US20110208711A1 - Search annotation and personalization services - Google Patents

Abstract

Various features are disclosed for storing and providing access to event data reflective of user-generated events, including events associated with search query submission of users. One such feature enables users to annotate search results, to later recall and view these annotations, and to publish the annotations to other users. Another feature involves recording event data reflective of search result viewing events of users, and using this event data to personalize search results pages for particular users.

Description

PRIORITY CLAIM
• This application is a continuation of U.S. application Ser. No. 11/325,009, filed Jan. 4, 2006, which is a division of U.S. patent application Ser. No. 10/612,395, filed Jul. 2, 2003, the disclosure of which is hereby incorporated by reference.
BACKGROUND
• 1. Field
• The present disclosure relates to computer-based systems for capturing, persistently storing, and serving event data reflective of user-generated events, including search-related events.
• 2. Description of the Related Art
• Web site systems commonly include one or more mechanisms for capturing and storing information about the browsing activities or “clickstreams” of users. The captured clickstream data is commonly used to personalize web pages for recognized users. Typically, however, the captured clickstream data either provides only very limited information about each user's browsing history, or is captured in a format that is of only limited use for personalization.
• For example, some web sites maintain a real time record of each item selection, browse node selection, and search query submission performed by each user during browsing of an electronic catalog. Such browse histories are useful, for example, for generating personalized item recommendations, and for displaying navigation histories to assist users in returning to previously accessed content. However, these types of records typically lack the level of detail and structure desired for flexibly building new types of real-time personalization applications.
• Some systems also maintain web server access logs (“web logs”) that contain a chronological record of every HTTP request received by the web site, together with associated timestamp and user ID information. For web pages that are generated dynamically, the web query logs may also record the identities of items presented to users within such pages (commonly referred to as item “impressions”). While these logs typically contain more detailed browse history information, they are maintained in a format that is poorly suited for the real-time extraction and analysis of users' clickstream histories. Although web logs can be mined for information useful to various personalization functions, the task of mining a large web log can take many hours or days, potentially rendering the extracted data stale by the time it is available for use. Further, much of the detailed information contained in a web log is disregarded during the mining process, and is thus effectively lost for purposes of personalization.
SUMMARY
• An event history server system is disclosed that persistently stores event data descriptive of events that occur during browsing sessions of web site users. The event data is stored in association with the IDs of the corresponding users, and is made available in real time to web site applications that may use the event data to personalize web pages for specific users. In one embodiment, the event history server records event data descriptive of substantially every selection event (e.g., mouse click) of every user of a web site. The event history server may also record event data descriptive of other types of browsing events, such as impressions (i.e., items presented to users on dynamically generated web pages) and mouse-over events.
• The event data stored for each recorded event is preferably stored within a database as an event object. Each event object may, for example, include identifiers of the general event type (e.g., mouse click, impression, etc.) and type of display element involved (e.g., catalog item, browse node, search result item, etc.), an event value (e.g., the text of a selected URL), a timestamp indicating of the event's date and time of occurrence, and associated context information. A query interface of the event history server enables applications, such as personalization applications, to retrieve a particular user's event data by general event type, type of display element involved, time of occurrence, and possibly other criteria. The query interface also preferably supports queries with the semantics of “has user X accessed URL Y before?” and “when did user X access URL Y?” The query interface may more generally support queries with the semantics of “does an event of type T and value V exist within the event history of user X?”
• In one embodiment, users are provided an option to annotate particular search results, and to publish the annotations to other users. The annotations may be stored as events that can be retrieved via the query interface. The system may also provide functionality for users to organize and search their respective event histories.
• Neither this summary nor the following detailed description purports to define the invention. The invention is defined by the claims.
BRIEF DESCRIPTION OF THE DRAWINGS
• FIG. 1 illustrates a web site system that includes an event history server system according to one embodiment of the invention.
• FIG. 2 illustrates a set of software components that may be executed by the cache layer servers and storage layer servers ofFIG. 1.
• FIG. 3 illustrates an example web search results page in which search results are annotated to indicate the user's prior browsing history with respect to such items.
• FIG. 4 illustrates a process for identifying search result items that have previously been accessed by the user.
• FIG. 5 illustrates a process for determining whether the user previously submitted the same search query, and if so, whether any new search result items have been located in the current search.
• FIG. 6 illustrates the use of a Bloom filter to efficiently determine whether a user has accessed a particular object.
• FIG. 7 illustrates how the event history server may be used to collect event data reported by a browser component, such as a browser toolbar.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
• A preferred embodiment of a web site system that includes an event history server system will now be described with reference to the drawings. As will be recognized, many of the inventive features embodied within the disclosed system may be implemented or used without others. Numerous implementation details will be set forth in the following description in order to illustrate, but not limit, the invention.
I. Overview
• FIG. 1 illustrates aweb site system30 that includes an event history server system32 (“event history server”) according to one embodiment of the invention. Theevent history server32 provides a service for persistently recording, and providing real-time access to, event data indicative of events that occur during browsing sessions of users. The event data preferably describes the clickstreams of the users, and may also describe other types of events such as mouse-over and impression events. As described below, theevent history server32 also provides an API (application program interface) through which applications and services can flexibly retrieve, and request information about, the event data of specific users.
• As illustrated inFIG. 1, theweb site system30 includes one or moreweb server machines34 that receive and process requests fromuser computers36 and/or other types of devices (PDAs, mobile phones, etc.). Processes running on theweb server machines34 communicate withvarious web applications38, which may be implemented as web services. Theseapplications38 may run on theweb server machines34, but more typically run on separate server machines (not shown). The types of web applications provided within a given system will typically depend upon the nature and purpose of the system. For example, an Internet search engine site may include one ormore search applications38 for enabling users to conduct keyword searches of an index of web pages. An electronic commerce site may include web applications for performing such tasks as conducting searches of an electronic catalog, browsing an electronic catalog via a browse tree, generating personalized recommendations, placing orders, and managing personal account data. For purposes of this description, it may be assumed that at least some of theweb applications38 supply personalized web page content based on user-specific event data stored by theevent history server32.
• As depicted inFIG. 1, eachweb server machine34 preferably runs anevent reporting component35 that sends updates to theevent history server32 in response to actions performed by online users. (A given user need not be an individual, but may, for example, be a group of individuals that share auser computer36 or a user account). Each such update contains or otherwise specifies event data descriptive of a particular event that has occurred during a browsing session of a user, such as a mouse click, a search query submission, or an impression. Theevent reporting component35 may include a set of APIs (application program interfaces) for making calls to theevent history server32, and may include configuration parameters specifying the types of actions for which events/updates are to be generated. APIs are also provided for allowingapplications38 to submit queries to theevent history server32. As described in section V below, browsing events may additionally or alternatively be reported to theevent history server32 by a browser-based event reporting component (FIG. 7) that runs on theuser computers36.
• The set of data stored by theevent history server32 for a particular event is preferably stored as an “event object.” In one embodiment, for example, theevent history server32 persistently stores event objects describing substantially every selection action or “mouse click” of every recognized user of theweb site system30. This information may include, for example, the URL accessed, the time/date of the access, and associated context information. The event history server may also record impression events reflective of specific items presented to users within dynamically generated web pages. For example, when a user views a dynamically-generated web page that includes a personalized list of items selected from a database, the event history server may store an event object for each such item, or may store a single event object that contains the list of items. Additional examples of the types of events that may be recorded, and of the event data that may be recorded for such events, are provided below. The event objects are preferably stored and indexed within theevent history server32 to permit retrieval based on time-of-occurrence, general event type (e.g., mouse click versus impression), type of display element involved (e.g., catalog item, browse node, or Web search result URL), user ID, and other object properties.
• The event data captured by theevent history server32 reflects actions performed by users during browsing of a particular web site or set of web sites hosted by theweb site system30. This captured data may, in some embodiments, also reflect actions performed by users during browsing of external, independent web sites. For example, users may be permitted or required to download to their computers36 a browser plug-in, such as a browser toolbar, that reports all URL accesses (and possibly other types of events) to the event history server32 (seeFIG. 7, discussed below).
• Some or all of theweb applications38 preferably act as clients of theevent history server32. As depicted inFIG. 1, the web applications interact with theevent history server32 primarily by sending event queries to theevent history server32 in order to retrieve or obtain information about specific events. Theses queries preferably include requests for fixed-size blocks of event data, such as “recall last N events of type T for user Y,” or “recall next N events of type T for user Y.” For example, aweb application38 may request the event objects for the last fifty search queries submitted by a particular user, or may request the event objects for the last fifty browse nodes viewed by a particular user. The retrieved event objects may be used by theweb applications38 for any suitable purpose, such as generating personalized item recommendations or displaying browse history data to users to facilitate navigation.
• Theevent history server32 also preferably supports queries of the semantic form “does an event of type T and value V exist within the event history of user Y?” and “when did an event of type T and value V occur within the event history of user Y?” For example, a web application can query theevent history server32 to find out whether a particular user has accessed a particular URL before, and if so, when. As described below, one application of this feature involves generating a personalized web search results page (seeFIG. 3) that highlights the search results/URLs that have already been accessed by the particular user, and indicates when the viewed URLs were last accessed. Additional examples of the types of queries that may be supported by theevent history server32, and of applications which make use of the captured event data, are provided below.
• Although the event queries are depicted inFIG. 1 as emanating from theweb applications38, theevent history server32 may also respond to event queries (and possibly updates) from other types of components. For example, a data mining application may retrieve and analyze event data of users for purposes of enhancing browsing of the web site; and a security application or administrator may retrieve clickstream data to analyze a security breach. Application components and services that send updates and/or queries to theevent history server32 are referred to herein as “clients,” and are depicted inFIG. 1 as being part of aclient layer39.
• As illustrated inFIG. 1, theevent history server32 consists of two layers: acache layer40 and apersistent storage layer44. Thecache layer40 is made up of a set of physicalcache layer servers42, each of which runs service-level software70 (FIG. 2) for responding to updates and queries from clients. Eachcache layer server42 includes acache43 that temporarily stores event data in the form of event objects. Thecache43 may be implemented using non-volatile disk storage, random access memory, or a combination thereof. As described below, thecache43 may also store user-specific Bloom filters for responding to certain types of queries. Communications between theclient layer39, thecache layer40 and the persistent storage layer preferably occur over a local area network.
• Thecache layer servers42 are preferably partitioned by browsing session ID, meaning that eachsuch server42 only stores event data (event objects) associated with its respective range or group of session IDs. Thus, for example, when a user starts a new browsing session, that session is assigned to a particularcache layer server42 which thereafter services all updates and event queries corresponding to that browsing session. In one implementation, a total of nine dual-CPUcache layer servers42 are provided within thecache layer40, one of which is used as a hot spare. Thecaches43 are preferably implemented as stateless, write-through caches, facilitating the addition, removal, backup, and rebooting of themachines42.
• As described below, thecache layer40 acts essentially as an intermediary between theclient layer39 and thepersistent storage layer44, allowing the system to operate at higher throughput levels. Although the use of acache layer40 is preferred, theevent history server32 may alternatively be implemented without a cache layer. Further, where thecache layer40 is provided, it may be bypassed where caching would not be beneficial, such that some requests are serviced directly by thepersistent storage layer44.
• With further reference toFIG. 1, thepersistent storage layer44 is made up of a set of physicalstorage layer servers46. In the illustrated embodiment, twostorage layer servers46 are provided, although a greater or lesser number may be used. Eachstorage layer server46 is responsible of maintaining a complete copy of the event data persistently stored by thesystem32. Thus, thestorage layer servers46 are mirrors of each other when two or more are provided. Because thestorage layer servers46 are mirrored, a singlestorage layer server46 can go down or be taken off line without affecting the overall functionality of the event history service.
• Eachstorage layer server46 may, for example, include several terabytes of disk drive storage. If approximately 100M events are recorded per day, and an average of thirty bytes of data are stored for each recorded event, eachstorage layer server46 will store approximately 3 GB (gigabytes) of event data per day, or about one terabyte per year. Under such a usage scenario, eachstorage layer server46 is capable of storing and serving several years worth of event data. When the storage capacity of astorage layer server46 is reached, new disk drives may be added to the storage layer to increase its capacity, and/or old event data may be purged or archived.
• As described below and depicted inFIG. 2, eachstorage layer server46runs database software62 for storing and managing the event data, and runs service-level code60 for processing updates and queries received from thecache layer40. Thestorage layer servers46 may be implemented using low cost commodity hardware, and may use Berkeley databases, or another type of relatively low cost database, to store the event data.
• Theevent history server32 responds to updates generally as follows. When an update is sent to thecache layer40 by aweb server machine34, thecache layer server42 assigned to the corresponding session ID updates itsrespective cache43 to include the event object specified by the update. If the event is for a recognized user, thecache layer server42 also sends the update to each of the storage layer servers46 (preferably using a publish-subscribe protocol), and eachsuch server46 stores the associated event object within itspersistent storage64 in association with the corresponding user ID. Eachcache layer server42 preferably aggregates multiple events/updates for sending to thestorage layer servers46, so that theseservers46 receive updates from the cache layer in batches. The event objects are stored in thecache43 in association with the corresponding session ID and/or user ID.
• If the event data specified within an update is for an unrecognized user, thecache layer server42 stores the event object in its cache43 (in association with the corresponding session ID), but does not send the update to thepersistent storage layer44. The user may be unrecognized if, for example, the user (1) is new to the web site, (2) is accessing the web site from aparticular computer36 for the first time and has not logged in, or (3) is accessing the web site from acomputer36 that is configured to block cookies and has not logged in. Throughout a session, the cache layer will thus collect event data for an unrecognized user.
• If an unrecognized user logs in, creates an account, or otherwise becomes recognized during the browsing session, the associatedcache layer server42 associates collected event data for that user with the user's ID. In addition, thecache layer server42 sends a series of updates to thestorage layer servers44 to persistently store this collected event data in association with the user ID. The system thus allows a user's events to be persistently captured even though the user may not be recognized at the time such events occur. Further, during the period in which the user is unrecognized, clients of theevent history server32 can retrieve and request information about the cached event data of the unrecognized user based on the user's session ID.
• Theevent history server32 responds to event queries from clients generally as follows. When acache layer server46 receives an event query from a client, it initially checks itsrespective cache43 to determine whether the relevant event data resides therein, and responds to the query if the data is present. If the relevant event data does not reside in thecache43, the query is passed to one of thestorage layer servers46. A load balancing algorithm may be used to select between thestorage layer servers46 for this purpose. The selectedstorage layer server46 responds to the query by generating a response (which may include requested event objects), and returning this response to thecache layer server42 from which the query was received. Thecache layer server42 then passes this response to the requesting client.
• If the response includes event data retrieved from persistent storage, thecache layer server42 stores this event data in itsrespective cache43 by default. In the preferred embodiment, an event query may indicate that the retrieved event data should not be cached; this feature may be used, for example, to inhibit caching when a large quantity of event data is read from persistent storage for purposes of off-line data mining.
• As indicated by the foregoing, the illustrated embodiment of theevent history server32 captures data descriptive of browsing events as such events occur, and makes such event data available topersonalization applications38 in real time (i.e., substantially immediately). In addition, the event data is made available in a form that allowsapplications38 to limit their queries to the specific types and items of event data needed to perform specific personalization tasks. Further, unlike systems that rely on the results of an off-line data mining analysis, theapplications38 have access to the “raw” event data itself, as opposed to merely a summary of such data. Users of the event history server may, in some embodiments, be given the option to control whether their respective browsing histories are to be recorded by the event history server (e.g., an opt-in or opt-out option may be provided). Additional features and benefits of the disclosed architecture are discussed below.
II. Event Object Content and Retrieval
• As indicated above, the set of data stored for a given event is stored by theevent history server32 as an event object. In one embodiment, each event object includes the following components: Subject, Value, Tag, and Time. Each of these components is described below. In one embodiment, these components are used to capture data regarding three general types of events: mouse clicks, impressions, and mouse-over events.
• The Subject of the event object is a code that indicates, for mouse click, mouse over, and impression events, the type of display element involved (e.g., an item, a browse node, an external URL, or link for submitting a search query). One or more subject codes may also be defined for describing search query submissions from users. The Subject of an event, together with the event's Tag (described below), fully specify the type of the event (e.g., mouse click of browse node, impression of catalog item, etc.).
• Table 1 below provides examples of some of the event subjects that may be supported, and indicates the data stored in the Value field for each such event subject. These examples assume that theweb site system30 hosts an electronic catalog that may be browsed and searched by users to locate items (products, new articles, etc.), and also assumes that theweb site system30 implements a search engine for locating external web sites and pages. As will be recognized, the types of events recorded within a particularweb site system30 will depend largely on the nature and purpose of that system, and may vary significantly from those listed in Table 1.
• Further, the event types may be varied or extended in order to support additional application features. For example, in one embodiment which is not represented in Table 1, an “annotation” event type is defined for purposes of storing annotations entered by users. One application for annotation events involves allowing users to annotate their respective search results, and to later recall and review such annotations. Users may also be given the ability to publish their annotations to other users. Unlike mouse clicks, impressions, and mouse over events, annotation events represent explicit requests by users to store event data for later retrieval.

• TABLE 1
  Event
  Subject	Description	Value
  Catalog	User selection, impression, or mouse-over of a	Item ID
  item	catalog item
  Browse	User selection, impression, or mouse-over of a	Browse node ID
  node	browse node (item category)
  Internal	Search query submission for conducting an	Text of search term(s)
  Search	internal search, such as a search for items in an	entered by user
  query	electronic catalog
  Web search	Search query submission for conducting a	Text of search term(s)
  query	general Web/Internet search that is not limited	entered by user
  	in scope to any particular set of Web sites
  Web search	URL selected by user from web search results	Two separate strings are
  URL	page. This information may be captured, for	stored, one which contains
  	example, by initially directing the user's	the text of the URL, and the
  	browser to an internal URL used for tracking	other of which contains the
  	purposes, and then redirecting the browser to	page's display title. Strings
  	the external URL/web site selected by the user.	that exceed a particular
  		length are truncated.
  Feature	User access to a particular web site feature,	Feature ID
  	such as an item recommendations service
  Other URL	URL access event that does not fall within one	Text of URL and associated
  	of the above categories	display title

• The Tag may be implemented as a set of flags indicating some or all of the following: (1) the general type of the event (e.g., mouse click, impression, or mouse-over), (2) whether the Value portion of the event has been truncated (e.g., because of the excessive length of a particular text string), (3) whether the event is “undisplayable,” meaning that it cannot be viewed by the user, and (4) whether the event is transient versus persistent.
• A Tag's “undisplayable” flag may be used, for example, to allow users to effectively remove events from their viewable event histories. For example, theweb site system30 may provide anapplication38 and associated user interface through which users can view and search their respective event histories, and “delete” selected events from such histories. When a user deletes a particular event (such as particular search query submission or browse node access), the corresponding event object is marked by theevent history server32 as “undisplayable” to prevent the user from viewing the associated event, but remains accessible to clients of theevent history server32.
• The “transient/persistent” flag may be used to mark those events that can be permanently deleted from persistent storage at a certain point of time. This feature may be used to purge event objects that are of little or no value after a certain time period, so that the associated persistent storage is made available for storing other data.
• The Time component is a numerical value indicating the time of occurrence of the event, and may be expressed, for example, in seconds since 1970. When the persistent/transient flag is set to “transient,” an additional value may be included specifying the time of expiration of the event object. Expired event objects may be deleted from persistent storage periodically by a background task, or using any other appropriate method.
• The query set implemented by theevent history server32 preferably allows clients to retrieve the event objects for a given user or session based on event Subject, Value, Tag, and Time. For example, a client can request the event objects for all impressions (or all mouse click events) of a particular type of display element, or for all impressions (or all mouse click events) of a particular display element type and value. In addition, the query set preferably allows clients to specify an event time range (e.g., “last 10 days,” or “since Feb. 10, of 2003”).
• As mentioned above, the query set also supports queries of the following form: “does an event of type T and value V exist within history of user X?” The type of the event may be specified in the query in terms of the general event type (e.g., mouse click or impression), the type of display element involved, or both. For example, a query of the form “does event of type=Web search query and Value=comet Halley exist in history of user X?” would reveal whether user X has conducted a general Web search using the query “comet Halley.” Further, the query set supports queries of the type “when did event of type T and value V occur in the history of user X?”
III. Software Architecture
• FIG. 2 illustrates the primary software components that run on the cache layer andstorage layer servers42,46 in one embodiment. As illustrated, eachstorage layer server46runs service code60 for processing queries and updates from thecache layer40. The service code communicates withdatabase software62, such as software used to implement Berkeley databases. Thedatabase software62 manages one or more databases of event data indisk drive storage64. In one embodiment, eachstorage layer server46 runs multiple pub-sub type service processes, each of which corresponds to a respective set or range of user IDs and exclusively accesses its own Berkeley database; this architecture greatly simplifies the management of the databases. Eachstorage layer server46 additionally runssynchronization service code68 that is responsible for maintaining the storage layer servers in synchronization. The synchronization service is used, for example, to synchronize the persistent databases after one storage layer server is temporarily taken off line.
• Eachcache layer server42 runs cachelayer service code70 that accesses itsrespective cache43 of event data. Event data is preferably stored in thecache43 both by user ID (if the user is recognized) and session ID. As illustrated inFIG. 2, the cachelayer service code70 includescomponents72,74 for processing updates and queries received from clients.
• As further depicted inFIG. 2, both types ofservers42,46 also preferably include software components for generating and processing Bloom filters. As described below, each Bloom filter is a condensed representation of some portion of a user's event data, and may be used to evaluate whether the user has performed a particular action without having to retrieve the associated event objects. For example, in one embodiment, Bloom filters are used to reduce the processing and data retrieval needed to determine whether a given user has accessed a given URL.
IV. Example Search Personalization Applications
• As indicated above, one application of theevent history server32 involves generating a personalized search results page identifying any search result items that were previously accessed by the particular user. The search results page may further indicate the time each such item was accessed. This feature may be applied both to internal catalog searches (in which case the search results page may, for example, indicate those catalog items for which the user has viewed an item detail page), and to web searches (in which case the search results page may indicate which of the external web pages has been viewed).
• FIG. 3 illustrates an example web search results page according to one embodiment of this feature. In this example, the web search query “lord of the rings” has produced a search results listing in which each search result item (three shown) is an external web page or site having a corresponding URL. In one embodiment, the links contained within the search result listings are internal links (i.e., they point back to the web server system30) that are used to immediately redirect the user's browser to the external page of interest; this allows theevent history server32 to record event objects describing the search results (external URLs) selected by the user for viewing. In another embodiment, the user's selections of external URLs are logged by firing a JavaScript component that detects and reports the user's selections.
• Thefirst item100 in the list ofFIG. 3 includes the annotation “viewed,” indicating that the user previously accessed this particular URL. The user can thus readily identify those web pages that have already been viewed. Because the user's event history is stored on the server side (rather than merely on the user's computer36), the “viewed” status may be properly indicated even if the user only viewed the particular web page from adifferent computer36. By hovering the mouse cursor over thisitem100, the user can also view the date he/she last accessed this URL, as shown by the mouse-over text “previously viewed on February 14.” The date-of-access text may alternatively be displayed in-line with thesearch results text100. Various other types of information can be incorporated into the annotations based on the user's event history, such as the number of times the user has accessed the particular URL, and the text of a prior search query that uncovered this URL (if different from the current search query).
• FIG. 4 illustrates the general process by which asearch application38 may interact with theevent history server32 to generate search results pages of the type shown inFIG. 3. It is assumed in this example that the user is recognized by theweb site system30. As depicted byblock110, the search application initially receives a search query, and executes the search to generate a set of search results. If the search is a general web search, each search result item will be in the form of a URL of a web page that is responsive to the search, and may be displayed together with textual excerpts extracted from such web pages (as inFIG. 3). As is known in the art, the execution of such searches typically involves comparing the search query to an index generated by a web crawler program. If the search is directed to an electronic catalog hosted by theweb site system30, each search result item may be in the form of an item description that may be clicked on to access a corresponding item detail page. In some embodiments, the search results returned in a given search may include both external web pages and items selected from an internally-hosted electronic catalog.
• As depicted byblock112, the search application then sends a separate query to theevent history server32 for each search result item—or at least those that are to be displayed on the current search result page—to determine whether the user previously accessed that item. For general web searches, each such query may be in the form of “has user X selected URL=<URL value> before?” For catalog searches, each query may be in the form of “has user X selected catalog item=<item ID> before?” As described below, theevent history server32 preferably uses Bloom filters to efficiently process these types of queries, although the use of Bloom filters may alternatively be omitted.
• As depicted byblock114, for each search result item found to have been previously accessed by the user (if any), an additional query is preferably sent to theevent history server32 to request the date (time) of the last user's last access to that item. Finally, as depicted inblock116, a search results page is generated with embedded viewed-item annotations of the type shown inFIG. 3. The search application may also take the viewed/not-viewed status of each search result item into consideration in ranking/ordering the search result items for display.
• As will be apparent, the process depicted byFIG. 4 can be varied such that theevent history server32 is not actually accessed in response to the search queries. For example, to reduce the load on theevent history server32, the event data indicative of specific search result URLs selected by specific users can be periodically retrieved from theevent history server32 and stored on a separate “search personalization” server. This separate server can then be assigned to the task of responding to requests of the type “has user X accessed external URL Y?” The same is true for the process ofFIG. 5, described below.
• The search application's user interface may also provide an option for the user to restrict the scope of the search to items previously viewed, items not previously viewed, or items viewed within a particular time period (e.g., the last seven days). This feature may be implemented using the same process flow as inFIG. 4, except that the query results returned by theevent history server32 will be used to determine which search result items are to be displayed.
• As will be apparent, the foregoing search results personalization features, as well as those described below with reference toFIG. 5, may also be applied to searches for other types of items. For example, in the context of an online auction site, the results of an auction search may be annotated to indicate which of the located auctions have been viewed by the user, when each auction was viewed, and possibly whether the user has submitted a bid on each such auction. Similarly, the results of a search of bulletin board postings or blog (web log) postings may be annotated to indicate if/when specific postings were viewed by the user.
• Referring again toFIG. 3, the secondsearch result item120 illustrates another feature that may be implemented using theevent history server32. This feature involves determining whether the user previously submitted the same search query, and if so, whether any new search result items (i.e., items not present in the prior search results set) have been found. In this particular example, the search results page reveals that the user previously conducted the same search on February 14, and that the item corresponding to the URL “lordotrings.com” is a new item that did not come up in the prior search.
• FIG. 5 illustrates the general process by which asearch application38 may interact with theevent history server32 to implement this “new-item annotations” feature. This process may be combined with that ofFIG. 4 to generate annotated search results pages of the type shown inFIG. 3. As illustrated inFIG. 5, the search application initially receives and executes the search query (block130), and queries theevent history server32 to determine whether the user has previously submitted the same query (block132). If the user has not submitted this query, the search results page is generated (block140) without adding any “new item” annotations.
• If the same search query was previously submitted, theevent history server32 is again queried for the time of the last submission (block134). In addition, as depicted inblock136, theevent history server32 is queried to determine which of the current search result items, if any, have not been displayed to the user (i.e., have not been the subject of an impression event); this query may optionally be limited in scope to impressions occurring at the time of or shortly after the prior search. As depicted byblock138, if any new search result items exist in the current search result set, they are annotated as shown inFIG. 3 (item120). The relevance rankings of the new items may also be augmented to increase the likelihood that the new items will be displayed.
• One variation of the method shown inFIG. 5 is to allow the user to explicitly limit the (repeat) search to items that did not come up in a prior execution of the same search query. This option may, for example, be provided as a check box on the web site's search page.
IV. Use of Bloom Filters to Determine Whether User Previously Viewed or Accessed a Given Element
• As mentioned above, Bloom filters may be used by theevent history server32 to reduce the need for persistent data retrieval when responding to a query of the form “does event of type T and value V exist in history of user X?” For example, Bloom filters may be used to determine whether a particular user has viewed or selected a particular URL.
• By way of background, a Bloom filter is a bit sequence or array generated according to a set of hash functions. Bloom filters are used to quickly test whether a particular item is a member of a large set of items. One common application for Bloom filters is to test whether a given object (as identified by the object's URL) is currently stored in a cache of web pages. Specifically, when an object is added to the cache, the hash functions are applied to the object's URL to determine which of the bits in the Bloom filter are to be turned ON. When an object is requested, these hash functions are again applied to the requested object's URL, and a test is then performed to determine whether all of the corresponding bits in the Bloom filter are turned ON. If one or more of the bits are not ON, the object is not stored in the cache. If, on the other hand, all of the bits are ON, there is a very high likelihood that the requested object is in the cache. Thus, “false positives” or “false hits” are possible, but “false negatives” or “false misses” generally are not.
• According to one aspect of the invention, one or more Bloom filters are generated for a given user to describe some aspect or segment of that user's event history. The Bloom filters are preferably generated and stored by the persistent storage layer44 (FIG. 1), and are passed to thecache layer40 for purposes of responding to queries. Different Bloom filters may be generated for a given user for responding to different types of queries. For example, different types of Bloom filters may be generated for different types of activity (e.g., URL accesses, URL impressions, URL accesses plus impressions, catalog item accesses, mouse over events, etc.). In addition, different Bloom filters may be generated for different time periods.
• FIG. 6 illustrates the general process by which a Bloom filter may be used to test for URL accesses by a recognized user, User X. Although this diagram is specific to URL accesses, the illustrated process flow also applies to other types of event activity. In step1 ofFIG. 6, astorage layer server46 generates one or more Bloom filters for User X. This step may, for example, be performed in response to receiving an associated query or update for User X. To generate a Bloom filter specifically for URL accesses, the storage layer service code (FIG. 2) initially retrieves URL access events for User X, applies a set of hash functions to each accessed URL, and sets the associated Bloom filter bits. As discussed below, the size of the Bloom filter may be selected to enable storage of a desired number of events while maintaining a false-positive rate below a desired threshold. Bloom filters generated by onestorage layer server46 may be passed to the other storage layer server(s) by the synchronization service. The bloom filters may additionally or alternatively be generated by thecache layer servers42.
• In step2 ofFIG. 6, a cache layer server receives a query of the form “has User X accessed URL Y?” In the context of the search results annotation features described above, many such queries may be received for a given user for purposes of generating a single search results page. In step3, thecache layer server42 retrieves the relevant Bloom filter from one of thestorage layer servers46 if the Bloom filter is not already stored in thecache43.
• Instep4, thecache layer server42 tests the relevant Bloom filter to see if the corresponding bits for the URL are ON. If one or more of the bits are OFF (meaning that no accesses to URL Y exist in the relevant event history of User X), thecache layer server42 returns an answer of NO (step5), without passing the query to the persistent storage layer. If, on the other hand, all of the bits are ON (meaning that the URL access very likely exists within User X's event history), the query is passed to one of the storage layer servers46 (step6) to check User X's actual event data for the URL access. In step7, thestorage layer server46 returns a response to the query via thecache layer server42. As an alternative to querying the storage layer server (step6), the cache layer server can be designed to simply return a YES response when the Bloom filter test is positive, although this approach may cause in accurate results to be presented to users on rare occasions.
• As updates reflective of URL accesses are thereafter received for User X, the cache layer server42 (but preferably not the storage layer server46) updates its copy of the associated Bloom filter to reflect these updates. For example, if User X selects a search result item (URL) from a web search results page of the type shown inFIG. 3, the associated Bloom filter will be updated within thecache43 to set the bits associated with the selected URL. Thus, the Bloom filter used by the cache layer to respond to queries reflects the most recent browsing activities of the relevant user. At some point, the Bloom filter may be purged from thecache43 due to inactivity, at which time the updated Bloom filter is written to the persistent storage layer (in place of the original version) for further use.
• At some point, a given Bloom filter may reach its capacity, meaning that it cannot store additional events without exceeding a desired average false positive rate. At this point, the Bloom filter may be replaced with a larger Bloom filter (e.g., 8 Kilobytes rather than 4 Kilobytes) in order to provide greater event capacity.
• The foregoing description focuses on the generation and use of Bloom filters for recognized users. Bloom filters may also be generated for unrecognized users by thecache layer servers42. For example, at the outset of a browsing session, the assignedcache layer server42 may generate a Bloom filter for a user, and may thereafter update the Bloom filter with new events for that user. This “session-specific” Bloom filter may be used to respond to queries in the same way as described above.
V. Browser-Based Reporting of Event Data
• FIG. 7 illustrates an embodiment in which theevent history server32 records event data reported by a browser-basedevent reporting component160. Theevent reporting component160 runs as a component of aweb browser162 on some or all of thecomputers36 of users of theevent history server32. As illustrated, theevent reporting component160 is preferably implemented as part of abrowser toolbar component164. Thetoolbar component164 may, for example, be provided as an optional browser plug-in that can be downloaded and installed by users. Theevent reporting component160 may alternatively be implemented as a standalone browser plug-in, as part of a plug-in other than a toolbar, or as part of the native code of theweb browser162.
• The browser-basedevent reporting component160 preferably reports event data for all web sites and pages accessed by the user. For example, theevent reporting component160 may report every mouse click or other selection event on every web page accessed by the user. Theevent reporting component160 may also report other types of browsing events, such as mouse-over events, impressions, selections of the “back” button on theweb browser162, etc.
• The browser-basedevent reporting component160 may take the place of the server-basedevent reporting component35 ofFIG. 1. Alternatively, both types ofevent reporting components35,160 may be used within a given system. For example, for users that do not have thetoolbar164 installed, theevent history server32 may only capture data reported by the server-basedevent reporting component35, in which case the event data may only reflect events that occur during browsing of one or more specific web sites. For users that have thetoolbar164 installed on theirrespective computers36, the captured event data may also extend to other web sites.
• The event data collected from the browser-basedevent reporting component160 may be used to provide a variety of different personalization services to users. For example, a service may be provided for allowing users to view a listing of all web sites they have respectively accessed that satisfy some user-specified criteria. Using this service, users may, for example, view listings of all payment transactions they have respectively made on the web, or view a history of all travel-related web sites they have accessed. The data fields included within the event objects may be supplemented as needed to implement such a service.
• As depicted inFIG. 7, personalized toolbar content may be generated by a server-sidetoolbar personalization application170. Thisapplication170 may use event data retrieved from theevent history server32 to generate context-sensitive toolbar messages for display to users. For example, when a user accesses a particular web site, thetoolbar personalization application170 may retrieve and display a history of transactions performed by the user on that web site. Further, the toolbar may be personalized with real time information about all of the web sites or pages visited during the current browsing session.
VI. Other Personalization Applications
• As will be recognized, numerous other types of personalization applications and features are made possible by theevent history server32. As mentioned above, one such application involves allowing users to view, organize, and possibly annotate their respective event histories. This may be accomplished in part by providing a user interface, such as a set of web pages, through which users can create event history folders, and select events to add to such folders. An event search engine may also be provided through which users can search their respective event histories by event type, event value, event time-of-occurrence, and various other criteria. As mentioned above, users may also be permitted to “delete” specific events from their respective event histories.
• Although this invention has been described in terms of certain preferred embodiments and applications, other embodiments and applications that are apparent to those of ordinary skill in the art, including embodiments which do not provide all of the features and advantages set forth herein, are also within the scope of this invention. Accordingly, the scope of the present invention is intended to be defined only by reference to the appended claims

Claims (19)

1. A system that provides functionality for conducting searches, the system comprising:

a search engine system that is responsive to search queries received over a network from user computing devices of users by generating and returning search results pages listing corresponding search results, said search results including uniform resource locators of responsive resources, said search engine system providing functionality, including a user interface, that enables users to annotate particular search results, and to subsequently recall such annotations for viewing; and

a server that persistently stores event data in association with particular users of the search engine system, said event data comprising search result annotations created by particular users, said server including a query interface that enables the search engine system to selectively retrieve the persistently stored event data associated with particular users.

2. The system ofclaim 1, wherein the search engine system additionally provides functionality for the users to publish their respective search result annotations to other users.

3. The system ofclaim 1, wherein the user interface includes functionality for users to view and organize their respective event histories.

4. The system ofclaim 1, wherein the server is configured to store a search result annotation submitted by a user as an event object that is separately retrievable via the query interface.

5. The system ofclaim 4, wherein the event object includes an event type identifier and an event timestamp.

6. The system ofclaim 1, wherein the server is additionally configured to store event data descriptive of search result viewing events, and the search engine system is configured to personalize search results page for a user with annotations indicative of when the user previously viewed particular search results, said annotations based on the event data descriptive of search result viewing events.

7. The system ofclaim 1, wherein the server stores event data for each of a plurality of defined event types, including an annotation event type.

8. The system ofclaim 7, wherein the query interface provides functionality for selectively retrieving the event data based on at least user ID, event type, and event time-of-occurrence.

9. The system ofclaim 1, wherein the server comprises at least one persistent storage layer machine that persistently stores the event data, and comprises a plurality of cache layer machines that cache event data of particular users during browsing sessions of such users, wherein the search engine system is configured to send queries to the cache layer machines to request event data.

10. A method, comprising:

receiving a search query submission from a user computing device associated with a user;

executing the search query to identify a plurality of search results that are responsive to the search query;

outputting a representation of the search results to the user computing device for presentation to the user;

providing an option for the user to annotate the search results, and receiving a resulting search result annotation submitted by the user;

persistently storing the search result annotation in association with the user;

subsequently retrieving the persistently stored search result annotation by execution of a query; and

outputting the search result annotation for display to the user;

said method performed in its entirety by a computer-implemented system that comprises a plurality of server machines.

11. The method ofclaim 10, further comprising, by said computer-implemented system, providing an option for the user to publish the search result annotation to other users.

12. The method ofclaim 10, further comprising, by said computer-implemented system, publishing the search query annotation to other users.

13. The method ofclaim 10, wherein persistently storing the search result annotation comprises persistently storing, on a server, an event object that includes at least the search result annotation and an event timestamp.

14. The method ofclaim 13, wherein retrieving the search result annotation comprises receiving said query from an application, and, in response to the query, returning the event object to the application.

15. The method ofclaim 10, further comprising recording, in association with the user, event data that identifies specific search result URLs selected by the user for viewing, and using said event data to personalize search results pages for the user.

16. The method ofclaim 10, wherein the computer-implemented system is a web site system that hosts a web site that provides functionality for conducting Internet searches.

17. A system, comprising

a computer system comprising a plurality of servers, said computer system configured to perform a process that comprises:

receiving a search query submission from a user computing device associated with a user;

executing the search query to identify a plurality of search results that are responsive to the search query;

outputting a representation of the search results to the user computing device for presentation to the user;

providing an option for the user to annotate the search results, and receiving a resulting search result annotation submitted by the user;

persistently storing the search result annotation in association with the user;

subsequently retrieving the persistently stored search result annotation; and

outputting the search result annotation for display to the user.

18. The system ofclaim 17, wherein the computer system is additionally configured to provide an option for the user to publish the search result annotation to other users.

19. The system ofclaim 17, wherein the computer system is additionally configured to publish the search query annotation to other users.

Images