US20160026715A1 - Determining quality of tier assignments - Google Patents

Determining quality of tier assignments Download PDF

Info

Publication number
US20160026715A1
US20160026715A1 US14/875,666 US201514875666A US2016026715A1 US 20160026715 A1 US20160026715 A1 US 20160026715A1 US 201514875666 A US201514875666 A US 201514875666A US 2016026715 A1 US2016026715 A1 US 2016026715A1
Authority
US
United States
Prior art keywords
search engine
search
tier
data storage
based
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Abandoned
Application number
US14/875,666
Inventor
Mikhail Bilenko
Miles Arthur Munson
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Microsoft Technology Licensing LLC
Original Assignee
Microsoft Technology Licensing LLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to US11/964,729 priority Critical patent/US8024285B2/en
Priority to US13/210,797 priority patent/US9177042B2/en
Application filed by Microsoft Technology Licensing LLC filed Critical Microsoft Technology Licensing LLC
Priority to US14/875,666 priority patent/US20160026715A1/en
Assigned to MICROSOFT CORPORATION reassignment MICROSOFT CORPORATION ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: MUNSON, MILES ARTHUR, BILENKO, MIKHAIL
Assigned to MICROSOFT TECHNOLOGY LICENSING, LLC reassignment MICROSOFT TECHNOLOGY LICENSING, LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: MICROSOFT CORPORATION
Publication of US20160026715A1 publication Critical patent/US20160026715A1/en
Application status is Abandoned legal-status Critical

Links

Images

Classifications

    • G06F17/30864
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/951Indexing; Web crawling techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2455Query execution
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/248Presentation of query results
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/31Indexing; Data structures therefor; Storage structures
    • G06F16/316Indexing structures
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/958Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • G06F17/30477
    • G06F17/30554
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network-specific arrangements or communication protocols supporting networked applications
    • H04L67/10Network-specific arrangements or communication protocols supporting networked applications in which an application is distributed across nodes in the network
    • H04L67/1097Network-specific arrangements or communication protocols supporting networked applications in which an application is distributed across nodes in the network for distributed storage of data in a network, e.g. network file system [NFS], transport mechanisms for storage area networks [SAN] or network attached storage [NAS]

Abstract

Technologies pertaining to computing a tiering policy that defines how digital items are desirable stored across a plurality of different storage tiers are described herein. A data repository that comprises data that is indicative of historic user interaction with a search engine is accessed. Subsequently, a tiering policy for digital items that are retrievable by way of the search engine is computed based at least in part upon the data that is indicative of the historic user interaction with the search engine. Retrieval times for digital items in the data storage tiers differ across the data storage tiers.

Description

    CROSS REFERENCE TO RELATED APPLICATIONS
  • This application is a continuation of U.S. patent application Ser. No. 13/210,797, filed on Aug. 16, 2011, which is a continuation of U.S. patent application Ser. No. 11/964,729, filed on Dec. 27, 2007. The entireties of these applications are incorporated herein by reference.
  • BACKGROUND
  • Search engines have enabled users to quickly access information over the Internet. Specifically, a user can submit a query to a search engine and peruse ranked results returned by the search engine. For example, a user can provide a search engine with the query “Spider” and be provided with web pages relating to various arachnids, web pages relating to automobiles, web pages relating to films, web pages related to web crawlers, and other web pages. Search engines may also be used to return images, academic papers, videos, and other information to an issuer of a query.
  • Operation of a search engine may include employment of web crawlers to locate and store a large amount of information (e.g., web pages) that is available on the World Wide Web. For example, web pages or information pertaining thereto may be stored in a search engine index, which is used (in connection with one or more search algorithms) when queries are received.
  • Conventionally, a search engine index is stored in several tiers, wherein different tiers provide different levels of performance. The tiering of the search engine index is analogous to the memory hierarchy used in computer architecture: overall storage capacity of the index is divided between different levels that vary in size, speed, latency, and cost. Higher tiers of the index typically have higher speed but have smaller capacity and higher cost. Accordingly, it is desirable to carefully index web pages to maximize efficiency of the search engine.
  • One manner for tiering web pages that has been used is to select a tier of an index in which to place a web page as a function of the web page's relative importance as determined by some metric, such as a static rank of the web page. Specifically, a number of links to a web page may be used to select a tier of an index in which to locate the web page. The relative importance of the page, however, is not necessarily indicative of whether the page is frequently accessed, and thus may be suboptimal for indexing web pages in a search engine index. Evaluating tier assignment is a difficult problem, however, because it is unclear which metrics capture the quality of a particular allocation of web pages to the tiers.
  • SUMMARY
  • The following is a brief summary of subject matter that is described in greater detail herein. This summary is not intended to be limiting as to the scope of the claims.
  • Various technologies relating to tiering digital items (such as web pages) are described herein. User interaction with a search engine, database management system, or the like can be monitored and data can be collected relating to such user interaction. For example, queries submitted by users, search results (e.g., digital items) provided in response to the queries, and user actions with respect to the search results can be monitored and retained. In a particular example, a toolbar on a browser can be used to collect the user history data. Based at least in part upon the user history data, an indication of quality of a tier assignment for searchable digital items can be generated, wherein a tier assignment indicates to which of several tiers searchable digital items are assigned. The indication of quality of the tier may be a value that accords to a defined tier assignment quality metric, which is described in detail herein.
  • In an example, the indication of quality may be determined by ascertaining several parameters. For instance, the indication of quality of the tier assignment may be based at least in part upon weights that are assigned to observed queries. In an example, the weights may be indicative of relative importance of the queries, and may be based at least in part upon frequency of issuance of the queries. In another example, the indication of quality of the tier assignment may be based at least in part upon a probability that, for a particular query and a determined system load (e.g., how busy a system is when the query is received), retrieval of digital items will end in a specified tier. The probability may be determined for multiple tiers. In yet another example, the indication of quality of the tier assignment may be based at least in part upon a measure of search result quality obtained when retrieval ends in a particular tier. Normalized Discounted Cumulative Gain, Mean Average Precision, Q-measure, or other suitable mechanisms for measuring information retrieval loss or search result quality may be used in connection with determining the measure of tiering quality.
  • In addition, an improved tier assignment can be generated based at least in part upon the indication of quality of tier assignment and/or the user history data. For example, the indication of quality of tier assignment may conform to a defined tier assignment quality metric, and an improved tier assignment may be optimized or substantially optimized with respect to the metric. Furthermore, a tiering policy can be updated based at least in part upon the improved tier assignment. A tiering policy is a policy that is used to assign digital items to tiers, and can take into account various features that correspond to a digital item, such as a number of times the digital item has been accessed by a user, size of the digital item, and the like. The tiering policy can be updated through the use of machine learning techniques, for example.
  • Other aspects of the present application will be appreciated upon reading and understanding the attached figures and description.
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • FIG. 1 is a functional block diagram of an example system that facilitates determining an indication of quality of a tier assignment.
  • FIG. 2 is a functional block diagram of an example component that generates an indication of quality of a tier assignment.
  • FIG. 3 is a functional block diagram of an example system that facilitates generating an improved tier assignment.
  • FIG. 4 is a functional block diagram of an example system that facilitates generating an improved tier assignment.
  • FIG. 5 is a flow diagram that illustrates an example methodology for generating an indication of quality of a tier assignment.
  • FIG. 6 is a flow diagram that illustrates an example methodology for generating an indication of quality of a tier assignment.
  • FIG. 7 is a flow diagram that illustrates an example methodology for outputting a tier assignment that is optimized or substantially optimized with respect to a tier assignment quality metric.
  • FIG. 8 is a flow diagram that illustrates an example methodology for updating a tiering policy.
  • FIG. 9 is an example computing system.
  • DETAILED DESCRIPTION
  • Various technologies pertaining to determining quality of a tier assignment, generating an improved tier assignment, and automatically updating a tiering policy will now be described with reference to the drawings, where like reference numerals represent like elements throughout. In addition, several functional block diagrams of example systems are illustrated and described herein for purposes of explanation; however, it is to be understood that functionality that is described as being carried out by certain system components may be performed by multiple components. Similarly, for instance, a single component may be configured to perform functionality that is described as being carried out by multiple components.
  • With reference to FIG. 1, an example system 100 that facilitates outputting an indication of quality of a tier assignment with respect to a tiered storage system (not shown) is illustrated. Pursuant to an example, a tiered storage system may be a search engine index with multiple tiers, wherein a first (highest) tier may be more costly and have a relatively small amount of storage space, but retrieval time for digital items retrieved from the first tier may be less than retrieval times for digital items retrieved from other tiers (lower tiers). A second tier may be less expensive and have more storage space than the first tier, but retrieval time may be greater when compared to retrieval time corresponding to the first tier. In another example, the tiered storage system may include tiers of storage used in connection with a database management system. For example, a server used in a database management system may have a hard drive, random access memory, and high-speed random access memory, which can each be a tier.
  • The system 100 includes a data store 102 that comprises user history data 104. The user history data 104 may include, for example, queries issued by users, search results provided to the users in response to the queries, search results selected by users in response to being provided with the search results, and/or other suitable information. In an example, the user history data 104 can be accumulated by monitoring user interaction with respect to a search engine. For instance, a toolbar plugin may be installed in a browser, and queries entered into the browser may be collected by the toolbar plugin, as well as search results returned in response to the queries, user selection of particular search results, and the sequence of pages viewed by the user after submitting the query.
  • A receiver component 106 receives a subset of the user history data 104. A quality indicator component 108 is in communication with the receiver component 106 and receives the subset of user history data 104 from the receiver component 106. The quality indicator component 108 can generate an indication 110 of quality of a tier assignment, wherein the tier assignment indicates where digital items are to be assigned in a tiered storage system. For instance, the indication of quality may conform to a tier assignment quality metric, which is described in detail below. In addition, operation of the quality indicator component 108 is described in greater detail below.
  • Now referring to FIG. 2, an example of the quality indicator component 108 is illustrated. The quality indicator component 108 includes a weight determiner component 202, a load determiner component 204, a tier determiner component 206, and a utility determiner component 208. The weight determiner component 202 determines a weight that is assigned to each query used by the quality indicator component 108 to generate an indication of quality of a tier assignment corresponding to a tiered storage system. In an example, the weight determined by the weight determiner component 202 may be based at least in part upon frequency of issuance of the query (as ascertained from query logs, for example).
  • The load determiner component 204 determines the system load observed when a particular query was executed by a search component (e.g., search engine, database system, . . . ). The system load may be based at least in part upon a number of queries processed by the search component while the particular query was processed, a number of processing cycles dedicated to retrieving search results while the particular query was executed, or how “busy” the search component was in general.
  • The tier determiner component 206 can determine a probability that a certain tier will be the last tier searched over for digital items (with respect to the particular query) under the system load determined by the load determiner component 204. Generally, when a query is entered into a search component (e.g., a search engine), retrieval is first performed in higher tiers that are typically smaller but have faster access and retrieval times when compared to lower tiers. Depending on the number and quality of results obtained in the higher tiers as well as a current system load, retrieval may or may not be performed in lower tiers. Accordingly, as noted above, the tier determiner component 206 can determine a probability that a certain tier will be the last tier searched over for digital items (with respect to the particular query and under the determined system load). The probability can be determined for each tier in a tiered storage system.
  • The utility determiner component 208 determines an indication of search result quality (with respect to a particular query) when retrieval ends in a certain tier, wherein the indication of search result quality can be computed using any suitable metric. In an example, Normalized Discounted Cumulative Gain (NDCG) can be used to determine the indication of search result quality. In another example, Mean Average Precision (MAP) can be used to determine the indication of search result quality. In yet another example, Q-measure can be used to determine the indication of search result quality. Accordingly, it can be discerned that the utility determiner component 208 can utilize any suitable mechanisms/metrics to determine an indication of search result quality with respect to the particular query when retrieval ends in the certain tier.
  • The weight determined by the weight determiner component 202, the system load determined by the load determiner component 204, the probability determined by the tier determiner component 206, and the indication of search result quality determined by the utility determiner component 208 may be used by the quality indicator component 108 to determine an indication of quality of a tier assignment.
  • Pursuant to an example, the following algorithm can be used to define a metric of tier assignment quality, and can be employed by the quality indicator component 108 to determine an indication of quality of a tier assignment:
  • TQ ( T ( D ) , L ) = q Q w ( q ) t = 1 k P ( t | q , T ( D ) , L ) × Utility ( t , q , T ( D ) ) , ( 1 )
  • where D={d1, . . . , d|D|} is the set of all digital items (di) that are to be stored in k tiers T1, . . . Tk that have corresponding capacities |T1|, . . . , |Tk|; t(di) is the tier assignment for each item in the set of digital items D, where t(di) can have values l, . . . , k; T(D)={t(di), . . . , t(di)} is the overall set of tier assignments; TQ(T(D),L) is a measure of tier assignment quality for a current system load L; Q is a set of all possible queries; w(q) is a weight (e.g., relative importance) of a query q; P(t|q,T(D),L) is the probability that the t-th tier will be the lowest tier visited during retrieval under the current system load L; and Utility(t,q,T(D)) is a measure of search result quality obtained when retrieval ends in the t-th tier. Algorithm (1) thus computes an expectation of overall tier assignment quality over all possible queries for the given tier assignment over the probability distribution of ending retrieval in each tier.
  • It can be discerned that the number of all possible queries, however, is infinite. Accordingly, a set of observed queries Q′ may be used by the quality indicator component 108 as an approximation of the distribution of all possible queries. In an example, these observed queries Q′ can be randomly selected from a data repository that includes multiple observed queries (e.g., the user history data 104), where the probability of selecting any query q∈Q′ can be computed as the likelihood of selecting a random query received by a search component (e.g., search engine, database management system, . . . ). In another example, the set of observed queries Q′ may be selected such that they are representative of all possible queries. For instance, the queries Q′ may be selected such that a number of queries that have a certain length (as measured in words, characters, or the like) do not exceed a threshold. In addition, queries that are directed at different subject matter can be selected. In yet another example, the queries Q′ may be selected based upon an amount of user data that is associated with such queries. For instance, the queries Q′ may be limited to queries that have sequential user data associated therewith, such as user clicks on one or more search results and/or advertisements that are provided in response to the queries. It is to be understood that any suitable manner for selecting a subset of observed queries is contemplated and intended to fall under the scope of the hereto-appended claims.
  • For every selected query q in Q′, a relevant result set R(q)={dz,l, . . . , dq,M} can be constructed by the quality indicator component 108 that includes no more than M items, wherein the items may be partially ordered from most relevant to least relevant. In an example, the result set may incorporate digital items that are frequently selected/visited by users following submission of the query to a search component, where frequency of selection/visitation can be combined with the time that users spent viewing the digital items; and/or digital items returned by a search component as relevant results for the query across all tiers of a tiered storage system.
  • Using the queries Q′ and corresponding result sets, the following algorithm can be used to define a metric of tier assignment quality, and can be employed by the quality indicator component 108 to determine an indication of quality of a tier assignment:
  • TQ ( T ( D ) , L , Q ) = q Q w ( q ) t = 1 k P ( t | q , T ( D ) , L ) × Utility ( t , R ( q ) , T ( D ) ) , ( 2 )
  • where TQ(D),L,Q′) is a measure of tier assignment quality for a current system load L with respect to the set of queries Q′; and Utility(t,R(q),T(D)) is a measure of search result quality obtained when retrieval ends in the t-th tier.
  • As noted above, the quality indicator component 108 can determine an indication of quality of a tier assignment. More particularly, the weight determiner component 202 can determine weights (w) for each query in the set of queries Q′. The load determiner component 204 can determine the system load L present for each query in the set of queries Q′. The tier determiner component 206 can determine P(t|q,T(D),L), and the utility determiner component 208 can determine Utility(t, R(q),T(D)). In an example, utility determiner component 208 can use normalized discounted cumulative gain (NDCG) to determine Utility(t,R(q),T(D)). The utility determiner component 208 can employ other mechanisms to measure utility; examples include Mean Average Precision (MAP), and Q-measure. These examples are not intended to be limiting, as other mechanisms to measure utility may be employed and are contemplated.
  • In a particular example, the utility determiner component 208 can utilize the following algorithm to determine the measure of search result quality when retrieval ends in the t-th tier, wherein the algorithm is a modification of NDCG:
  • Utility NDCG ( t , R ( q ) , T ( D ) ) = N d R t ( q ) 2 rel ( d ) - 1 log ( rank ( d ) + 1 ) ( 3 )
  • where N is a normalization factor, Rt(q) is the ordered subset of digital items in R(q) stored in tiers 1 through t, rel(d) is a relevance score for digital item d, and rank(d) is the rank position in Rt(q)of the digital item. Note that rank(d) can depend on t if more relevant digital items reside in lower (deeper) tiers; these may not be retrieved if retrieval does not go beyond tier t. As noted above, using a modification of NDCG is but one possible measure of search result quality for a particular query given current tier assignments, and other measures can be utilized, such as the proportion of relevant results retrieved, etc.
  • As can be discerned from the above, the user history data 104 (FIG. 1) can be used to construct the set of queries Q′ and the corresponding result set R(q) that can be employed to evaluate a tier assignment. P(t|q,T(D),L) can be instantiated for a particular system to reflect a tiering policy used in tiered storage system for forwarding queries to the t-th tier under an observed load L, provided a current tier assignment T(D). Then, given alternative tier assignments (e.g., T1(D) and T2(D)), a preferred assignment can be selected by computing TQ. Additionally, the quality indicator component 108 can use TQ to investigate the expected quality of search results under varying loads (and thus the quality of tier assignments under different loads), as well as for different instantiations of a tiering policy used for forwarding queries to different tiers, as described in detail below.
  • Referring now to FIG. 3, an example system 300 that facilitates automatically updating a tier assignment with respect to a tiered storage system is illustrated. The system 300 includes a tiered storage system 302 that may include a plurality of tiers, wherein each of the tiers may be used to store one or more digital items, such as web pages, images, documents, and/or the like. A search component 304 performs searches for digital items stored in the tiered storage system 302 based at least in part upon received queries. For example, the search component 304 can be a search engine that is configured to search through a tiered search index in response to receiving a query. In another example, the search component 304 may be a portion of a database management system used to search tiers of storage (e.g., memory, hard drive, . . . ) in response to receipt of a query. In yet another example, the search component 304 may be a desktop search module used to search items on a computer. Other search components are also contemplated.
  • The data store 102 retains user history data 104 that can be received from the search component 304. For example, queries provided to the search component 304, user actions upon being provided with search results, and sets of search results provided to the user in response to the query can be stored in the user history data 104. The receiver component 106 receives a subset of the user history data 104. As described above, the quality indicator component 108 can generate the indication 110 of quality of a tier assignment. In an example, the indication 110 may be stored in a computer readable medium upon being generated by the quality indicator component 108.
  • An update component 306 can receive the indication 110 and an output an improved tier assignment 308 based at least in part upon the indication 110. For example, the update component 306 can receive other possible tier assignments and corresponding indications of quality and select a tier assignment that corresponds to a highest indication of quality. For example, the update component 306 may use heuristics to determine an optimal or substantially optimal tier assignment (with respect to a defined tier assignment quality metric). In another example, machine learning techniques, which will be described in greater detail below, can be utilized by the update component 306 to output the improved tier assignment 308. Digital items 310 may then be assigned to the tiered storage system 302 based at least in part upon the improved tiering assignment 308.
  • With more detail relating to the update component 306, the indication 110 of quality of an initial tier assignment can provide a basis for developing algorithms/techniques for identifying improved tier assignments for digital items. Given a space of possible tier assignments T={T(1)(D), . . . , T(N)(D)}, identifying a tier assignment T*(D) that has an optimal or substantially optimal indication of tier quality as output by algorithm (2) can be defined as follows:
  • T * ( D ) = argmax T i ( D ) T TQ ( T ( i ) ( D ) , L , Q ) . ( 4 )
  • The set of possible tier assignments T can be defined as a set of alternative assignments or groups of assignments that are parameterized by some variables, such as parameters of a static ranking scheme. Then the update component 306 can use machine learning techniques to search a set of alternative assignments to identity one of such assignments as being optimal or substantially optimal. For example, the update component 306 may use a neural network, a regression tree, a Bayesian network, or any other suitable machine learning technique to determine a tiering assignment that optimizes or substantially optimizes the indication 110.
  • Furthermore, update component 306 can determine a tiering policy 312 that is used to assign the digital items 310 to particular tiers in the tiered storage system 302 based at least in part upon the improved tier assignment 308 and/or a subset of the user history data 104. A tiering policy may be used to determine which tiers of the tiered storage system 302 to use when storing digital items. For instance, the tiering policy 312 may take into account various features of searchable digital items that may be returned in response to one or more queries. Such features may include a static ranking derived from a link structure (e.g., page rank of a digital item), a rank of a domain that includes the digital item, a popularity of the digital item among search engine results, a number of words in a digital item, color spectrums of images in a digital item, etc. Each of these features may be parameterized by the update component 306. In other words, the features may be assigned weights that are used by the tiering policy 312 to assign a corresponding digital item to a tier of the tiered storage system 302. The update component 306 can use machine learning techniques to learn the weights that are to be assigned to the features, and the tiering policy may be used to assign digital items to tiers of the tiered storage system 302.
  • With reference now to FIG. 4, an example system 400 that facilitates updating a tier assignment based on multiple possible tier assignments is illustrated. The system 400 includes the quality indicator component 108 that can generate an indication of quality of tier assignments. More specifically, the quality indicator component 108 can generate indications of quality of a first tier assignment 402 through an Nth tier assignment 404 based at least in part the user history data 104. The update component 306 can receive the indications of quality (which may be values that correspond to a defined tier assignment quality metric) and combine several different tier assignments in such a manner that a resulting improved tier assignment 406 has a higher quality (as determined by the quality indicator component 108) than any of the individual tier assignments. The update component 306 can combine different tier assignments based at least in part upon the indications of quality corresponding to the tier assignments 402-404 and/or a subset of the user history data 104.
  • In more detail, combining tier assignments may be a particular instantiation of algorithm (4), where the set T of possible assignments may be a set of possible combinations of individual tier assignments. The set of possible combinations can be parameterized by some variables, such as parameters of a static ranking scheme. The update component 306 can use machine learning techniques to determine a combination of individual tier assignments that is optimal or substantially optimal with respect to a defined tier assignment quality metric. In addition, as discussed above, the update component 306 can generate or update the tiering policy 312 that is used to assign digital items to tiers of a tiered storage system based at least in part upon the improved tier assignment 406.
  • With reference now to FIGS. 5-8, various example methodologies are illustrated and described. While the methodologies are described as being a series of acts that are performed in a sequence, it is to be understood that the methodologies are not limited by the order of the sequence. For instance, some acts may occur in a different order than what is described herein. In addition, an act may occur concurrently with another act. Furthermore, in some instances, not all acts may be required to implement a methodology described herein.
  • Moreover, the acts described herein may be computer-executable instructions that can be implemented by one or more processors and/or stored on a computer-readable medium or media. The computer-executable instructions may include a routine, a sub-routine, programs, a thread of execution, and/or the like. In addition, tier assignments in a search engine and/or database management system can be determined based at least in part upon the methodologies described herein. Still further, results of acts of the methodologies may be stored in a computer-readable medium, displayed on a display device, and/or the like.
  • Referring specifically to FIG. 5, an example methodology 500 for determining an indication of quality of a tier assignment is illustrated. The methodology 500 starts at 502, and at 504 user history data is received. For example, the user history data can include queries that were issued by users, search results provided to the users in response to the queries, user selections of the search results and the sequence of pages viewed by the user after issuing the query. The user history data may also include labeled data, wherein relevance of search results to queries is explicitly defined by users.
  • At 506, an indication of quality of a tier assignment is generated based at least in part upon a subset of the user history data. The methodology 500 completes at 508.
  • Turning now to FIG. 6, a methodology 600 that facilitates determining an indication of quality of a tier assignment with respect to a tiered storage system is illustrated. The methodology 600 starts at 602, and at 604 a weight assigned to a query is determined. For example, the weight may depend on frequency of issuance of the query. In another example, a user or users may explicitly assign a weight to the query to indicate a relative importance of the query.
  • At 606, a system load background for the query is determined. As noted above, the system load may be related to a number of queries that are being processed by a search component, such as a search engine or database management system, at a time that the query is processed.
  • At 608, a probability that a certain tier will be a lowest tier visited when the search engine is under the system load is determined. For example, this probability can be determined for each tier used to store searchable digital items.
  • At 610, an indication of quality of a tier assignment is determined, where the tier assignment is used to store digital items that correspond to the query in a tiered storage system. The indication of quality is determined based at least in part upon the weight, the system load, and the determined tier probability. In an example, the determined indication of quality may be stored, at least temporarily, in a computer-readable medium. The methodology 600 ends at 612.
  • Referring now to FIG. 7, a methodology 700 for determining an optimal or substantially optimal tier assignment (e.g., optimized or substantially optimized for a defined tier assignment quality metric) is illustrated. The methodology 700 starts at 702, and at 704 a plurality of different tier assignments are received. At 706, user history data is received. As noted above, the user history data may include queries, search results provided in response to the queries, and/or user selections of search results provided in response to the queries.
  • At 708, indications of quality are determined for a subset of the plurality of different tier assignments. At 710, tier assignments are combined such that the resulting combination has a higher indication of quality than any individual tier assignment. The methodology 700 ends at 712.
  • With reference now to FIG. 8, a methodology 800 that facilitates updating a tiering policy is illustrated. In an example, a search engine that uses a tiering policy to assign digital items to tiers of a search engine index may use acts of the methodology 800 to update the tiering policy. The methodology 800 begins at 802, and at 804 user history data is received. At 806, an indication of quality of a tier assignment is determined. At 808, an improved tier assignment is determined based at least in part upon the user history data and/or the indication of quality determined at 806. At 810, a tiering policy is updated based at least in part upon the user history data and the improved tier assignment. For instance, the improved tier assignment may contemplate digital items that are related to the user history search data, and the tiering policy may be used to assign digital items that were not contemplated in the improved tier assignment to particular tiers. The methodology 800 ends at 812.
  • Now referring to FIG. 9, a high-level illustration of an example computing device 900 that can be used in accordance with the systems and methodologies disclosed herein is illustrated. For instance, the computing device 900 may be used in a search engine system. In another example, the computing device 900 may be used in a database management system. The computing device 900 may be a server, or may be employed in devices that are conventionally thought of as client devices, such as personal computers, personal digital assistants, and the like. The computing device 900 includes at least one processor 902 that executes instructions that are stored in a memory 904. The instructions may be, for instance, instructions for implementing functionality described as being carried out by one or more components discussed above or instructions for implementing one or more of the methods described above. The processor 902 may access the memory by way of a system bus 906. In addition to storing executable instructions, the memory 904 may also store digital items, at least a portion of a tier assignment, indications of quality of one or more tier assignments, etc.
  • The computing device 900 additionally includes a data store 908 that is accessible by the processor 902 by way of the system bus 906. The data store 908 may include executable instructions, one or more tier assignments, indications of quality of tier assignments, user history data, labeled data, etc. The computing device 900 also includes an input interface 910 that allows external devices to communicate with the computing device 900. For instance, the input interface 910 may be used to receive queries from a user by way of a network. The computing device 900 also includes an output interface 912 that interfaces the computing device 900 with one or more external devices. For example, the computing device 900 may display search results by way of the output interface 912.
  • Additionally, while illustrated as a single system, it is to be understood that the computing device 900 may be a distributed system. Thus, for instance, several devices may be in communication by way of a network connection and may collectively perform tasks described as being performed by the computing device 900.
  • As used herein, the terms “component” and “system” are intended to encompass hardware, software, or a combination of hardware and software. Thus, for example, a system or component may be a process, a process executing on a processor, or a processor. Additionally, a component or system may be localized on a single device or distributed across several devices.
  • It is noted that several examples have been provided for purposes of explanation. These examples are not to be construed as limiting the hereto-appended claims. Additionally, it may be recognized that the examples provided herein may be permutated while still falling under the scope of the claims.

Claims (20)

What is claimed is:
1. A method comprising:
storing a search engine index across multiple storage tiers in accordance with a tier assignment policy, the multiple storage tiers comprise a first storage tier and a second storage tier, the first storage tier associated with a first access speed and the second storage tier associated with a second access speed that is slower than the first access speed, the tier assignment policy based upon observed user interaction with a search engine; and
executing a search over at least a portion of the search engine index in response to receipt of a query, wherein search results retrieved when executing the search over the search index are based upon the tier assignment policy.
2. The method of claim 1, wherein executing the search over at least the portion of the search engine index comprises executing the search over at least the portion of the search engine index based upon a load of the search engine at a time of receipt of the query.
3. The method of claim 2, further comprising updating the observed user interaction with the search engine with the query and the load of the search engine at the time of receipt of the query.
4. The method of claim 3, further comprising:
updating the tier assignment policy responsive to updating the observed user interaction with the search engine; and
storing the search engine index across the multiple storage tiers in response to updating the tier assignment policy.
5. The method of claim 1, further comprising:
determining the tier assignment policy based upon the observed user interaction with the search engine, the observed user interaction with the search engine comprises:
queries previously submitted to the search engine; and
tiers in the multiple storage tiers searched accessed by the search engine when searches were executed based upon the queries.
6. The method of claim 5, wherein the observed user interaction with the search engine further comprises search results selected by users who issued the queries.
7. The method of claim 1, further comprising determining the tier assignment policy based upon the observed user interaction with the search engine, the observed user interaction with the search engine comprises queries previously submitted to the search engine and respective loads of the search engine when the queries were submitted.
8. The method of claim 1, further comprising determining the tier assignment policy based upon the observed user interaction with the search engine, the observed user interaction with the search engine comprises queries previously submitted to the search engine and respective weights assigned thereto, the weights are indicative of frequency of issuance of the queries.
9. The method of claim 1, further comprising determining the tier assignment policy based upon the observed user interaction with the search engine, the observed user interaction with the search engine comprises values indicative of search result quality that respectively correspond to queries previously received at the search engine.
10. A system comprising:
at least one processor; and
memory that stores instructions that, when executed by the at least one processor, cause the at least one processor to perform acts comprising:
receiving a query; and
executing a search over at least a portion of a search engine index in response to receipt of the query, the search engine index stored over multiple storage tiers of a tiered storage system in accordance with a tier assignment policy, the multiple storage tiers comprise data storage devices having different access speeds, the tier assignment policy defines which data storage devices store respective portions of the search engine index, and wherein executing the search comprises returning search results based upon the query and the tier assignment policy.
11. The system of claim 10, wherein executing the search comprises searching over the portion of the search engine index based upon a computational load of the search engine at a time that the query is received.
12. The system of claim 10, the data storage devices comprise different storage capacities.
13. The system of claim 10, the portion of the search engine index is stored in a first data storage device having a first access speed, and wherein executing the search comprises:
searching over the portion of the search engine index stored in the first data storage device in response to receipt of the query; and
refraining from searching over a second portion of the search engine index stored in a second data storage device based upon a load of the search engine a time that the query is received.
14. The system of claim 10, wherein the tier assignment policy is based upon static rankings assigned to documents indexed by the search engine index.
15. The system of claim 10, wherein the tier assignment policy is based upon numbers of words in documents indexed by the search engine index.
16. The system of claim 10, wherein the tier assignment policy is based upon values that are indicative of qualities of tier assignments of documents that are indexed by the search engine index.
17. A computer-readable data storage device comprising instructions that, when executed by a processor, cause the processor to perform acts comprising:
storing a portion of a search engine index in a data storage device in a tiered data storage system, the data storage device has a first access speed, the tiered data storage system comprises the data storage device and a second data storage device that has a second access speed that is different from the first access speed, the second data storage device comprises a second portion of the search engine index, wherein the portion of the search engine index is stored in the data storage device based upon a tier assignment policy and the second portion of the search engine index is stored in the second data storage device based upon the tier assignment policy; and
executing a search over at least the portion of the search engine index in the data storage device in response to receipt of a query, wherein executing the search comprises returning search results based upon the query and the tier assignment policy.
18. The computer-readable data storage device of claim 17, wherein the tier assignment policy is based upon loads observed at the search engine when queries were received by the search engine.
19. The computer-readable data storage device of claim 17, the acts further comprising updating the tier assignment policy based upon the search results.
20. The computer-readable data storage device of claim 17, wherein executing the search over at least the portion of the search engine index comprises searching over the portion of the search engine index in the data storage device while refraining from searching over the second portion of the search engine index in the second data storage device.
US14/875,666 2007-12-27 2015-10-05 Determining quality of tier assignments Abandoned US20160026715A1 (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US11/964,729 US8024285B2 (en) 2007-12-27 2007-12-27 Determining quality of tier assignments
US13/210,797 US9177042B2 (en) 2007-12-27 2011-08-16 Determining quality of tier assignments
US14/875,666 US20160026715A1 (en) 2007-12-27 2015-10-05 Determining quality of tier assignments

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US14/875,666 US20160026715A1 (en) 2007-12-27 2015-10-05 Determining quality of tier assignments

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
US13/210,797 Continuation US9177042B2 (en) 2007-12-27 2011-08-16 Determining quality of tier assignments

Publications (1)

Publication Number Publication Date
US20160026715A1 true US20160026715A1 (en) 2016-01-28

Family

ID=40799712

Family Applications (3)

Application Number Title Priority Date Filing Date
US11/964,729 Active 2030-06-23 US8024285B2 (en) 2007-12-27 2007-12-27 Determining quality of tier assignments
US13/210,797 Active US9177042B2 (en) 2007-12-27 2011-08-16 Determining quality of tier assignments
US14/875,666 Abandoned US20160026715A1 (en) 2007-12-27 2015-10-05 Determining quality of tier assignments

Family Applications Before (2)

Application Number Title Priority Date Filing Date
US11/964,729 Active 2030-06-23 US8024285B2 (en) 2007-12-27 2007-12-27 Determining quality of tier assignments
US13/210,797 Active US9177042B2 (en) 2007-12-27 2011-08-16 Determining quality of tier assignments

Country Status (5)

Country Link
US (3) US8024285B2 (en)
EP (1) EP2248055A4 (en)
JP (1) JP5372955B2 (en)
CN (1) CN101911061B (en)
WO (1) WO2009085410A1 (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090254523A1 (en) * 2008-04-04 2009-10-08 Yahoo! Inc. Hybrid term and document-based indexing for search query resolution
US8615477B2 (en) * 2010-06-08 2013-12-24 Microsoft Corporation Monitoring relationships between digital items on a computing apparatus
US8463036B1 (en) * 2010-09-30 2013-06-11 A9.Com, Inc. Shape-based search of a collection of content
US8700583B1 (en) * 2012-07-24 2014-04-15 Google Inc. Dynamic tiermaps for large online databases
US9501506B1 (en) 2013-03-15 2016-11-22 Google Inc. Indexing system
US9483568B1 (en) 2013-06-05 2016-11-01 Google Inc. Indexing system
US10146872B2 (en) * 2014-07-16 2018-12-04 Excalibur Ip, Llc Method and system for predicting search results quality in vertical ranking
US10140299B2 (en) * 2014-12-31 2018-11-27 Rovi Guides, Inc. Systems and methods for enhancing search results by way of updating search indices

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3662400A (en) * 1970-04-28 1972-05-09 Hinderstein & Silber Subsidiary document identification system
US4195342A (en) * 1977-12-22 1980-03-25 Honeywell Information Systems Inc. Multi-configurable cache store system
US6763351B1 (en) * 2001-06-18 2004-07-13 Siebel Systems, Inc. Method, apparatus, and system for attaching search results
US20070027700A1 (en) * 2005-07-29 2007-02-01 Sivajini Ahamparam System and method for global informaiton delivery management through a reporting hiearachy
US20070043970A1 (en) * 2005-08-22 2007-02-22 Ethan Solomita Approach for managing interrupt load distribution
US20070214133A1 (en) * 2004-06-23 2007-09-13 Edo Liberty Methods for filtering data and filling in missing data using nonlinear inference
US20070244857A1 (en) * 2006-04-17 2007-10-18 Gilbert Yu Generating an index for a network search engine
US20070294615A1 (en) * 2006-05-30 2007-12-20 Microsoft Corporation Personalizing a search results page based on search history

Family Cites Families (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0981630A (en) 1995-09-13 1997-03-28 Toshiba Corp Information providing system with evaluation function
US6272507B1 (en) * 1997-04-09 2001-08-07 Xerox Corporation System for ranking search results from a collection of documents using spreading activation techniques
US6138118A (en) * 1998-07-30 2000-10-24 Telcordia Technologies, Inc. Method and system for reconciling concurrent streams of transactions in a database
US6845370B2 (en) * 1998-11-12 2005-01-18 Accenture Llp Advanced information gathering for targeted activities
US7181459B2 (en) * 1999-05-04 2007-02-20 Iconfind, Inc. Method of coding, categorizing, and retrieving network pages and sites
US6175830B1 (en) * 1999-05-20 2001-01-16 Evresearch, Ltd. Information management, retrieval and display system and associated method
US6516337B1 (en) * 1999-10-14 2003-02-04 Arcessa, Inc. Sending to a central indexing site meta data or signatures from objects on a computer network
NO313399B1 (en) * 2000-09-14 2002-09-23 Fast Search & Transfer Asa The process feed to search and analysis of information in a computer network
KR20020077502A (en) * 2000-12-22 2002-10-11 코닌클리케 필립스 일렉트로닉스 엔.브이. Meta data category and a method of building an information portal
KR20020069892A (en) 2001-02-28 2002-09-05 주식회사 인터넷과 꿈 A methode of offer for appraisal information by internet site
CA2458908A1 (en) * 2001-08-31 2003-03-13 Arkivio, Inc. Techniques for storing data based upon storage policies
WO2003083643A1 (en) * 2002-03-25 2003-10-09 Morciz Michael Z Accessing deep web information using a search engine
KR100700376B1 (en) 2002-09-11 2007-03-27 한국과학기술정보연구원 Real-time quality measurement method of bibliographic database
US7917483B2 (en) * 2003-04-24 2011-03-29 Affini, Inc. Search engine and method with improved relevancy, scope, and timeliness
KR20040098889A (en) 2003-05-16 2004-11-26 엔에이치엔(주) A method of providing website searching service and a system thereof
US7146353B2 (en) * 2003-07-22 2006-12-05 Hewlett-Packard Development Company, L.P. Resource allocation for multiple applications
US7240064B2 (en) * 2003-11-10 2007-07-03 Overture Services, Inc. Search engine with hierarchically stored indices
JP2005173876A (en) * 2003-12-10 2005-06-30 Hitachi Ltd Cache server
US7734561B2 (en) * 2003-12-15 2010-06-08 International Business Machines Corporation System and method for providing autonomic management of a networked system using an action-centric approach
US20050198007A1 (en) * 2004-03-02 2005-09-08 Tehuti Networks Ltd. Method, system and algorithm for dynamically managing a connection context database
US7779464B2 (en) * 2004-06-14 2010-08-17 Lionic Corporation System security approaches utilizing a hierarchical memory system
US7379947B2 (en) * 2004-07-30 2008-05-27 Microsoft Corporation Efficiently ranking web pages via matrix index manipulation and improved caching
US20060015498A1 (en) * 2004-08-13 2006-01-19 Edgar Sarmiento Search engine
US20060150094A1 (en) * 2004-12-31 2006-07-06 Zakir Patrawala Web companion
US7603343B2 (en) * 2005-02-04 2009-10-13 Microsoft Corporation Quality of web search results using a game
US7873624B2 (en) * 2005-10-21 2011-01-18 Microsoft Corporation Question answering over structured content on the web
US20070244868A1 (en) * 2005-10-31 2007-10-18 Grubb Michael L Internet book marking and search results delivery
US8229897B2 (en) * 2006-02-03 2012-07-24 International Business Machines Corporation Restoring a file to its proper storage tier in an information lifecycle management environment
US20070239747A1 (en) * 2006-03-29 2007-10-11 International Business Machines Corporation Methods, systems, and computer program products for providing read ahead and caching in an information lifecycle management system

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3662400A (en) * 1970-04-28 1972-05-09 Hinderstein & Silber Subsidiary document identification system
US4195342A (en) * 1977-12-22 1980-03-25 Honeywell Information Systems Inc. Multi-configurable cache store system
US6763351B1 (en) * 2001-06-18 2004-07-13 Siebel Systems, Inc. Method, apparatus, and system for attaching search results
US20070214133A1 (en) * 2004-06-23 2007-09-13 Edo Liberty Methods for filtering data and filling in missing data using nonlinear inference
US20070027700A1 (en) * 2005-07-29 2007-02-01 Sivajini Ahamparam System and method for global informaiton delivery management through a reporting hiearachy
US20070043970A1 (en) * 2005-08-22 2007-02-22 Ethan Solomita Approach for managing interrupt load distribution
US20070244857A1 (en) * 2006-04-17 2007-10-18 Gilbert Yu Generating an index for a network search engine
US20070294615A1 (en) * 2006-05-30 2007-12-20 Microsoft Corporation Personalizing a search results page based on search history

Also Published As

Publication number Publication date
US9177042B2 (en) 2015-11-03
WO2009085410A1 (en) 2009-07-09
CN101911061B (en) 2013-04-10
US8024285B2 (en) 2011-09-20
JP2011508340A (en) 2011-03-10
CN101911061A (en) 2010-12-08
EP2248055A4 (en) 2012-01-11
JP5372955B2 (en) 2013-12-18
EP2248055A1 (en) 2010-11-10
US20110302146A1 (en) 2011-12-08
US20090171867A1 (en) 2009-07-02

Similar Documents

Publication Publication Date Title
Liu et al. Distributed nonnegative matrix factorization for web-scale dyadic data analysis on mapreduce
Webber et al. A similarity measure for indefinite rankings
Carmel et al. Personalized social search based on the user's social network
US8219593B2 (en) System and method for measuring the quality of document sets
AU2011202345B2 (en) Methods and systems for improving a search ranking using related queries
US7162522B2 (en) User profile classification by web usage analysis
Rafiei et al. Diversifying web search results
JP4750456B2 (en) Content propagation for enhanced document retrieval
US6728704B2 (en) Method and apparatus for merging result lists from multiple search engines
Sparck Jones Search term relevance weighting given little relevance information
Fuxman et al. Using the wisdom of the crowds for keyword generation
JP5572596B2 (en) Personalize the ordering of place content in search results
CN102725759B (en) Semantic directory for search results
US7647314B2 (en) System and method for indexing web content using click-through features
US8572074B2 (en) Identifying task groups for organizing search results
KR101311050B1 (en) Ranking functions using document usage statistics
US7716202B2 (en) Determining a weighted relevance value for each search result based on the estimated relevance value when an actual relevance value was not received for the search result from one of the plurality of search engines
US7428538B2 (en) Retrieval of structured documents
US7827181B2 (en) Click distance determination
US8166032B2 (en) System and method for sentiment-based text classification and relevancy ranking
US8001121B2 (en) Training a ranking function using propagated document relevance
Kolda et al. Higher-order web link analysis using multilinear algebra
US7856446B2 (en) Method and apparatus for determining usefulness of a digital asset
US20110131160A1 (en) Method and System for Generating A Linear Machine Learning Model for Predicting Online User Input Actions
AU2006279520B2 (en) Ranking functions using a biased click distance of a document on a network

Legal Events

Date Code Title Description
AS Assignment

Owner name: MICROSOFT TECHNOLOGY LICENSING, LLC, WASHINGTON

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:MICROSOFT CORPORATION;REEL/FRAME:036738/0352

Effective date: 20141014

Owner name: MICROSOFT CORPORATION, WASHINGTON

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:BILENKO, MIKHAIL;MUNSON, MILES ARTHUR;SIGNING DATES FROM 20071221 TO 20071225;REEL/FRAME:036738/0136

STCB Information on status: application discontinuation

Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION