System to generate related search queries

Abstract

Systems and methods described herein may operate to receive, at a search service, a first query from a user device corresponding to a current user of the search service, analyze existing search queries performed by users of the search service, and selectively provide, to the user device, a second query of the existing search queries as at least one of: a refinement of the first query based on determining that the second query includes all keywords in the first query, or an alternative to the first query based on determining that the second query lacks at least one keyword in the first query.

Description

BACKGROUND

Networked commerce systems, such as Internet-based auction systems, depend upon user specified searches to locate items of commerce. While some users are highly skilled at locating desired items, a large segment of users lack knowledge helpful in navigating such systems. As a result, inefficient use of the system and a lack of user traffic to some commerce items can be experienced.

To increase sales, some commerce systems have provided recommendations to users based upon purchasing performance of prior users. For example, when viewing a product the system may indicate that users who purchased the product also purchased a second identified product. This system does not assist the user in navigating but merely provides suggestions to increase sales. What is needed is a system to assist users in navigating a network based commerce system.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a page view of a user interface;

FIG. 2 is a flow chart illustrating a method of one embodiment;

FIGS. 3-5 are additional page views of user interfaces:

FIG. 6 is a flow chart illustrating a method of one embodiment;

FIGS. 7, 8A, 8B, 9A, 9B, and 10-13 are additional page views of user interfaces;

FIG. 14 is block diagram illustrating a network-based computer system;

FIG. 15 is a database diagram illustrating an example database;

FIG. 16 illustrates example logic for generating recommendation queries based on past user interaction; and

FIG. 17 shows a diagrammatic representation of a machine in the example form of a computer system.

DETAILED DESCRIPTION

Methods and systems to search or access a data source are described. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the invention. It will be evident, however, to one skilled in the art that the invention may be practiced without these specific details.

In various embodiments, a first query may be received, at a search service, from a user device corresponding to a current user of the search service. Existing search queries performed by users of the search query may be analyzed. As a result of the analyzing, a second query of the existing search queries may be selectively provided, to the user device, as at least one of: a refinement of the first query based on determining that the second query includes all keywords in the first query, or an alternative to the first query based on determining that the second query lacks at least one keyword in the first query. Other embodiments are possible. More detailed explanations of various embodiments are provided below with respect to FIGS. 1-17.

In various embodiments, new recommendations are provided to a search interface to assist users in navigating toward new searches that are likely to generate results aligned with the user's searching intentions. In one embodiment, an algorithm analyzes previous search sessions to determine a “next best place” for the user to go. In an example of a commerce system, new information placement may be added to the top of search and listings pages to show links to new searches that can be run by the user. On a periodic basis, the search recommendations may be updated, for example based on the most current changes in user behavior. For example see the page view 100 of FIG. 1. Where a search term “Ferrari” in “all categories” provides related search suggestions of “Lamborghini”, “Porche”, “Bentley”, “Aston Martin” and “Maserati.”

An embodiment seeks to improve the searching experience by directing users to searches that have proven successful in the past. Users who execute one of the search recommendations may be more successful searching for, and locating, data items of interest. The recommended additional search options may prove useful to non-expert users of a data resource, specifically those users who need more guidance on which search terms they should use to successfully find data items on a data resource.

User provided searches that are determined to be of poor quality can include offensive terms, terms that are not relevant, and terms that drive the user more in a particular direction than another (e.g., toward a particular type of brand instead of a competing brand, in a commerce environment). As such, recommended alternate searches can be provided.

In various embodiments, a search and recommendation engine seeks to provide recommendations for a further search (e.g., query terms) based on the observed (or recorded) behavior of users of the search and recommendation engine with respect to an information resource. For example, the search and recommendation engine, after having received a first search query including the term “Toyota”, may recommend a further search query including the term “Honda”. The recommendation of the further search query is based upon, in one embodiment, having observed (or recorded) that a certain threshold number of users provided search queries including the term “Honda”, subsequent to providing a search query including the term “Toyota”.

Further, in various embodiments, whether or not the search and recommendation engine recommends the further search query (and also how the search engine ranks the further search query) may be based on post-search user or system behavior (or activities) with respect to the information resource. For example, where the search engine recorded that the further search query (e.g., including the term “Honda”) previously delivered less than a predetermined number of search results, a recommendation ranking with respect to the further search query (relative to other candidate search queries) may be reduced.

Where the search and recommendation engine recorded the presence (or absence) of certain post-search user behavior, this information may be utilized to rank the further search query and/or utilized to determine whether to provide the further search query as a recommended search query. For example, in the context of a commerce system, where the recorded past user behavior indicated that, subsequent to a particular candidate search recommendation, a threshold number of users purchased (or entered into bidding or some other transaction activity) with respect to one or more products/services identified by the particular candidate search recommendation, such a particular candidate search recommendation may receive a higher recommendation ranking than a further candidate search recommendation that did not result in transaction activity. In one example, the search and recommendation engine may track the number of users that performed search A, then performed search B. and then performed certain post-search activity with respect to a result set returned by search B. Accordingly, it will be the appreciated that a recommended search query (e.g., search B) may or may not include terminology that is common with a preceding, base search query (e.g., search A) that is utilized to generate (or identify) the recommended search query.

In various embodiments, an algorithm implemented by the search and recommendation engine analyzes tracked (or logged) session data to determine other searches that are later in the path of page views for users who ran the same (or a similar) query as is currently being run, and who were eventually successful in a specified or predetermined activity (e.g., in placing a bid, in a network-based auction system).

Referring to both FIG. 2 and Table 1 a method is described for an algorithm 105 to generate search recommendations that can be offered to a user. At operation 110, Tracked data is monitored and raw session information is written based on LOOKBACK_PERIOD and SAMPLING_RATE configurations. At operation 120, the session data is trimmed by describing each unique page view by two variables only: the search phrase and the category constraint. The search term and category constraint are an input variable that refers to a combination of a keyword (or other search phrase or input) and a category constraint. The keyword can be null if there is a category constraint and the category constraint can be the root category if there is a keyword applied. For each Searching page (A), the search phrase (A1) and the category constraint (A2) are logged.

At operation 130, for every unique page view (A), it is determined which page was viewed immediately after (B), and which page was viewed immediately after that (C). At 140, each entry in the table (A, then B, then C) is split into two rows, where the first row is the unique page view in question followed by the next page view (A then B) and the second row is the page view in question followed by the page view after that (A then C). Now, each unique page view should have two rows in this table, unless the page view was the last or second to last in the user's session.

At operation 150, all Search page pairs are eliminated where the category constraint is not the same for both pages, or the search phrase is the same for both pages. This may leave a list of Searching page pairs where the category constraint has not changed but the search phrase has. In another embodiment, the search pairs are not eliminated where the category constraint is not the same, but the search phrase is the same. In this embodiment, the system generates recommendations to the user to perform the same search phrase in a different category. Further, where both the category constraint and search phrase change the system can provide a recommendation of expected success based upon prior user performances.

The number of times that each pair of Searching pages (A then B) appears in the exact same order is counted at operation 160 and that number is logged for each Searching page pair (A then B occurred N times). At operation 170, all Searching page pair counts (A then B occurred N times) are eliminated where the number of times that sequence was run (N) was less than the MIN_TRAFFIC_COUNT parameter.

All Searching page pairs are optionally eliminated at operation 180 where the category constraint is in the Mature Audiences category, any word in either of the two search phrases is on a blacklist for that site, or the search phrase contained complex search operators (such as minus, parentheses, etc.) For each starting Searching page (A), at operation 190 the method finds and records the series of “next Searching pages” (B's) in descending order of how many times the Searching page pair was counted (For A: B occurred 12 times, C occurred 9 times. D occurred 6 times. E occurred 4 times). Finally, at operation 195, the list of “next Searching pages” for each starting Searching page (A) is truncated so that it does not exceed the MAX_REL_SEARCHES parameter.

TABLE 1
Configuration	Description	Configurable By
SAMPLING_RATE	This is the percentage of	Site
	sessions that are fed into
	the Related Searches raw
	data in step 1 (below).
LOOKBACK_PERIOD	This is the amount of time	Site
	the Related Searches
	algorithm may look back
	to consider which
	searches are related to
	which Searching pages.
	Data older than this
	period may be ignored.
UPDATE_PERIODICITY	This is the length of time	Global
	that should go by between
	updates to the Related
	Searches output. It
	establishes a frequency
	for updating the
	recommendations and
	feeding those updates into
	Production.
MIN_TRAFFIC_COUNT	This is the minimum	Site
	number of users who may
	have independently run
	the same sequence of
	searches for a Related
	Search recommendation
	to be considered good
	enough to be counted.
MAX_REL_SEARCHES	This is the maximum	Global
	number of Related
	Searches that should be
	stored for a given search
	phrase-category pair.

Displaying Related Searches on Search, Listings, and Cross-Stores Search

The Related Searches module may appear on Searching pages as an additional line of links below the search box, as shown in the page view 300 of FIG. 3. The Related Searches placement is based on a current query that the user is running. To determine which related searches to display, the system may analyze the search phrase and the category constraint, ignoring everything else, except attributes.

Related Searches may not appear if certain predetermined conditions are present, for example the current query contains an attribute constraint, or the current query is a title and description search. Whether the Related Searches placement appears, in various embodiments, depends on whether there are recommendations available given the current search phrase and category constraint. Whether the Related Searches function is invoked also depend on whether other types of recommendations that also appear on the Searching page, in various embodiments.

Resolving Recommendation Conflicts

In one embodiment, there are multiple types of search modification (or enhancement) recommendations that can appear on Searching pages. For example, Related Searches, Spell Check. RIS (Recommendation in Search) and PTS (Product Titles in Search) search modification recommendations. Recommendations developed by the Relates Searches algorithm could conflict/overlap with some of the recommendations from the other systems. To avoid redundancy, conflict solution logic may determine how the recommendations co-exist, as explained below.

Resolving Conflicts with Spell Check

If a Spell Check recommendation surfaces (based on the existing surfacing rules), then the Related Searches module may not appear, in various embodiments. See FIG. 4 for a page view 400 of an example spell check recommendation correcting the term “chars” with “chairs”.

Resolving Conflicts with PTS

If one or more PTS (Product Titles in Search) recommendations surfaces (based on existing surfacing rules), then the Related Searches module may appear, but the list of recommendations appearing may be filtered to avoid displaying recommendations that overlap too closely with PTS.

In various embodiments, the only recommendations that may be allowed in the Related Searches module while PTS recommendations appear are ones where the search terms in the recommendation are not a superset of the search terms in the current query. Related Searches recommendations where additional keywords are added to the query but no keywords are changed may not be acceptable when PTS appears.

For example, if the user was searching on “Top Gun” in the DVDs category PTS recommendations appear; in this case a Related Searches recommendation for “Mission Impossible” is allowed, but a recommendation for “Top Gun DVD New” is not allowed. Referring to FIG. 5, a page view 500 example is illustrated wherein a search for “Deception Point” provides related searches for “Angels and Demons” and “Digital Fortress”.

Resolving Conflicts with RIS

The potential conflict between Related Searches and RIS (Recommendation in Search) can be handled by a new configuration (RIS_CONFLICT). There may be three possible values for RIS_CONFLICT, and the setting may determine exactly how Related Searches and RIS interact, see Table 2. As known to those in the art, a ‘widget’ is typically a rectangular section (like a module) on a web page used to demarcate different logical pieces of information on a web page. Most web pages are composed of various widgets showing different types of information. Sometimes users are given the choice to customize what widgets they want to see and which widgets they don't want to see.

TABLE 2
RIS_CONFLICT Setting             Required Interaction
RIS Wins                         If a RIS recommendation surfaces
                                 (based on the existing surfacing rules),
                                 then a Related Searches widget may not
                                 be shown
Related Searches Wins            If based on the rules described above, a.
                                 Related Searches widget would appear,
                                 than no RIS recommendations may be
                                 shown
RIS and Related Searches Coexist Related Searches and MS
                                 recommendations can appear
                                 independent of each other; however, in
                                 the event that the Related Searches
                                 widget does appear, the maximum
                                 number of RIS recommendations that
                                 could otherwise be shown (currently set
                                 to 3) would be reduced by 1 (down to 2)

The default setting for RIS_CONFLICT for all sites may be “RIS and Related Searches Coexist”. The example page view 700 of FIG. 7 shows how Related Searches and RIS recommendations may appear together based on this setting.

Determining the Recommendations to Display

When displayed subject to the conflict resolution rules above, the Related Searches widget may appear immediately below the search box. Recommendations may always appear as individual links oriented horizontally in a single row. Five site-specific configurations may govern the display properties of the Related Searches widget, as shown in Table 3.

TABLE 3
		Default
Configuration	Description	Values
MIN_ATTEMPTED_SA_RECOS	This is the minimum	All sites: 0
	number of search
	alternative-type
	recommendations that
	may be displayed before
	refined search-type
	recommendations are
	displayed
MIN_ATTEMPTED_RS_RECOS	This is the minimum	All sites: 0
	number of refined
	search-type
	recommendations that
	may be displayed before
	search alternative-type
	recommendations are
	displayed
MAX_REL_SEARCHES	This represents the	All sites: 5
	absolute maximum
	number of Related
	Searches
	recommendations
	allowed to appear in the
	Related Searches widget,
	regardless of space
	constraints.
TOTAL_CHAR_LIMIT	This represents the	All other
	maximum number of	sites: 100
	characters allowed for
	the sum of all the
	Related Searches
	recommendations
	displayed in the widget
	(including comma
	separators and
	spaces), ensuring the
	recommendations
	do not take
	up more than one line
	in the HTML.
RECO_TYPE_ORDER	This is the order in	All sites:
	which search	search
	alternatives and search	alternatives
	refinements relative	appear
	to each other.	before
		search
		refinements

Turning off Related Searches with MAX_REL_SEARCHES can be performed as follows. If MAX_REL_SEARCHES is set to 0, then Related Searches may not appear on any page (e.g., search, listings, cross-stores search, dynamic landing page (DLP), Personalized web page). Otherwise, MAX_REL_SEARCHES only has impact on the Searching pages (e.g., search, listings, cross-stores search). In other words, MAX_REL_SEARCHES influences search, listings, and cross-stores search only, but works as a universal feature shut-off switch when set to 0.

Given that MAX_REL_SEARCHES is not set to zero, recommendations may be selected to appear on search, listings, and cross-stores searches as follows. If the sum of MIN_ATIEMPTED_RS_RECOS and MIN_ATTEMPTED_SA_RECOS is greater than MAX_REL_SEARCHES, then both MIN_ATTEMPTEDRS_RECOS and MIN_ATTEMPTED_SA_RECOS may be ignored. (This situation makes no sense, and is an indication that the parameters have been incorrectly set.)

Referring to FIG. 6 a flow chart 600 is described for managing the recommendations to display. At operation 610 all the available recommendations coming from the Related Searches system for the current search phrase-category pair are fetched. At operation 620 the recommendations are organized into their respective types. If all of the keywords of the current search phrase can be found in the search phrase of the recommendation, then the recommendation type is refined search, else, the recommendation type is search alternative.

At operation 630 the top M search alternative-type recommendations are chosen (based on frequency count, as described in the Related Searches algorithm section), where M is the setting for MIN_ATTEMPTED_SA_RECOS. Also choose the top N refined search-type recommendations (based on frequency count, as described in the Related Searches algorithm section), where N is the setting for MIN_ATTEMPTED_RS_RECOS. The chosen recommendations are sorted at operation 640 by type, then by frequency count highest to lowest, where the preferred type is determined by the setting of RECO_TYPE_ORDER.

At operation 650, the next best recommendations of either type are chose until the total number of recommendations chosen is equal to MAX_REL_SEARCHES. These recommendations may be ordered after the recommendations already selected. The sort order within this set may be based on frequency count only. Finally, at operation 660 the complete list of chosen recommendations truncated starting from the last recommendation, until the total number of characters of the recommendations (taking into consideration the four characters that may separate each recommendation) is not greater than the MAX_CHAR parameter.

If, after executing these rules, there are no Related Searches recommendations to display, then the Related Searches widget may not appear at all, and the space it otherwise would have taken up would collapse. When presented, the recommendations in the Related Searches widget may always be sorted by highest-relevance first.

Display Properties

In various embodiments, Related Searches recommendations may be displayed based on the following rules. The section label may be “Related Searches” or the section label may be “Hot Keywords.” After each recommendation link (except the last recommendation link), a comma (not hyperlinked) may be shown. Additionally, characters may separate each recommendation, any words in a recommendation link that are being used in the current query may appear in bold, and any words in a recommendation link that are not being used in the current query may not appear in bold. For double byte sites the font size may be standard and for all other sites, the font size may be small.

FIG. 8A illustrates an example page view 800 for use in an English language location other Western locations. Similarly FIG. 8B illustrates an example page view 810 for a non-English speaking location.

Navigation

Clicking on a Related Search recommendation may navigate the user to another Searching page where the previous search phrase is altered to the new search phrase but all other search parameters/filters/sorts have been maintained. For example, if no category constraint had been applied before the recommendation was clicked, then no category constraint would be applied after. If a category constrain had been applied before the recommendation was clicked, then the same category constraint would also be applied after. Any additional filters (search options, tab selections, etc.) would also be maintained after the recommendation was clicked. The search that the user had applied would continue to be applied after the recommendation was clicked. If the user was in cross-stores search, he would remain in cross-stores search. If he was in core search, he would remain in core search.

ssPageName Tracking

ssPageNames may be added to the end of the anchors for each recommendation link. The formatting of the appended ssPageName may be &ssPageName=RELS:SA<X>:RS<Y>:<TYPE> where <X> may be an integer value (0−n). It may be the count of the number of search alternative-type recommendations that were displayed in the related searches widget <Y> may be an integer value (0−n). It may be the count of the number of refined search-type recommendations that were displayed in the related searches widget <TYPE> may be one of two values: RS or SA. RS may be the value if the recommendation type (for that link) is refined search and SA may be the value if the recommendation type (for that link) is search alternative

Examples of ssPageNames:

Example 1: If two search alternative recommendations and three refined search recommendations are shown, the ssPageName for each search alternative recommendation may be &ssPageName=RELS:SA2:RS3:SA and the ssPageName for each refined search-type recommendation may be &ssPageName=RELS:SA2:RS3:RS.

Example 2: If zero search alternative recommendations and four refined search recommendations are shown, the ssPageName for each refined search-type recommendation may be &ssPageName=RELS:SA0:RS4:RS.

Recommendation Organization

If there are no recommendations available for a current search, such as used by a DLP (dynamic landing page), then the Related Searches module may not appear at all. A DLP is an intermediate search page which provides the user with multiple navigation options based upon their original search query, therefore for the purposes of this invention it can be classified as a search page. If there are recommendations available for the current DLP search, then the Related Searches module may appear. To determine how to display the recommendations, they may first be organized into search refinements and search alternatives based on the same logic as described above.

If there are search refinements but not search alternatives to display, the search refinements may take up the entire module. Search refinements may be shown in three columns, ordered first by row, then by column, as shown below:

Link 1	Link 2	Link 3
Link 4	Link 5	Link 6
Link 7	Link 8

A maximum of 15 search refinements can be shown when only search refinements appear.

If there are search alternatives but not search refinements to display, the search alternatives may take up the entire module. Search alternatives may be shown in three columns, ordered first by row, then by column, as shown below:

Link 1	Link 2	Link 3
Link 4	Link 5	Link 6
Link 7	Link 8

A maximum of 15 search alternatives can be shown when only search alternatives appear.

If both types of recommendations (search refinements and search alternatives) are available for the DLP, then the recommendations may be split into two sections by a graphical horizontal separator. For each section, related searches may be shown in three columns, ordered first by row, then by column, as shown below:

Link 1	Link 2	Link 3
Link 4	Link 5	Link 6
Link 7	Link 8

The ordering of the two subsections in the DLP may depend on the RECO_PREFERENCE_TYPE parameter setting described above. If RECO_PREFERENCE_TYPE is Search Refinements, then search refinements may appear above search alternatives. If RECO_PREFERENCE_TYPE is Search Alternatives, the search alternatives may appear above search refinements.

Regardless of whether the widget is split, a label for the applicable recommendation type may appear above the recommendations. The label for search refinements can be “Search Refinements:” and the label for search alternatives can be “Search Alternatives:”.

In the base template, the total number of recommendation links appearing may be shown. (See site text “Top <N> related searches for <X>”.) N always represents the total number of both types of recommendations. This supplemental text is not included in the specific item template.

A “Base Template” user interface 900 design with only search refinements appearing is illustrated in FIG. 9A. A “Base Template” user interface 910 design with both recommendation types appearing is illustrated in FIG. 9B. Similarly, a “Specific Item Template” interface design 1000 with only refines searches appearing is illustrated in FIG. 10. FIG. 11 illustrates a “Specific Item Template” interface 1100 design with both recommendation types appearing.

Displaying Related Searches on the Personalized Web Page

Search pages, listing pages, and DLPs may require new functionality to write information to a cookie. When viewed, each of these types of pages may update the cookie. The following logic may be used to determine whether the Search results pages, Listings pages, and DLPs may update this cookie. If the search contains an attribute value constraint, do not update the cookie. If the search is a title and description search, do not update the cookie. If the search contains any complex search operators (such as minus or parentheses), do not update the cookie. Once these checks have been made, count the characters (bytes in the case of double-byte language) in the search phrase and assign that value to X, count the number of characters of the category ID and assign that value Y and sum X and Y. If X+Y is greater than MAX_COOKIE_BYTE_COUNT, do not update the cookie. Else, update the cookie with the exact string of the query and the category ID.

This information may be sufficient to be able to make recommendations in the personalized web page. The MAX_COOKIE_BYTE_COUNT setting may be 40. This ensures that no more than 40 bytes are taken up in the cookie, while maximizing the likelihood of being able to surface high quality recommendations on the personalized web page. This proposed value may be vetted through the appropriate channels to gain agreement.

Add Related Searches to Personalized Web Page

A new module within the personalized web page may be created to display recommendations coming from the Related Searches system. The input for the Related Searches module may be the search phrase and the category constraint of the last search that the user ran. This “last search” may be from the current session or from a previous session. It may be available in the cookie. The output may be a list of recommendation links. The related searches widget may appear in the order of event recency, which is how the existing modules may already be ordered in production. The related searches widget may be titled “Related Searches”.

If the Last Search Did not Include a Category Constraint:

The related searches widget may contain a subtitle of “Your last search was <X>. Here are some related searches.”<X> represents the search phrase of the last search. <X> may be hyperlinked and clicking it may navigate the user to the search results page for that search phrase. The requirements for displaying the recommendation links within the new Related Searches widget are exactly the same as for DLPs (described earlier). A Personalized web page 1200 for searching without category constraint is illustrated in FIG. 12.

If the Last Search Did Contain a Category Constraint:

The related searches widget may contain a subtitle of “Your last search was <X> (in <Y>). <X> represents the search phrase of the last search. If the category constraint was a meta level category, <Y> is the meta category name. If the category constraint was an L2 category, <Y> is the meta category name followed by “>” followed by the L2 category name. If the category constraint was L3 or below, <Y> is the meta category name followed by ellipses ( . . . ) followed by “>” followed by the name of the category that was applied as a constraint.

The entire string “<X> (in <Y>)” is hyperlinked and may navigate to run a search with that same search phrase and that same category constraint. When there is a category constraint, the subheadings for Search Refinements and Search Alternatives may also be modified. The subheading for the Search Refinements portion may be “Search Refinements (in the same category)”. The subheading for the Search Alternatives portion may be “Search Alternatives (in the same category)”.

The actual search refinement and search alternative links may all navigate to a search where those keywords are used, but constrained to the same category as the last search (which the recommendations are based on) was constrained to. FIG. 13 illustrates a Personalized web page 1300 for searching with category constraint.

Click-Through and Bid-Through Tracking

Click-through tracking may be accomplished through ssPageNames as defined above. Bid-through tracking and click-through rates may be determined. A variable may be added to indicate whether the Searching page contains related searches recommendations or not and whether the user clicked on a related searches recommendation to arrive at the current page. The values for this new variable may be for example 00, 01, 10, or 11, based on the following rules. The first digit may be 0 if the user did not click on Related Searches to arrive at the current page. The first digit may be 1 if the user clicked on Related Searches to arrive at the current page. The second digit may be 0 if Related Searches recommendations are not being shown on the current page. The second digit may be 1 if Related Searches recommendations are being shown on the current page.

Terminology

The term “listing” or “item” is used to provide an example of data, and may refer to any data item, description, identifier, representation or information pertaining to a listing, service, offering or request. For example, a listing may be an auction or fixed-price offering (e.g., products such as goods and/or services), an advertisement, or a request for a listing or service. For the purposes of this specification, the word “term” is synonymous with the word “phrase” and is also intended to include a plurality of words. Thus. “term” or “phrase” can be used to refer to any entry (or entries) a user enters into a search field when requesting a search of a data system. The term “term-category pair” (or phrase-category pair) may refer to a search term or phrase associated with a particular data category.

Transaction Facility

FIG. 14 is block diagram illustrating a network-based computer system 1410, within which various embodiments may operate. While an exemplary embodiment of the present invention is described within the context of the network-based computer system 1410 for the purposes of illustration, the invention will find application in many different types of computer-based, and network-based, facilities and data processing systems.

The network-based computer system 1410, includes one or more of a number of types of front-end servers that each includes at least one Dynamic Link Library (DLL) to provide selected functionality. The system 1410 includes page servers 1412 that deliver web pages (e.g., mark-up language documents), picture servers 1414 that dynamically deliver images to be displayed within Web pages, listing servers 1416 that facilitate category-based browsing of listings, search servers 1418 that handle search requests to the system 1410 and facilitate keyword-based browsing of data, and ISAPI servers 1420 that provide an intelligent interface to a back-end of the system 1410. The system 1410 also includes e-mail servers 1422 that provide, inter alia, automated e-mail communications to users of the network-based computer system 1410. In one embodiment, one or more administrative application functions 1424 facilitate monitoring, maintaining, and managing the system 1410. One or more API servers 1426 may provide a set of API functions for querying and writing to the network-based computer system 1410. APIs may be called through the HTTP transport protocol. In one embodiment, information is sent and received using a standard XML data format. Applications utilized to interact (e.g., upload transaction listings, review transaction listings, manage transaction listings, etc.) with the network-based computer system 1410 may be designed to use the APIs. Such applications may be in an HTML form or be a CGI program written in C++, Perl, Pascal, or any other programming language. Exemplary APIs are more fully described in co-pending U.S. patent application Ser. No. 09/999,618, herein incorporated by reference.

The page servers 1412, API servers 1426, picture servers 1414, ISAPI servers 1420, search servers 1418, e-mail servers 1422 and a database engine server 1428 may individually, or in combination, act as a communication engine to facilitate communications between, for example, a client machine 1430 and the network-based computer system 1410. In addition, the page servers 1412, API servers 1426, picture servers 1414, ISAPI servers 1420, search servers 1418, e-mail servers 1422 and database engine server 1428 may individually, or in combination, act as a transaction engine to facilitate transactions between, for example, the client machine 1430 and the network-based computer system 1410. Furthermore, the page servers 1412, API servers 1426, picture servers 1414. ISAPI servers 1420, search servers 1418, e-mail servers 1422 and database engine server 1428 may individually, or in combination, act as a display engine to facilitate the display of listings on, for example, the client machine 1430.

The back-end servers may include the database engine server 1428, a search index server 1432 and a credit card database server 1434, each of which maintains and facilitates access to a respective database.

In one embodiment, the network-based computer system 1410 is accessed by a client program, such as for example a browser 1436 (e.g., the Internet Explorer distributed by Microsoft Corp. of Redmond, Wash.) that executes on the client machine 1430 and accesses the network-based computer system 1410 via a network such as, for example, the Internet 1438. Other examples of networks that a client may utilize to access the network-based computer system 1410 include a wide area network (WAN), a local area network (LAN), a wireless network (e.g., a cellular network), the Public Switched Telephone Network (PSTN) network, or the like. The client program that executes on the client machine 1430 may also communicate with the network-based computer system 1410 via the API servers 1426.

Database Structure

FIG. 15 is a database diagram illustrating an exemplary database 1540, maintained by and accessed via the database engine server 1428, which at least partially implements and supports the network-based computer system 1410.

In one embodiment, the database engine server 1428 may maintain two databases, a first database being maintained for listing (or offering) information that is not included within a virtual “store”, and a second database for listing (or offering) information that is presented via a virtual “store” supported by the network-based computer system 1410.

The database 1540 may, in one embodiment, be implemented as a relational database, and includes a number of tables having entries, or records, that are linked by indices and keys. In an alternative embodiment, the database 1540 may be implemented as collection of objects in an object-oriented database.

The database 1540 includes a user table 1542 that contains a record for each user of the network-based computer system 1410. A user may operate as a seller, a buyer, or both, when utilizing the network-based computer system 1410. The database 40 also includes listings tables 44 that may be linked to the user table 42. The listings tables 44 may include a seller listings table 46 and a bidder listings table 48. A user record in the user table 42 may be linked to multiple listings that are being, or have been, listed or offered for sale via the network-based computer system 10. In one embodiment, a link indicates whether the user is a seller or a bidder (or buyer) with respect to listings for which records exist within the listings tables 44.

The database 1540 also includes one or more divisions in the form of categories provided in category tables 1550. Each record within the category table 1550 may describe a respective category. In one embodiment, listings provided by the system 10 are arranged in the categories. These categories may be navigable by a user of the network-based computer system 1410 to locate listings in specific categories. Thus, categories provide a mechanism to locate listings that may be browsed. In addition or instead, an alphanumeric search mechanism may be provided by the search servers 1420 to allow a user to search for specific listings using search terms or phrases. In one embodiment, the category table 1550 describes multiple, hierarchical category data structures, and includes multiple category records, each of which describes the context of a particular category within the multiple hierarchical category structures. For example, the category table 1550 may describe a number of real, or actual, categories to which listing records, within the listings tables 1544, may be linked.

The database 1540 also includes one or more attributes tables 1552. Each record within the attributes table 1552 describes a respective attribute associated with a listing. In one embodiment, the attributes table 1552 describes multiple, hierarchical attribute data structures, and includes multiple attribute records, each of which describes the context of a particular attribute within the multiple hierarchical attribute structures. For example, the attributes table 1552 may describe a number of real, or actual, attributes to which listing records, within the listings tables 1544, may be linked. Also, the attributes table 1552 may describe a number of real, or actual, attributes to which categories, within the category table 1550, may be linked.

The database 1540 may also include a note table 1554 populated with note records that may be linked to one or more listing records within the listings tables 1544 and/or to one or more user records within the user table 1542. Each note record within the note table 1554 may include, inter alia, a comment, description, history or other information pertaining to a listing being offered via the network-based computer system 1410, to a user of the network-based computer system 1410. The database 1540 may also include a targeted site table 1556 populated with targeted site records that may be linked to one or more listing records within the listings tables 1544 and/or to one or more user records within the user table 1542.

A number of other example tables may also be linked to the user table 1542, namely a user past aliases table 1558, a feedback table 1560, a feedback details table 1562, a bids table 1564, an accounts table 1566, and an account balances table 1568. In one embodiment, the database 1540 also includes a batch table 1570, a batch listings table 1572, and a listings wait table 1574.

In one embodiment, the system 1410 generates search recommendations to a user of the system 1410. The search recommendations may be based on past user interaction of the particular user with the system 1410, and search terms used in the network-based computer system 1410 (or any other systems associated with the network-based computer system 1410).

Referring to FIG. 16, reference numeral 1680 generally indicates example logic for generating recommendation queries based on past user interaction in the form of past bidding (and/or buying) history of the user, and the search terms. As shown at block 1682 past bidding (and/or buying) data of participating users is gathered at a data warehouse. In addition, popular search terms or phrases are gathered at block 1684 that, together with the past bidding (and/or buying) data is used to generate recommendation queries (see block 1686). Thus, the data warehouse may identify and store search terms that are used most frequently (popular search terms) across one or more predetermined number of sites (e.g., web sites) associated with the network-based computer system 1410, and also identify data uniquely associated with each user. As shown at block 1688, the popular search terms may then be passed periodically (e.g., on a daily basis) to a production facility, where the production facility may then project the popular search data against current listing inventory (see block 1690). In one embodiment, a search is conducted through each category, at each category level, using each popular search term. All popular search terms that match at least a predetermined number of listings (e.g., 50 listings), located in each particular category, may be stored along with total number of listings located in the particular category using the popular search term. Thus, each category may have a number of popular search terms or phrases (e.g., from 0 to a predetermined number) assigned to it along with a measurement of the popularity of the search term in that category. Thus, the system 10 allows a search to be conducted through current listings based on popular searches (based on interaction of all users) and unique historical interaction of an

FIG. 17 shows a diagrammatic representation of a machine in the example form of a computer system 1700 within which a set or sequence of instructions, for causing the machine to perform any one of the methodologies discussed herein, may be executed. In alternative embodiments, the machine may comprise a network router, a network switch, a network bridge, Personal Digital Assistant (PDA), a cellular telephone, a web appliance, set-top box (STB) or any machine capable of executing a sequence of instructions that specify actions to be taken by that machine.

The computer system 1700 includes a processor 1702, a main memory 1704 and a static memory 1706, which communicate with each other via a bus 1708. The computer system 1700 may further include a video display unit 1710 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)). The computer system 1700 also includes an alphanumeric input device 1712 (e.g., a keyboard), a cursor control device 1714 (e.g., a mouse), a disk drive unit 1716, a signal generation device 1718 (e.g., a speaker) and a network interface device 1720 to interface the computer system to a network 1722.

The disk drive unit 1716 includes a machine-readable medium 1724 on which is stored a set of instructions or software 1726 embodying any one, or all, of the methodologies described herein. The software 1726 is also shown to reside, completely or at least partially, within the main memory 1704 and/or within the processor 1702. The software 1726 may further be transmitted or received via the network interface device 1720. For the purposes of this specification, the term “machine-readable medium” shall be taken to include any medium which is capable of storing or encoding a sequence of instructions for execution by the machine and that cause the machine to perform any one of the methodologies of the present invention. The term “machine-readable medium” shall accordingly be taken to include, but not be limited to, solid-state memories, optical and magnetic disks, and carrier wave signals. Further, while the software is shown in FIG. 17 to reside within a single device, it will be appreciated that the software 1726 could be distributed across multiple machines or storage media, which may include the machine-readable medium.

Although the invention has been described with reference to various embodiments, it will be evident that various modifications and changes may be made to these embodiments without departing from the broader spirit and scope of the invention. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense.

Claims

1. A method comprising: tracking activity with respect to a first query result of a first query and a second query result of a second query, the activity being performed by users of a network publication system;counting, from the tracked activity, a number of search events corresponding to users that viewed the second query result subsequent to the first query result and performed post search activity on the second query result;determining, using one or more processors, that the counted number of search events that corresponds to users that viewed the second query result subsequent to the first query result and performed post search activity on the second query result exceeds a predefined threshold;counting a further number of search events corresponding to users that viewed a third query result after the first query result and performed post search activity on the third query result, the third query result presented in response to a third query;determining that the further number of search events exceeds the counted number of search events;designating the third query as a recommended query for the first query;receiving the first query from a client device of a subsequent user;based on receiving the first query from the client device, providing a first set of results to the client device to be displayed in a first widget of a web interface; andbased on the number of search events exceeding the predefined threshold and receiving the first query, recommending the second query within a second widget of the web interface;wherein the web interface is customizable by the subsequent user regarding visibility of the first widget and the second widget.

2. The method of claim 1, wherein: the tracking the activity includes determining that a category constraint for the first query is the same as a category constraint for the second query; andthe counting the number of search events is based on the determining that the category constraint for the first query is the same as the category constraint for the second query.

3. The method of claim 2, further comprising causing display of an indication that the category constraint for the first query is the same as the category constraint for the second query via a search user interface on the client device.

4. The method of claim 1, wherein: the tracking the activity includes determining that a search phrase for the first query is not the same as a search phrase for the second query, andthe counting the number of search events is based on the determining that the search phrase for the first query is not the same as the search phrase for the second query.

5. The method of claim 1, wherein the recommending the second query includes causing display of the second query in a search user interface on the client device.

6. The method of claim 5, wherein the causing the display of the second query in the search user interface includes causing display of a link in the search user interface, the link being selectable to perform a search using the second query.

7. The method of claim 1, wherein: the second query result indicates a product page published by the network publication system, the product page featuring a product; andthe post search activity on the second query result includes purchasing the product featured in the product page.

8. The method of claim 1, further comprising: determining that the second query includes all keywords in the first query, andcausing display of an indication that the second query is a refined query for the first query based on the determining that the second query includes all the keywords in the first query.

9. The method of claim 1, further comprising: determining that the second query does not include all keywords in the first query, andcausing display of an indication that the second query is an alternative query for the first query based on the determining that the second query does not include all keywords in the first query.

10. A system comprising: one or more processors; andone or more non-transitory computer readable media containing instructions that, when executed by the one or more processors, cause the system to perform operations comprising:track activity with respect to a first query result of a first query and a second query result of a second query, the activity being performed by users of the system,count, from the tracked activity, a number of search events corresponding to users that viewed the second query result subsequent to the first query result and performed post search activity on the second query result;determine that the counted number of search events that corresponds to users that viewed the second query result subsequent to the first query result and performed post search activity on the second query result exceeds a predefined threshold;count a further number of search events corresponding to users that viewed a third query result after the first query result and performed post search activity on the third query result, the third query result resented in response to a third query;determine that the further number of search events exceeds the counted number of search events;designate the third query as a recommended query for the first query;receive the first query from a client device of a subsequent user;based on receiving the first query from the client device, provide a first set of results to the client device to be displayed in a first widget of a web interface; andbased on the number of search events exceeding the predefined threshold and receiving the first query, recommend the second query within a second widget of the web interface;wherein the web interface is customizable by the subsequent user regarding visibility of the first widget and the second widget.

11. The system of claim 10, wherein the operations further comprise: determine that a category constraint for the first query is the same as a category constraint for the second query; andcount the number of search events based on the determining that the category constraint for the first query is the same as the category constraint for the second query.

12. The system of claim 11, wherein the operations further comprise: cause display of an indication that the category constraint for the first query is the same as the category constraint for the second query via a search user interface on the client device.

13. The system of claim 10, wherein the operations further comprise: determine that a search phrase for the first query is not the same as a search phrase for the second query; andcount the number of search events based on the determining that the search phrase for the first query is not the same as the search phrase for the second query.

14. The system of claim 10, wherein the operations further comprise cause display of a link in the search user interface, the link being selectable to perform a search using the second query.

15. The system of claim 10, wherein: the second query result indicates a product page published by the system, the product page featuring a product; andthe post search activity on the second query result includes purchasing the product featured in the product page.

16. A non-transitory machine-readable medium storing instructions that, when executed by one or more processors of a machine, cause the machine to perform operations comprising: tracking activity with respect to a first query result of a first query and a second query result of a second query, the activity being performed by users of a network publication system;counting, from the tracked activity, a number of search events corresponding to users that viewed the second query result subsequent to the first query result and performed post search activity on the second query result;determining that the counted number of search events that corresponds to users that viewed the second query result subsequent to the first query result and performed post search activity on the second query result exceeds a predefined threshold;counting a further number of search events corresponding to users that viewed a third query result after the first query result and performed post search activity on the third query result, the third query result resented in response to a third query;determining that the further number of search events exceeds the counted number of search events;designating the third query as a recommended query for the first query;receiving the first query from a client device of a subsequent user;based on receiving the first query from the client device, providing a first set of results to the client device to be displayed in a first widget of a web interface; andbased on the number of search events exceeding the predefined threshold and receiving the first query, recommending the second query within a second widget of the web interface;wherein the web interface is customizable by the subsequent user regarding visibility of the first widget and the second widget.

17. The system of claim 10, wherein the operations further comprise: determining that the second query includes all keywords in the first query, andcausing display of an indication that the second query is a refined query for the first query based on the determining that the second query includes all the keywords in the first query.

18. The system of claim 10, wherein the operations further comprise: determining that the second query does not include all keywords in the first query, andcausing display of an indication that the second query is an alternative query for the first query based on the determining that the second query does not include all keywords in the first query.