UNIVERSAL QUERY SEARCH RESULTS

Abstract

Generally described, embodiments of the present invention provide the ability to generate a universal query result set from several different search index result sets by including identifications of items from the different search index result sets in an accurate manner. After a query has been submitted, search index result sets are received from several different search indexes, an allocation score for each search index is computed and a universal item score for the top-level item identified in each search index is computed. The method then combines the allocation score and the universal item score for the top level item for each search index result set and adds the item with the highest combined score to the universal query result set.

Description

CROSS-REFERENCES TO RELATED APPLICATIONS

This application is a continuation of U.S. patent application Ser. No. 13/934,128 filed Jul. 2, 2013, entitled “UNIVERSAL QUERY SEARCH RESULTS,” Attorney Docket Number 20347.0015.CNUS00 which is a continuation of U.S. patent application Ser. No. 12/793,628, filed Jun. 3, 2010, entitled “UNIVERSAL QUERY SEARCH RESULTS,” Attorney Docket Number 20347.0015.PCUS00 of which is a continuation of U.S. patent application Ser. No. 11/506,495, filed Aug. 18, 2006, entitled “UNIVERSAL QUERY SEARCH RESULTS,” Attorney Docket Number 20347.0015.NPUS00, of which are hereby incorporated in their entirety by reference.

BACKGROUND

With the increase in electronically accessible material, the ability to Find that material in an organized and efficient manner is becoming more difficult. One technique typically used for finding and organizing electronically accessible material is through the use of search engines. For example, a user may submit a query containing one or more parameters that are used to search for items that are identified in response to the query. The parameters are compared to an index, referred to herein as a “search index,” containing identifications of electronically accessible material, and ordered listings of items matching the query parameters are returned. The search index, upon identifying items matching the query parameters, typically organizes or ranks the identification of those items based on how close they match the query parameters. The organized items are then returned to the searching entity in the form of a query result set.

While the above searching technique works well for searching a single search index and obtaining an ordered query result set from that search index, it does not provide an efficient means for searching multiple indexes and obtaining a single ordered query result set. Using the above technique, an entity would have to individually search each search index and then manually review each of the different query results sets. One technique that has been implemented to attempt to resolve this deficiency is through the simple merging of multiple query result sets based on the ranking of individual items identified within those sets. However, because different search indexes utilize different properties and definitions for determining the relevance of items matching the parameters of a query, the simple merged query result set will likely not contain an accurate ordering of the item identifications contained therein. Because different search indexes utilize different definitions and parameters for ranking matching items, an item of higher actual relevance to the query may be ranked lower than another item from a different search index that is of less actual relevance because of the different factors used in ranking those items by the different search indexes.

Accordingly, there is a need for a system and method for merging identifications of items from separate query result sets into a single query result set such that those identifications of items are more accurately ordered based on actual relevance to the query.

SUMMARY

This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This summary is not intended to identify key features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

In accordance with an aspect of the present invention, a method for generating a universal query result set is provided. Generally described, the method includes receiving a query containing at least one parameter and submitting that query to a plurality of search indexes. After the query has been submitted, search index result sets are received from each of the plurality of search indexes. The search index result sets each contain different lists of items that, based on the information maintained by the respective search index, best matches the parameters of the query. After the search index result sets have been received, an allocation score for each search index is computed and a universal item score for the top-level item identified in each search index is computed. For each search index, the method then combines the allocation score and the universal item score for the top level item from the same search index. The item with the highest combined score is then added to the universal query result set. Generally, the item with the highest combined score will be the item that has the highest probability of satisfying the query.

In accordance with another aspect of the present invention, a computer system having a computer-readable medium including a computer-executable program therein for performing a method for generating a universal query result set in response to a query is provided. The computer-executed method includes queuing a first search index result set and a second search index result set. Each of the index result sets are ordered lists identifying items related to the query. Additionally, because each of the index result sets are provided by different search indexes, the lists themselves will be different. Based on items identified in the lists, an item having a highest probability of satisfying the query is determined and added to the universal query result set.

In accordance with another aspect of the present invention, a computer-readable medium having instructions stored thereon that direct a computing system to generate a universal query result set from a group of search index result sets is provided. The instructions direct the computing system to submit a query to a plurality of search indexes and queue the search index result sets as they are received. Each of the plurality of search index result sets are ordered lists identifying items related to the query. Additionally, a query i:r;tdex association for each of the search indexes is determined for the given query. Based on the items contained in the queued search index result sets and the query index associations, a universal query result set that includes a list of items obtained from the plurality of search index result sets is created. The universal index result set includes an ordered list based on a universal probability of those items satisfying the query, wherein the list of items contains items obtained from different search index result sets.

DESCRIPTION OF THE DRAWINGS

The foregoing aspects and many of the attendant advantages of this invention will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

FIG. 1 is a state diagram illustrating a universal query routine for obtaining search results from multiple different search indexes, according to embodiments of the present invention;

FIG. 2 is a block diagram of the search index result sets obtained in response to a query, generated in accordance with an embodiment of the present invention;

FIG. 3 is a graphical representation of a universal query result set generated by combining the identification of items from the different search index result sets returned in response to a query, in accordance with an embodiment of the present invention;

FIG. 4 is a flow diagram of a universal query routine in which results from separate search indexes are to be obtained, combined, and returned for a received query, in accordance with an embodiment of the present invention;

FIG. 5 is a flow diagram of the query index association subroutine for computing query index associations for each determined search index, in accordance with an embodiment of the present invention;

FIG. 6 is a flow diagram of the allocation score subroutine for computing an allocation score for each determined search index for which a search index result set is being obtained, in accordance with an embodiment of the present invention; and

FIG. 7 is a flow diagram of a universal item score subroutine for computing a universal item score for the next top order item listed in each of the search index result sets that have been queued, in accordance with an embodiment of the present invention.

DETAILED DESCRIPTION

FIG. 1 is a state diagram illustrating a universal query routine for obtaining search results from multiple different search indexes, according to embodiments of the present invention. Through the use of a query controller 113, search results (referred to herein as “search index result sets”) may be obtained from multiple search indexes, such as a books search index 101, DVD search index 103, music search index 105, software search index 107, and apparel search index 111. After obtaining the search index result sets, the query controller combines identifications of items from different search index result sets to generate a universal query result set. As described in more detail below, the identifications of items selected and organized into the universal query result set are organized such that the items listed in the universal query result set are those items, out of all items identified by the different search index result sets, that have the highest probability of satisfying the query.

While the state diagram of FIG. 1 illustrates the different search indexes 101-111 as being different physical objects, as will be appreciated by one of skill in the relevant art, the different search indexes may all be on one computing device, multiple different computing devices, or any combination there between. Additionally, the query controller 113 may reside on a computing device with one or more of the search indexes, on a stand-alone computing device, or a computing device that is used for submitting a query. Similarly, the search indexes may identify any type of item and those listed and used herein are provided as examples only. Moreover, more than one search index may identify the same (partially or wholly) or similar items. Additionally, the search indexes may be part of a single entity maintaining multiple search categories or randomly dispersed throughout a computing network, such as the Internet.

At an initial point, a query is submitted to the query controller 113 using a computing device, such as computing device 115. The query has one or more parameters, or search terms, that are used to identify items that are relevant to the query. A query may be submitted automatically, through an API, or input by a user who desires to obtain universal search results.

Upon receiving a query, such as a query containing the search parameter “Howard,” the query controller 113 determines an appropriate set of search indexes for which query result sets are to be obtained. For example, based on the query parameters submitted to the query controller 113, the query controller 113 may identify a subset from a larger set of search indexes that are most likely relevant to the query parameters. For example, the query controller 113 may identify similar historical queries and select search indexes that provided identifications of items for those historical queries, or select search indexes that provided identifications of items that were selected from those historical queries. Still further, the query may include additional parameter identifiers that may be used by the query controller to identify the most likely search indexes for use in identifying items relevant to the query. For example, in addition to a query containing the parameter “Howard” the entity that generated the query may also identify that the desired item will be of a particular type, such as entertainment, movies, books, music, etc. Alternatively, the query controller 113 may identify all search indexes associated with that query controller 113 for use in obtaining index query result sets.

Upon identification of the search indexes, the query controller 113 submits the query to the determined search indexes and obtains search index query result sets from each of the different search indexes 101-111. Search index result sets may be generated and returned in any manner. For embodiments of the present invention, the only assumption made about returned search index result sets is that the items listed in the set are ordered or identified by relevance with respect to the submitted query.

In addition to submitting query parameters to determined search indexes and obtaining search index result sets, the query controller 113 computes, or determines, query index associations for each of the determined search indexes, as described in more detail below with respect to FIG. 5. Based on the computed query index associations and the obtained search index result sets, the query controller 113 determines the identified item having the highest probability of satisfying the query.

FIG. 2 is a block diagram of the search index result sets obtained in response to a query having the parameters “Howard” submitted to the books search index 101, DVD search index 103, music search index 105, software search index 107, and apparel search index Ill. As shown, the content of the search index result sets are ordered by relevance with respect to the submitted query parameters. For example, using the query parameters of “Howard,” the books search index 101 returns an ordered search index result set 201 that includes an identification of five items matching the parameter “Howard.” Similarly, the DVD search index returns an ordered search index result set 203 listing DVDs that match the search parameter “Howard.” The music search index 105 returns an ordered search index result set 205 listing music, such as CDs, matching the search parameter “Howard.” The software search index 107 returns an ordered software search index result set 207 identifying software that matches the query parameter “Howard.” Finally, the apparel search index 111 returns an apparel search index result set 211 identifying an ordered list of apparel items having a name that matches the query parameter “Howard.” While this example uses five different search indexes, any number of search indexes may be used for obtaining search index result sets. Similarly, each search index result set may identify any number of items.

Typically, search indexes contain internal relevance ranking functions that are used to order items to be returned in response to a submitted query. However, different search indexes, such as the book search index 101 and the apparel search index 111, often have different internal relevance ranking functions for use in ordering identifications of items returned in response to a query. As can be seen from FIG. 2, the ranking or relevance of items may be determined based on a variety of ranking criteria. For example, the books search index 101 and the DVD search index 103 may rank items based on a percentage value 213 of relevance. In contrast, the music search index 105 and software search index 107 may rank items based on a scaled range of numbers 215.

Any type and combination of ranking criteria may be used with embodiments of the present invention and those discussed herein are provided only as examples. Because different search indexes have different relevance ranking functions and order items based on different ranking criteria, a direct merging of results does not generate an accurate universal query result set. However, in accordance with embodiments of the present invention, the ranking criteria may be normalized across all search index result sets, and through the use of additional techniques of combining independent search results, discussed below, a higher likelihood of an appropriate listing of items from the different search index result sets may be obtained and used for generating a universal query result set. Normalization of multiple ranking criteria may be accomplished using any normalization technique. For example, all ranking criteria may be modified to a set percentage ranging (e.g., 0%-100%). Alternatively, all ranking criteria may be adjusted to all fall within a standard range (e.g., 0-100).

Returning to FIG. 1, the query controller 113 determines the highest probability matches from the different items from the independent search index result sets for use in adding to the universal query result set. Techniques for determining the highest probability matches are described in more detail below with respect to FIGS. 4-7. Generally described, the probability is determined based on existing content of the universal query result set, the likelihood of a particular search index returning an identification of an item that will satisfy the query, historical queries, and a relevance score of the items identified by the different search indexes. For example, FIG. 3 is a graphical representation of a universal query result set generated by combining the identification of items from the different search index result sets 201-211 returned in response to a query for the query parameter “Howard,” in accordance with an embodiment of the present invention. In particular, the universal query result set 300 identifies that the highest probability item that may satisfy the query parameter “Howard” is the book “Howard, the Autobiography” 301 and the second highest probability item being the DVD “Howard and God” 303.

In contrast, a simple merging of the independently returned search index result sets could result in an ordered listing of items having the book “Howard, the Autobiography” as the highest probability, followed by the book “Howard's Evil Empire.” As described above, such a simple merging may not result in an accurately ordered list of items because the different search indexes order items based upon different criteria. Because the query controller 113 utilizes embodiments of the present invention for identifying the items that have the highest probability of satisfying the query parameters, the items listed in the universal query result set may be ordered differently than a simple merging of search index results. Additionally, items from different search index results that are ranked using different ranking techniques may also be merged based on an overall probability. Utilizing an overall probability, items with a lower ranking, as assigned by the returning search index, may be ordered higher in the universal query result set than an item from another search index that has a higher ranking. For example, the query result item “Howard Visits Rome” 305 has a ranking of 95% as assigned by the DVD search index 103, whereas the item “Howard's Evil Empire” 307 has a ranking of 99% as assigned by the books search index I 01. Thus, as can be seen from the universal query result set 300 illustrated in FIG. 3, the items listed in the universal query result set are not simply an ordering of the different items based on rankings assigned by the different search indexes.

Once the desired number of items have been identified and added to the universal query result set, the query controller 113 returns the universal query result set to the source of the query. In alternative embodiments, the query controller 113 may generate and deliver the universal result set to a separate location independent of the location that submitted the query. The entity receiving the universal query result set may then interact with the universal result set as if it were a typical result set. As interaction with the universal result set occurs, the query controller 113 tracks and retains those interactions for use in generating future universal query results in response to queries containing similar or the same parameters. The query and the tracked interactions are referred to herein as a “historical query.” Because ordering of items in the universal query result is partially dependent upon results of similar queries, a subsequent query for the parameter “Howard” may return a different universal query result set. For example, if interactions with the universal query result set 300 (FIG. 3) often resulted in the purchase of the DVD “Howard and God” 303, a future query for the same parameter may result in the DVD “Howard and God” 303 being ordered as the most probable item satisfying the query.

FIG. 4 is a flow diagram of a universal query routine in which results from separate search indexes are to be obtained, combined, and returned for a received query, in accordance with an embodiment of the present invention. At an initial point, at block 401 a query is received and in response to receiving that query, the search indexes to be queried is determined, as illustrated by block 403. In an alternative embodiment, all available search indexes may be selected. Upon determination of the search indexes to be queried, at block 405 the received query is submitted to each of the determined search indexes. After submitting the query to the determined search indexes, search index result sets are received from each of the determined search indexes and queued for use in generating a universal query result, as illustrated by block 407. Additionally, if the search index results sets utilize different ranking criteria, the ranking criteria may be normalized to a common ranking criteria.

At subroutine block 409, a query index association is computed, or determined, for each of the determined search indexes, as described in more detail below with respect to FIG. 5. Additionally, an allocation score is computed for each determined search index, as illustrated by subroutine block 411. Techniques for computing allocation scores for determined search indexes are described in more detail below with respect to FIG. 6. At subroutine block 413, a universal item score is computed for each of the top ordered items listed in each of the search index result sets that have not previously been added to the universal query result set, as described in more detail below with respect to FIG. 7.

Upon computing the allocation scores and the universal item scores at subroutine blocks 411 and 413, at block 415, for each top order item from the queued search index result sets, or for each queued search index result set, the allocation score and universal item score are combined. Any technique may be used for combining the allocation score and the universal item score. For example, the allocation score and the universal item score may be combined using addition, multiplication, averaging, etc.

At block 417, the identification of the top order item with the highest combined score is added in an ordered fashion to the universal query result set. As the universal query routine 400 continues, identifications of items are selected and added in order to generate a complete universal query result set. At decision block 419, a determination is made as to whether additional results are needed for completion of the universal query result set. If it is determined at decision block 419 that additional results are needed, the universal query routine 400 returns to subroutine block 411 and continues. However, if it is determined at decision block 419 that additional results are not needed, at block 421 the universal query result set is returned and the universal query routine 400 completes, as illustrated by block 423.

FIG. 5 is a flow diagram of the query index association subroutine 409 for computing query index associations for each determined search index, in accordance with an embodiment of the present invention. At decision block 500 it is determined whether a query index association for the determined search indexes exist for the query. For example, as query index associations are generated for search indexes, based on a given query, they may be stored and reused for subsequent queries having the same or similar parameters. Alternatively, or in addition thereto, at predetermined times (e.g., nightly) query index associations may be computed for all search indexes based on prior queries. If it is determined at decision block 500 that query index associations exist for the determined search indexes for the query, at block 504 the query index associations are retrieved and returned at block 506. Query index associations for determined search indexes may be determined to exists for the query if the historical queries used to generate the query index associations have the same or similar parameters to that of the existing query, as described with respect to block 501 through 503, and 505. If it is determined at decision block 500 that query index associations for the determined search indexes do not exist for the query, at block 501, historical queries having the same or similar parameters are identified. Historical queries having similar parameters may be those that have at least one common parameter to that of the query. Upon determination of historical queries, at block 502 “appropriateness scores” for the determined search indexes are generated based on the actions tracked and stored as part of the identified historical queries. An “appropriateness score,” as used herein, is a score, such as a numerical value, that is computed and assigned to each determined search index based on actions recorded as part of the determined historical queries. For example, each historical query may contain one or more user actions. Actions include, but are not limited to, selecting an identified item for viewing, purchasing an identified item, selecting the item for potential purchase, selecting a category of an item, etc. The appropriateness score may be different based on the action. For example, the purchase of an item may be given a higher score than the viewing of an item. For each action, the search index, or indexes, from which the item was provided is identified and an appropriateness score for that action is assigned to the identified search index(es).

In addition to generating appropriateness scores based on the action, those scores may be weighted or adjusted, as illustrated by block 503. Any type of weighting may be applied to the appropriateness scores. For example, the appropriateness scores may be weighted or adjusted based on time, similarity of the historical query parameters to those of the current query parameters, etc. In one embodiment, appropriateness scores determined for older historical queries may be given a lower weighting than those of more recent historical queries. Similarly, appropriateness scores determined for historical queries that are similar to those of the current query may be given a lower weighting than those that exactly match the current query.

Based on the appropriateness scores assigned to the determined search indexes at block 502, query-index associations are computed for each determined search index, as illustrated by block 505. In one embodiment, the query index associations are computed for each search index by summing the assigned appropriateness scores. In essence, the query index association score for each search index is a probability estimate of a given query, and that all else being equal, the chance that the result that is desired originates from the determined search index. After the query index associations for each determined search index have been computed, at block 506, the query index association subroutine 409 returns the determined query index associations to the universal query routine 400 (FIG. 4), as illustrated by block 509 and completes.

In embodiments in which query index associations are computed at predetermined times (e.g., nightly), instead of re-computing all the appropriateness scores for historical queries each time, embodiments of the present invention may utilize the previously computed query index associations to generate updated query index associations that take into consideration recent queries. For example, for each query that had not previously been considered in generating query index associations, appropriateness scores are determined and assigned to the appropriate search indexes and current query index associations are computed for each search index for the query. The pre-existing query index associations for those search indexes for that same or similar query may then be weighted based on time to reduce the value of older query index associations and combined with the current query index association for the query and search index. Utilization of pre-computed query index associations may improve efficiency of real-time query processing.

FIG. 6 is a flow diagram of the allocation score subroutine 411 for computing an allocation score for each determined search index for which a search index result set is being obtained, in accordance with an embodiment of the present invention. At block 601, the composition of items listed in the universal query result set is determined.

For example, referring back to FIG. 3, it may be determined that the composition of the universal query result set being constructed lists one item from the book search index and two items from the DVD search index. Based on the determined composition of items listed in the universal query result set, as determined at block 601, and based upon the computed query index association for a determined search index, an allocation score for the determined search index may be determined, as illustrated by block 603. In particular, if the composition of listed items in the universal query result set determined at block 601 has a larger percentage of items from one search index, such as the DVD search index, the allocation score for the DVD search index will be lower because there is a lower probability that the next most likely item will also be identified from that same search index. Likewise, the allocation score computed for a determined search index other than the DVD search index, such as the book search index, will be a higher score because there is a higher probability that the next most likely matching result will come from that determined search index, instead of the DVD search index. Upon computation of the allocation score for a determined search index at block 603, at decision block 605, a determination is made as to whether additional determined search indexes exist for which an allocation score is to be computed. If it is determined at decision block 605 that additional determined search indexes exist, the allocation score subroutine 600 returns to block 603 and continues. However, if it is determined at decision block 605 that no additional determined search indexes exist for which an allocation score still needs to be computed, the allocation score subroutine 600 returns the computed allocation scores to the universal query search routine 400 (FIG. 4) as illustrated by block 607 and completes.

FIG. 7 is a flow diagram of a universal item score subroutine for computing a universal item score for the top order item listed in each of the search index result sets that have been queued, in accordance with an embodiment of the present invention. At block 701, a relevance score for a search index result set that has been queued by the universal query routine 400 is determined. A relevance score for a search index result set may be determined based on knowledge about the internal ranking functions of the determined search index from which the search index result set was obtained. As discussed above, while scores for items contained in different search index result sets received from different search indexes are not directly comparable, they can be used to compare items within a same index and a goodness score may be computed to represent, given the search index, how good a match the particular given result is.

Upon determination of a relevance score for a search index result set, at block 703, a universal item score for the top order item listed in the search index result set that has not already been included in the universal query result set is computed. A universal item score may be computed based on the relevance score of the search index result set from which it was obtained, the existing contents of the universal query result set, based on the relevance score assigned to the item in the search index result set by the determined search index that generated that set, or any combination thereof. At decision block 705, a determination is made as to whether additional search index result sets exist for which a universal item score for the next result in the search index result set is to be computed. If it is determined at decision block 705 that additional index result sets exist for which a universal item score for the top order item listed in the search index result set is to be computed, the universal item score subroutine 415 returns to block 701 and continues. However, if it is determined that additional search index result sets do not exist, at block 707, the computed universal item scores are returned to the universal query routine 400 as illustrated by block 707 and the routine 415 completes.

While illustrative embodiments have been illustrated and described, it will be appreciated that various changes can be made therein without departing from the spirit and scope of the invention.

Claims

1. A computer-implemented method, comprising:

under control of one or more computer systems configured with executable instructions, distributing a query to a plurality of search indexes, each search index corresponding to a category of items; receiving a plurality of search index result sets; determining probabilities of items in the plurality of search index result sets satisfying the query; normalizing at least one of a plurality of distinct internal ranking scales associated with the plurality of search indexes to a common ranking scale; and generating a universal query result set including at least a plurality of the items selected from the plurality of search index result sets in an order utilizing the common ranking scale based at least in part on the probabilities of the items satisfying the query.

2. The computer-implemented method of claim 1, wherein distributing the query to the plurality of search indexes includes:

providing a first copy of the query to a first search index and providing a second copy of the query to a second search index,

wherein receiving the plurality of search index result sets comprises receiving a first search index result set ranked according to a first ranking scale and receiving a second search index result set ranked according to a second ranking scale.

3. The computer-implemented method of claim 1, further comprising:

identifying historical queries having at least one similar search parameter to the query, and wherein generating the universal query result set is based at least in part on recorded actions associated with the historical queries.

4. The computer-implemented method of claim 3, further comprising:

determining an appropriateness score of each search index for the query, each appropriateness score corresponding to an appropriateness of the category of items for the query; and

applying a weight to the appropriateness score of each search index based at least in part on the recorded actions associated with the historical queries,

wherein generating the universal query result set is based at least in part on the probabilities of the items satisfying the query, the appropriateness score of each search index for the query, and the weight applied to each appropriateness score.

5. The computer-implemented method of claim 4, wherein determining the appropriateness score comprises modifying each appropriateness score differently based at least in part on different types of recorded actions associated with the historical queries that were submitted to a corresponding search index.

6. The computer-implemented method of claim 1, wherein each of the plurality of search index result sets is ranked in accordance with a different internal relevance ranking function.

7. The computer-implemented method of claim 6, wherein each of the different internal relevance ranking functions rank items in accordance with one of the plurality of distinct internal ranking scales.

8. A computer system, comprising:

a query controller configured to, at least: receive a query from a searching entity; distribute the query to a plurality of search indexes, each search index corresponding to a category of items; receive a plurality of search index result sets; determine probabilities of items in the plurality of search index result sets satisfying the query; normalize at least one of a plurality of distinct internal ranking scales associated with the plurality of search indexes to a common ranking scale; generate a universal query result set including at least a plurality of the items selected from the plurality of search index result sets in an order utilizing the common ranking scale based at least in part on the probabilities of the items satisfying the query; and provide the universal query result set to the searching entity; and

one or more processors collectively configured to facilitate at least the query controller.

9. The computer system of claim 8, wherein the query is received from a computing device distinct from the computer system comprising the query controller.

10. The computer system of claim 8 wherein the plurality of search indexes reside, at least in part, on at least one computing device distinct from the computer system comprising the query controller.

11. The computer system of claim 8 wherein distributing the query to the plurality of search indexes includes:

providing a first copy of the query to a first search index and providing a second copy of the query to a second search index,

wherein receiving the plurality of search index result sets comprises receiving a first search index result set ranked according to a first ranking scale and receiving a second search index result set ranked according to a second ranking scale.

12. The computer system of claim 8, the query controller further configured to:

identify historical queries having at least one similar search parameter to the query, and wherein generating the universal query result set is based at least in part on recorded actions associated with the historical queries.

13. The computer system of claim 12, the query controller further configured to:

determine an appropriateness score of each search index for the query, each appropriateness score corresponding to an appropriateness of the category of items for the query; and

apply a weight to the appropriateness score of each search index based at least in part on the recorded actions associated with the historical queries,

wherein generating the universal query result set is based at least in part on the probabilities of the items satisfying the query, the appropriateness score of each search index for the query, and the weight applied to each appropriateness score.

14. The computer system of claim 13, wherein determining the appropriateness score comprises modifying each appropriateness score differently based at least in part on different types of recorded actions associated with the historical queries that were submitted to a corresponding search index.

15. A non-transitory computer-readable storage medium storing instructions that, when executed by at least one processor, cause a computing device to:

distribute the query to a plurality of search indexes, each search index corresponding to a category of items;

receive a plurality of search index result sets;

determine probabilities of items in the plurality of search index result sets satisfying the query;

normalize at least one of a plurality of distinct internal ranking scales associated with the plurality of search indexes to a common ranking scale;

generate a universal query result set including at least a plurality of the items selected from the plurality of search index result sets in an order utilizing the common ranking scale based at least in part on the probabilities of the items satisfying the query.

16. The non-transitory computer-readable storage medium of claim 15, wherein distributing the query to the plurality of search indexes includes:

providing a first copy of the query to a first search index and providing a second copy of the query to a second search index,

wherein receiving the plurality of search index result sets comprises receiving a first search index result set ranked according to a first ranking scale and receiving a second search index result set ranked according to a second ranking scale.

17. The non-transitory computer-readable storage medium of claim 15, the instructions further configured to cause the computing device to:

identify historical queries having at least one similar search parameter to the query, and wherein generating the universal query result set is based at least in part on recorded actions associated with the historical queries.

18. The non-transitory computer-readable storage medium of claim 17, the instructions further configured to cause the computing device to:

determine an appropriateness score of each search index for the query, each appropriateness score corresponding to an appropriateness of the category of items for the query; and

apply a weight to the appropriateness score of each search index based at least in part on the recorded actions associated with the historical queries,

wherein generating the universal query result set is based at least in part on the probabilities of the items satisfying the query, the appropriateness score of each search index for the query, and the weight applied to each appropriateness score.

19. The non-transitory computer-readable storage medium of claim 18, wherein determining the appropriateness score comprises modifying each appropriateness score differently based at least in part on different types of recorded actions associated with the historical queries that were submitted to a corresponding search index.

20. The non-transitory computer-readable storage medium of claim 15, wherein each of the plurality of search index result sets is ranked in accordance with a different internal relevance ranking function.