18 Google Inc Contextes géographique et de mots-clés dans des applications intégrées
US8655761B2 (en) 2008-07-15 2014-02-18 Google Inc. Geographic and keyword context in embedded applications
US11120473B2 (en) 2008-07-15 2021-09-14 Google Llc Geographic and keyword context in embedded applications
CN111581510A (zh) * 2020-05-07 2020-08-25 腾讯科技（深圳）有限公司 分享内容处理方法、装置、计算机设备和存储介质
CN111581510B (zh) * 2020-05-07 2024-02-09 腾讯科技（深圳）有限公司 分享内容处理方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
WO2007145775A3 (fr) 2008-09-18

Similar Documents

Publication Publication Date Title
US10621240B2 (en) System and method for keyword extraction
US8200662B2 (en) System and method for application programming interfaces for keyword extraction and contextual advertisement generation
US8001105B2 (en) System and method for keyword extraction and contextual advertisement generation
US20080221987A1 (en) System and method for contextual advertisement and merchandizing based on an automatically generated user demographic profile
US20220222707A1 (en) System and Method For Contextual Advertising and Merchandizing Based On User Configurable Preferences
US11314926B2 (en) Selectively replacing displayed content items based on user interaction
KR102219344B1 (ko) 자동 광고 대행 서버, 자동으로 광고 매체를 위한 캠페인 정보를 생성하여 광고의 집행을 대행하는 방법 및 상기 방법을 실행하기 위한 컴퓨터 프로그램
US9262766B2 (en) Systems and methods for contextualizing services for inline mobile banner advertising
US8135617B1 (en) Enhanced hyperlink feature for web pages
JP5676557B2 (ja) 編集インターフェース
US20110213655A1 (en) Hybrid contextual advertising and related content analysis and display techniques
US20110015996A1 (en) Systems and Methods For Providing Keyword Related Search Results in Augmented Content for Text on a Web Page
US20130054672A1 (en) Systems and methods for contextualizing a toolbar
US20100241944A1 (en) Method and apparatus for associating advertising content with computer enabled maps
KR102191486B1 (ko) 자동 광고 대행 서버, 자동으로 광고 매체를 위한 캠페인 정보를 생성하여 광고의 집행을 대행하는 방법 및 상기 방법을 실행하기 위한 컴퓨터 프로그램
CN103150352A (zh) 用于生成相关搜索查询的系统
CN102257525A (zh) 用于基于先前捕捉的相关性数据重定向广告的系统和方法
US20110029384A1 (en) System and method for dynamic targeting advertisement based on content-in-view
WO2007145775A2 (fr) Extraction de mots-clés et production de publicité ciblée
US20170178179A1 (en) Modifying advertisement presentations
US11004118B1 (en) Identifying creative offers within online content
WO2013033445A2 (fr) Systèmes et procédés permettant de contextualiser une barre d'outils, une image et un bandeau publicitaire mobile en ligne
AU2015246138B2 (en) Contextual advertising based on user configurable preferences
Popper Bringing It All Together in a Framework of Sponsored Search

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 07795067

Country of ref document: EP

Kind code of ref document: A2

NENP Non-entry into the national phase in:

Ref country code: DE

NENP Non-entry into the national phase in:

Ref country code: RU

122 Ep: pct application non-entry in european phase

Ref document number: 07795067

Country of ref document: EP

Kind code of ref document: A2