KR20240015280A - Systems and methods for search query processing using trend analysis - Google Patents

Abstract

Systems and methods of processing search requests are provided to the queries received to provide a more accurate understanding of the information being sought.
Includes analysis. In one embodiment, a query is a unit that may contain one or more words or tokens of the query.
is parsed as , and the units are related to a concept network. Classify queries into subsets according to criteria of interest,
Trend analysis is achieved by comparing concept networks for different subsets. Quotations received continuously
Trend information can be used to improve the automatic discovery agent's response to information.

Description

Search query processing system and method using trend analysis {SYSTEMS AND METHODS FOR SEARCH QUERY PROCESSING USING TREND ANALYSIS}

This application relates to the following commonly assigned pending U.S. patent applications: “[0001] Universal search interface system and method

Preliminary Application No. 60/460,222, filed April 4, 2003, entitled "Act"; and "Using Clustering of Units

Preliminary Application No. 60/510,220, filed October 9, 2003, entitled “Search Processing System and Method.”

Each disclosure is incorporated herein by reference for all purposes. copy

The original invention relates to a search query processing system and method using trend analysis.

[0002] The emergence of the Internet and the large number of web pages and media content available to users through the World Wide Web (web)

There is a need to provide users with a concise approach to filtering and obtaining desired information from the Web.

I won. Search systems and processes have been developed to meet the needs of users seeking desired information.

all. Examples of these technologies can be accessed through Yahoo!, Google, and other sites. Normally used

You enter a query, and the search process responds with one or more links (if searching the web), documents, and/or references (if other search terms) related to your query. The links you answer are closely related to what the user was actually looking for

It could be, or it could be completely unrelated. The "relevance" of a result to a query depends in part on the search used.

It may be a function of the actual query entered as well as the robustness of the system (the underlying collection system).

Relevance may be determined subjectively by the user, or objectively by what the user is looking for.

It may be possible.

The query entered by the user typically consists of one or more words. For example, [0003] query would be "hawaii", "new

york city", "new york city law enforcement". Likewise, queries are all complete in the human brain.

It is not. In other words, humans don't naturally think about queries. These are query search engines or lookups

It is an artificial structure partly imposed by the needs of the library catalog. Humans are just single words

I don't naturally think about it either. What humans think is a natural concept. For example, "hawaii" and

"new york city" are quite different queries with regard to length measured in words, but they have one important thing:

They share features, that is, they each constitute a concept. But "new york city law enforcement"

is different because it constitutes two separate concepts: "new york city" and "law enforcement."

[0004] Additionally, humans think about logical relationships between concepts. For example, the police are important law enforcement agencies.

Therefore, “law enforcement” and “police” are related concepts, and users who type either of these concepts are

You may also be interested in sites related to other concepts, even if they do not contain the specific words or phrases you typed.

there is. As a result of this thinking pattern, humans naturally do not simplify long, variable sequences of single words.

Queries are constructed by inputting one or more natural concepts, and queries are generally related concepts known to the user.

does not include all. Additionally, user intent is not necessarily reflected in individual words in the query. yes

For example, "law enforcement" is a concept, but the individual words "law" and "enforcement" are combined and

They do not individually convey the same user intent.

[0005] Current technology at major search providers, such as MSN, Google or any other major search engine site, does not allow humans to

It doesn't understand queries in the same way as generating them. For example, existing search engines generally

It is not the related concept or fundamental natural concept that the user actually intends, but the exact information entered by the user.

Search for a word or phrase. This is probably because search providers recognize user intent and provide optimal search results and controls.

This is probably the most important reason that hinders the provision of content.

[0006] As can be seen, it provides results that are more consistent with the actual concepts that the user is interested in or enhances the user's experience.

Improved search and interface technologies are needed to help with this.

[0007] Embodiments of the present invention include analysis of received queries to provide a more sophisticated understanding of the information being sought.

Provides a system and method for processing search requests. A query can be defined by, for example, one or more words in the query or

It can be parsed into units that can contain tokens. Additional analysis is done on the unit representation of the query,

Detects patterns such as similar combinations of units used in other queries. relating to a similar set of secondary units

Units that arise from a query are grouped into clusters, and are grouped into clusters based on the similarity of the second units related to each.

Weights may be assigned to relationships between cluster members.

[0008] According to one form of the invention, the clustering is repeated for each different subset of queries, according to one or more criteria.

Queries are classified into subsets. In one embodiment, the criteria may be time, the user's personal characteristics, or demographics (e.g.

age, gender, known hobbies or user profiles), geographic classification criteria (e.g. physical location or IP address),

data), or longitudinal criteria that represent the circumstances or user behavior preceding the query (e.g., when the query is entered).

Where the user is on the website) Subset-specific concepts from at least two subsets

By comparing, trend information about units or clusters is generated.

The following detailed description assists in understanding the features and advantages of the present invention.

[0010] Figure 1 is a simplified high-level block diagram of an information retrieval and communication system according to an embodiment of the present invention.
2 is a simplified block diagram of an information retrieval and communication network delivering media content according to an embodiment of the present invention.
It is also a degree.
Figure 3 is a simplified block diagram of a query processing engine according to an embodiment of the present invention.
4 is a flow diagram of a process for performing trend analysis according to an embodiment of the present invention.
5 shows a unit dictionary and associated processing information, including, in some form, a query processing engine, according to an embodiment of the present invention.
This is a simplified block diagram of the system including.
Figure 6 is a graph of trend data showing the frequency of queries as a function of time of day.

1 is a diagram of an information retrieval and communication network 10 including a client system 20 according to an embodiment of the present invention.

Shows the class outline. In a computer network 10, a client system 20 has access to the Internet 40 or another communication network.

It is connected to any number of server systems (501 to 50N), for example, through a LAN or WAN connection. explained here

As described above, the client system 20 is configured to communicate with any server system 501 to 50N according to the present invention, e.g.

It is configured to access, receive, retrieve, and display other information, such as media content and web pages.

[0012] The various components of the system shown in Figure 1 include conventionally known components that do not need to be described in detail here.

Includes For example, client system 20 may include a personal desktop computer, workstation, laptop, PDA, cell phone,

or any WAP-enabled device or other computing device capable of interfacing directly or indirectly to the Internet.

It can be done. Client system 20 typically uses Microsoft Internet Explorer

TM

Browser, Netscape

Navigarot

TM

Browser, Mozilla

TM

Browser, Opera

TM

For browsers, cell phones, PDAs, or other wireless devices

A user of a client system (20), such as a WAP-enabled browser, can access the server system (501-50N) via the Internet (40).

Running a browsing program that allows you to access, process and view information and pages available from

all. Additionally, the client system 20 is typically a server system with a keyboard, mouse, touch screen, pen, etc.

(501-50N) or in connection with the display of pages, forms, and other information provided by

Interact with the graphical user interface (GUI) provided by the browser (on a monitor screen, LCD display, etc.)

and one or more user interface devices 22 for operation. The present invention is a specific global network

It is suitable for use on the Internet, also known as the Internet. However, an intranet instead of or in addition to the Internet;

Another network, such as an extranet, virtual private network (VPN), non-TCP/IP based network, or any LAN or WAN.

It must be understood that the size can be used.

According to one embodiment, the client device 20 and its components are all Intel Pentium

TM

[0013] Processor, AMD

Atholon

TM

Contains computer code that is executed using a central processing unit such as a processor or multiple processors

This is an operator that can be configured using an application. Data and Media Content as Disclosed Herein

The computer code that manipulates and configures the client system 20 to deliver, process, and display preferably

Although it is downloaded and stored on the hard disk, the entire program code or portions of it are stored in memory, better known as ROM or RAM.

It may be stored on other non-volatile memory media or devices, such as compact disk (CD) media, digital versatile disks, etc.

It may be provided on any medium that can store program code, such as (DVD) medium or floppy disk.

Additionally, the entire program code, or portions thereof, may be obtained from a software source, e.g., on a server system (501-50N).

may be transmitted and downloaded from one to the client system 20 via the Internet, any communication medium, and

Any protocol (e.g. TCP/IP, HTTP, HTTPS, Ethernet, or other conventional media and protocols)

may be transmitted over another network connection (e.g., an extranet, VPN, LAN, or other conventional network)

There is also.

[0014] The computer code implementing the forms of the present invention may be code such as C, C++, HTML, XML, Java, JavaScript, or any

It may be another suitable scripting language (e.g., VBScript), run on the client system 20, or

What may be any other suitable programming language that can be compiled to run on the client system 20

It must be recognized as In some embodiments, no code is downloaded to the client system 20, and the code required

code is executed by the server, or code that already exists on the client system 20 is executed.

[0015] Figure 2 illustrates another information retrieval and communication network 110 delivering media content according to an embodiment of the present invention.

Pay it out As shown, network 110 includes a client system 120 and one or more content server systems.

150 and a search server system 160. In network 110, client system 120 connects to Internet 140.

Or, it is communicatively connected to the server systems 150 and 160 through another communication network. As mentioned above,

Client system 120 and its components can be connected to server system 120 via the Internet 140 or other communications network.

(150, 160) and is configured to communicate with other server systems.

[0016] According to one embodiment, a client application (represented as module 125) running on client system 120

The client is configured to communicate with server systems 150, 160 and process and display data received from them.

Includes instructions for controlling the smart system 120 and its components. The client application module 125 is

As described above, it may be provided on any software storage medium such as floppy disk, CD, DVD, etc., but the client

The smart application 125 preferably operates on a remote server system (e.g., server system 150, server system 160, or

is transmitted and downloaded to the client system 120 from a software source, such as another remote server system.

For example, in one form, client application module 125 may include, for example, embedded JavaScript or

Various controls such as TivX controls that manipulate data and render data into various objects, frames, and windows.

It can be provided to the client system 120 through the Internet 140 as an HTML wrapper including.

[0017] Additionally, the client application module 125 is a specialized search module that processes search requests and search result data.

(126), data and media content, such as text and data frames and active windows, such as browser windows and

User interface module 127 for rendering in a dialog box, various applications running on the client 120

Processes data and media content, such as the application interface module 128, which interfaces with and communicates with

Includes various software modules to do this. Application interface module according to aspects of the present invention

Various applications executing on client system 120, which 128 is preferably configured to interface with.

Examples of applications include various email applications, instant messaging (IM) applications, browser applications, and document

Includes management applications and other applications. In addition, the interface module 127 is a client

It may include browsers such as the default browser configured in the system 120 or another browser. which

In an embodiment, client application module 125 is a general purpose application disclosed in preliminary application 60/460,222 cited above.

Provides features of the search interface.

[0018] According to one embodiment, the search server system 160 provides search result data and media control to the client system 120.

Configured to provide content, content server system 150 may be provided by, for example, search server system 160.

Data, such as a web page, and

It is configured to provide media content. In some variations, search server system 160 provides links to content.

and/or returns content as well as or instead of other references.

[0019] The search server system 160 of one embodiment refers to various collection techniques for populating one or more page indexes with, for example, pages, links to pages, data representative of the content of the indexed pages, etc.

do. These collection techniques use manual or semi-automatic classification to classify and sort web pages within a hierarchy.

It includes algorithms and interfaces, as well as automatic web crawlers, spiders, etc. In some form,

Search server system 160 also consists of search-related algorithms for processing and sorting web pages. search

Server system 160 is also preferably configured to record user query activity in the form of a query log file.

The search server system 160 responds to various search requests received from the client system, especially the search module 126.

is configured to provide data. Server systems 150, 160 are part of a single structure, for example Yahoo! Sae

It may be part of a distributed server system, such as that provided to users by

there is. The content server system 150 and the search server system 160 each have at least one server and related database.

database system, and may include multiple servers and associated database systems, or may be grouped into a single block.

Shown, but may also be geographically distributed. For example, all servers in search server system 160 are located in close proximity to each other.

They can be deployed in multiple locations (for example, in a server farm located in a single building or campus) or in locations that are far from each other.

may be distributed (one or more servers located in city A and one or more servers located in city B)

[0021] As used herein, a “server system” typically refers to a server system located locally or globally in one or more geographic locations.

A station contains one or more logically and/or physically connected servers. “Server” is usually a computer

Includes computing systems and related storage systems and database applications such as those known in the art.

[0022] According to one embodiment, search server system 160 accesses the page index(s) and

one or more page indexes that provide search results to users in response to search queries received from

It consists of an algorithm. In some forms, search server system 160 may further interact with client system 120.

It is configured to provide enhanced search query analysis and clustering capabilities in response to search requests. different side

In this example, search server system 160 includes all of the content and functionality of content server system 150.

[0023] In one embodiment, processes executing on search server system 160 may determine the context of the search query and/or search results.

It performs analysis and responds with search results grouped to reflect different situations. Many search terms do not match the intended situation.

It may have different meanings depending on the meaning. For example, if a user runs a search using the term "Java",

The intended situation is not clear. Users may be interested in the Java computer language, Java Island, Indonesia

You may be interested in coffee (colloquially known as java). The present invention is

Favorably analyze the color queries and/or results and group the results into situations for display on the user's computer 120.

get angry For example, in response to the search term “Java,” some embodiments of search server system 160 may

context or word concepts (more so if the situation is recognized): grouped into the computer languages Java, Island Java, and Coffee Java.

Returns search results. The system displays results in sets, with links provided relevant to each situation.

It can be configured to display only the context (with enough information for the user to distinguish between contexts) without any links.

You can also select the situation you want and display related links. Yahoo! network

In the system, for example, each context is a set of links to pages from the search index, associated with a sponsor match, and

A set of contexts is displayed with links, links related to directory matches, and links related to Inside Yahoo! (IY) matches.

It can be rayed.

[0024] In addition to words or phrases with ambiguous meanings such as “Java,” the system of the present invention necessarily provides ambiguous meanings in one embodiment.

It is configured to group results by context rather than by search term. An example is returned for the search term "Hawaii":

This is the result. Although the term "Hawaii" may not be inherently or naturally ambiguous,

The nature of the results returned can be very broad, relating to all sites that discuss or just mention Hawaii.

In order to provide more useful results to users, the system of the present invention preferably ensures that search results are actually relevant.

Organize search results with context by leveraging available knowledge. For example, for Hawaii the system

Returns results in various context groupings, including "Hawaii: Travel", "Hawaii: Climate", "Hawaii: Geography", and "Hawaii: Culture"

can do.

[0025] In some embodiments, a context identifier is stored in association with a page link in the index, so that when a search is performed, the identifier is

Links can be grouped accordingly. A page link may be associated with a number of context identifiers. These identifiers are

Preferably, the link will be automatically associated with the system when the user runs a related search, but with only one identifier.

The above indexes may be changed by the editor team and may be manually associated with links. In this way, multiple searches

The knowledge collected from the system is fed back to the system, making it more important to the user who requested the displayed search results.

Define and redefine context to make it useful.

[0026] In one embodiment, an algorithm on search server system 160 performs concept discovery or concept analysis of search terms to

Provides more meaningful results to users. For example, for the search phrase “New York City,” the user

It is fairly clear that there is interest in sites related to New York City (city or region) as opposed to other cities in the state.

Similarly, for "New York City law enforcement," the user is referring to New York City's law enforcement (e.g., larceny law enforcement).

It is clear that you are interested in sites related to people. However, most search engines just add the word in the search phrase.

Allows you to search using the individual words "New", "York", "City", "law", and "enforcement", regardless of the order in which they appear.

do. Other search engines may attempt to find the longest substring in the search phrase that appears in the index.

all. For example, if the index is not "New York City law enforcement" but "New York", "New York City", and

If included in "New York City law", search engines may use "enforcement" and

You can search using “New York City law.”

[0027] Embodiments of the present invention provide one or more concepts for constructing a search query by advantageously analyzing words in a search phrase.

Identify (unit).

[0028] Figure 3 is a system for performing concept analysis including concept discovery or trend analysis according to an embodiment of the present invention.

This is a block diagram of (300). One or more query log files 302 (or actual queries) are generated by the query engine (also known as a query engine).

is received by the query processing engine 304, and the query processing engine 304 generates the unit dictionary 306 as described later.

all. Query log files (or actual queries) are stored on the Internet or any network connection, such as LAN, WAN, or direct

It may be received from a variety of sources through links, distribution media (e.g., CD, DVD, floppy disk), etc. An example of a source is search server system 160 (FIG. 1), or multiple search servers in a distributed network of search servers.

160, or includes one or more content servers 150. Query log file sources are typically the same organization

or an entity, for example Yahoo! Server-related, but not necessarily server-related. (also known as query log)

Query log files use statistical methods, such as those that can be used in information theory or concepts such as mutual information.

Processed by the query engine 304. In a preferred form, different periods as desired, e.g. hours, weeks,

Daily query logs are used, although logs for liver etc. may be used. Typically, query logs are

Contains the actual query provided by , including the geolocation, timestamp, and

In response to some or all of the query, such as the system's IP address, cookies, client type (e.g., browser type), etc.

It may also contain additional information. The query processing engine 304 processes various query logs and creates units from them.

Create. Related statistics, such as units and frequencies of occurrence, are stored in a database file, here referred to as the unit dictionary 306.

Or it is stored in memory. The unit dictionary 306 provides information on subsequent queries by the search engine, as described later.

Can be used in responses.

In one embodiment, the query processor 304 includes a unit analysis module 310 and a clustering module 312. unit

The analysis module 310 processes the query log 302 to generate units. In a preferred embodiment, the system determines that the search term

Use the order in which they are presented to identify the units that make up the query. The unit can also be a word (e.g. "java")

It may also be a group of units that frequently appear adjacent to each other (e.g., "new york city"). Identification of units is

It is described in detail in Preliminary Application No. 60/460,222 cited above. Units generated by unit analysis module 310

These are stored in the unit dictionary 306 and, in some embodiments, provide statistical information (e.g., frequency of occurrence or one or more

(frequency of occurrence relative to other units of) may also be included. Information stored in the unit dictionary 306 is stored in the search engine.

It can also be used in responses to subsequent queries.

[0030] The clustering module 312 performs further analysis of the queries using the units identified by the unit analysis module 310.

By doing so, it creates a cluster or concept network that indicates similarities between different units. Day

As is generally known, a concept network is a structure in which each concept represents a relationship corresponding to a node,

Relationships between concepts are expressed as lines (or edges) connecting nodes. Edges have different weights

As can be expected, the weight of an edge reflects the strength or accuracy of similarity between concepts (nodes). this invention

According to an embodiment, "signature" units tend to appear in queries together with one or more identical groups of each other.

Clusters are created from units by identifying other units (“members” of the cluster). Member unit is a concept

It can act as a node in a network. Different expressions of connections or relationships between different units or concepts are used.

It may be used, and the term “concept network” as used herein should be understood to include other expressions.

[0031] For example, assume that a number of users are searching for information about their favorite pop music group. usual

As a result, these users can enter the group name (e.g. "Avril Lavigne" or "Celine Dion") and "lyrics", "mp3", and "characters".

Construct queries that include other words that reflect the type of information found, such as "ta tab", "discography", etc.

I do it. The clustering module 312 analyzes these queries to search for “lyrics”, “mp3”, “other tabs”, and “discography”.

Determine whether a group is a "signature" unit for a cluster of members containing the name of another pop music group. cluster

Examples of specific techniques for the creation of clusters that may be implemented in the clustering module 312 are provided in the preliminary application cited above.

It is disclosed in No. 60/510,220. Clustering module 312 advantageously stores data related to member units and signatures.

The included clustering information is stored in the unit dictionary 306. In this way, information available in the unit dictionary 306

Information about the viewing unit and its "neighbors" (i.e., other units that appear in queries with a given unit), as well as similar

It includes relationships with other units that may have neighbors.

[0032] In a preferred embodiment, the query processing engine 304 is provided with real-world knowledge or linguistic information about specific units or queries.

Although not necessarily required, clusters created by clustering module 312 may be associated with concepts' real-world relationships (e.g.

er, tend to reflect group units belonging to a common category). For example, a cluster might be called "New York

City", "San Francisco", and "Chicago", and the signature for that cluster would be "Hotel", "Restaurant", etc.

may also include "nightclubs" and "nightclubs". These clusters include New York, San Francisco, and Chicago, all of which are tourist destinations.

Although it reflects a destination (or city), the query engine 304 (FIG. 3) does not require any prior knowledge of the concept of "destination" (or "city").

There is no need to have either. This conceptual knowledge can sprout automatically from pattern analysis of queries. this specification

When labeling clusters with terms that convey linguistic meaning to humans, this facilitates understanding of the present disclosure.

You must understand that this is a convenience for doing this. In practice, it is used by the query engine 304 or the unit dictionary 306.

It is not necessary for any cluster labeling scheme to have these properties; for example, a cluster label can simply be

It can be a number, a pointer to the cluster's signature, etc.

[0033] A unit can belong to multiple clusters, and for example, an ambiguous unit such as “java” may eventually end up as “computer pro”.

It could be a “gramming” cluster, a “food and beverage” cluster, a “travel” or a “place” cluster. In some embodiments,

There may be units that do not belong to any cluster.

[0034] According to an embodiment of the present invention, the query processing engine 304 may provide information about time, geography, user demographics, (herein referred to as “longitudinal”),

Trend analysis module that finds trends in queries based on various criteria such as user history or situation (referred to as criteria)

(320) Also includes. The trend analysis module 320 performs the functions of the unit analysis module 310 and the clustering module 312.

Leverage this to your advantage to run concept analysis and/or concept discovery on different subsets of your queries, and

Compile. The resulting trend information is advantageously added to the unit dictionary 306.

[0035] The unit dictionary 306 may be implemented in any form, such as a magnetic disk, tape, or compact disk (CD).

It may be stored in any suitable storage medium, including optical storage media, etc. The contents of the unit dictionary 306 are

The units, as well as the statistical data generated by the unit analysis module 310, are determined by the clustering module 312.

Additional information about each unit, such as clustering information generated and trend information generated by the trend analysis module 320.

Includes. Information about units, clusters and/or trends can be retrieved from search engines to respond to subsequent queries.

can be used by

[0036] Figure 4 is a flow diagram of a process 400 that may be executed by an embodiment of the trend analysis module 320. Step 402

In , queries included in the query log are classified into subsets according to selected criteria. Standards are provided by log files

It can be determined based on arbitrary information about the query, and the queries can be classified into an arbitrary number of subsets according to predetermined criteria.

[0037] As an example, if the query log provides information related to the submission time of various queries, a "time" criterion may be used.

there is. Depending on the time criteria, queries can be made to analyze trends in user interest over time, for example week or week.

It can be classified by month. Queries can be based on time of day (e.g. morning, lunch, evening, night), weekdays vs. weekends, or

They can also be classified based on holidays, seasons, etc. Many variations are possible, and as just an example, 8 a.m. and a.m.

Queries received between 10:00 and 10:00 may be collected over a period of weeks, months, or other desired period. In some embodiments,

Lu's time is determined by the local time of the search server. Multiple locations to serve geographically different areas

If a search service is provided on a server, the server's local time approximately represents the user's local time. different

In embodiments, the user's local time may be used.

[0038] As another example, a “geography” criterion may be used if the query log provides information regarding the geographic origin of the query.

Geography can be related to physical geography. For example, what does it do to identify the likely location of the origin of a query?

IP addresses may be used and user profiles about the user making the query may be used to determine where the user resides.

Can be used to narrow down or query different locations or local/country sites related to different locations

Queries can also be categorized according to whether they were received from a mirror site of . Alternatively, geo-referencing the user's IP

It may also refer to "cybergeography", where an address (or a domain determined from an IP address) is treated as a geographic origin.

Physical and cyber geographic criteria may also be combined.

[0039] As a third example, if the query log provides information about the user's demographic characteristics, such as age, gender, etc. "

“Demographic” criteria may also be used. Any number and combination of demographic data may be used to analyze trends.

Demographic criteria can be defined for . For example, the demographic criterion is one “homeowner over 25 years old.”

a group of “female users”, another group of “male users who rent over 35 years of age”, and another group of “users under 25 years of age who own a car” group.

It may include groups, etc.

[0040] As a fourth example, the query log may contain information (hereinafter referred to as “user history” or “user history”) regarding user behavior prior to query registration.

A “longitudinal” or “user history” basis may be used, if it provides information (referred to as “contextual”), for example:

Portal sites (for example, the Yahoo! site at wwwyahoocom) are used in a variety of fields, such as shopping properties, and general web sites.

It includes “properties” such as directories and music properties, and each property allows the user to input a query.

Assume it includes a search interface. About the property the user was visiting when the query was entered

Information can be obtained and used on a “longitudinal” basis. As such, in one embodiment the longitudinal criterion is the “shopping” program.

One subset of queries originating from the user in Putty and one subset of queries originating from the user in the "General Directory" property.

It may contain other query subsets, etc. About a dialog box for users to enter a search query at any time.

having substantially continuous access, etc. (an example of such a dialog box is Preliminary Application No. 60/460,222 cited above)

In another embodiment (disclosed in), user history information is displayed when the user enters a query and

It may also contain the URL or other identifier of the web page it was on. The longitudinal criteria are URLs, parts of them (e.g.

main name), or information about the content of the page (for example, information that may be maintained in the page index).

Lee can be classified.

[0041] Those skilled in the art will understand that the above examples of criteria and subsets are illustrative and not limiting, and that any number of queries may be performed according to predetermined criteria.

It will be appreciated that criteria and subsets may be used. Classification of queries based on criteria determines whether the relevant criteria are "unknown" and/or

or it may contain a subset of "other" values, which may be resolved depending on the system configuration.

Maybe not.

[0042] At step 404, each subset of queries is provided to the clustering module 312 to form one or more subset-specific concept nets.

Create a work. In some embodiments, an existing unit dictionary 306 is used, while in other embodiments the subset is a unit dictionary 306.

It may be provided to the above analysis module 310 to generate a subset-specific set of units. Unit analysis and clustering minutes

This is disclosed in detail in Preliminary Application No. 60/510,220 cited above. Clustering analysis is advantageous for each quota.

This is performed individually on each subset, creating a set of concept networks for each subset. Resulting subset-specific

The positive concept network is returned to the trend analysis module 320.

[0043] At step 408, the trend analysis module 320 performs trend analysis using the subset-specific concept network.

In addition, it detects similarities and differences that reflect user behavior patterns that vary depending on the criteria. These other patterns

is referred to herein as a “trend”, and trend analysis generally involves clusters generated from different subsets (e.g.

For example, concept networks) and/or signatures to discover these differences. Trends and Trends

De-analysis can take many forms.

[0044] As an example, some trend analysis may be performed on specific search terms or

The relative frequencies of units can be compared. The units being compared may be, for example, different member units of a cluster.

It may be another signature unit for the cluster. For example, assuming there is a “Singer” cluster,

Some trend analysis of this cluster shows that for users under 21, “Avril Lavign” is preferred over “Celine Dion.”

While "e" appears to be the more popular curry term, the opposite appears to be true for users aged 35 and older.

It may appear. Alternatively, there is a "City" cluster, and the signature of this cluster includes the "Recruitment" and "Hotel" units.

Assuming that "jobs" is searched more frequently during the weekdays, any trend analysis of these signatures would show that "jobs" is searched more frequently during the weekdays, while on the weekends

“Hotel” may appear to be searched more frequently. (This means, for example, that users may search for jobs on weekdays.

(Reflects the tendency to plan vacations on weekends)

[0045] In some embodiments, trend analysis is used to detect more elusive trends in user interests and behavior.

It could be. For example, a well-known noun in one field (e.g. music) may be a well-known noun in another field (e.g.

Suppose you decide to go into politics. Although the overall interest of the noun doesn't change much,

There may be changes in the type of information. For example, search frequency for noun names and “song” or “album.”

decreases, while the frequency of searches for noun names and terms such as “politics” or “taxes” increases. different

Comparison of the neighbors of a noun name over a period of time reveals this trend.

[0046] At step 410, trend information is stored, for example, in a unit dictionary 306 for use in answering subsequent queries.

It's going to be great. Information may be stored in relation to the specific unit and/or cluster to which the information relates, or in relation to a signature or proposal relating to the unit or cluster.

[0047] The systems and processes described herein are examples and it is recognized that variations and modifications are possible. explanation in order

Process steps may be executed in parallel, the order of the steps may vary, and the steps may be modified or

I may be combined. Trend analysis may be performed simultaneously for any number of criteria (e.g.

For example, users under 21 years of age coming from the "shopping" sector), classification by criteria can be as loose as desired.

may be detailed. One subset-specific concept network may be compared to another subset-specific concept network.

There are different subset-specific concept networks, each with a "global" concept network generated from unclassified queries.

It can also be compared to twerking. Trend analysis refers to specific concept networks, e.g.

Generate meaningful data limited to those related to the above, or units used as a basis for suggesting related searches.

It may be successful. Trend analysis can be performed on units, clusters, signatures, or any combination thereof.

[0048] Like unit and cluster data, trend information can be advantageously updated or regenerated over time, allowing users to

Changes in interest (e.g., increased or decreased popularity of certain pop music performers, increased or decreased interest in job seeking versus traveling)

etc.) to capture changes in user behavior that can be reflected.

[0049] In another embodiment of the present invention, trend information provides search response to specific user interests by considering related trends.

It is used to help match the answer to the conditions. 5 shows a system 110 of FIG. 2 to respond to a query.

Indicates a method that can be used. The client 120 transmits a query to the search server system 160. search

Server system 160 transmits queries and/or their constituent units to concept server 180, which in turn uses the units.

Access before (306). The concept server 180 generates expressions from queries along with statistics and cluster information for various units.

Retrieves conceptual data related to a query, such as one or more separate units, as well as trend information related to the units.

Turn. This information can be accessed, for example, by hashing a query to identify the units contained within it and accessing the unit dictionary 306.

This can be derived by searching for an entry for each unit identified using In one embodiment, the returned information is

on trends related to one or more constituent units, or one or more clusters associated with any constituent unit.

Includes related information, units, statistics, and clusters.

[0050] Search server system 160 advantageously uses conceptual data received from concept server 180 to respond to queries.

do. Results returned by search server system 160 may advantageously relate to clusters and units and/or clusters.

Based on the understanding and extension of user needs captured by the unit, including trends,

Respond to your queries along with other relevant information, such as hints or tips about what you want to research.

Include the results of your answer.

[0051] For example, suppose a query contains an ambiguous term such as “Java” that can be used in more than one context.

all. These terms can be grouped into multiple clusters, such as the “Food and Beverage” cluster, the “Computer” cluster, and the “Location” cluster.

It can belong to a field. In some embodiments, search server system 160 may be configured to determine which cluster is most prominent, e.g.

You can use trend data to refine your response by choosing which positions should be given. For example

Well, users who search for "Java" during the week are likely to be most interested in computer languages, while users who search for "Java" on the weekend

Given that searched users are likely to be most interested in coffee, the order in which search results are presented may vary depending on the day of the week.

For example, on weekdays, results related to computer languages may be presented most prominently, while on weekends, results related to computer languages may be presented most prominently.

If so, results related to coffee can be presented most prominently. (Based on relationships with different clusters

(Some techniques for grouping search results by grouping are disclosed in preliminary application 60/510,220 cited above.)

[0052] As a second example, assume there is a four-character sequence (“WXYZ”) that is both the name of a rap group and the ticker symbol of a corporation.

do. A conceptual network or network based on queries received when the stock market was open versus queries received when the market was closed.

Comparing the clusters, in the former case WXYZ companies are searched more frequently, while in the latter case WXYZ

It may appear that lab groups are being searched for more frequently. Accordingly, whether the market was open when the query was received

Depending on whether it's closed, search results related to your company may be placed before or after results related to your lab group.

[0053] As a third example, think about the term “party,” which can be either a social event or a political party, depending on the context.

do. Some queries containing the term "party" may contain other disambiguating terms (e.g. "Republican").

party" or "party games"), but other queries may not include these terms.

there is. Trend analysis can be used to clarify responses to ambiguous queries. For example, Figure 6 shows

Frequency of queries containing the term "party" (in arbitrary units) divided by time of day (in 4-hour increments).

It is a graph of trend data. Such trend data may be subject to any query according to the process 400 described above.

This can be obtained by analyzing the data. Trend data set 602 (“X” symbol) indicates that “party” is a social event.

Matches queries that appear with other clearly related terms (e.g. “games”, “favors”, “supplies”). Trends

Data set 604 (“O” symbol) uses “party” as another term clearly related to a political party (e.g., “Republican”, “Democrat”).

Corresponds to queries that appear together with “party” and “candidate.” Trend data 606 (triangle) indicates that ambiguity is different from other terms in the query.

Matches queries containing "party" that cannot be resolved by reference to the term.

[0054] One way to resolve ambiguity in accordance with the present invention is to compare the shapes of different trends (e.g., as shown in Figure 6).

the first and/or second derivative of the data sets) and the trend for the ambiguous query set among the unambiguous sets.

It's about judging whether something matches me more closely than something else. In the example shown in Figure 6, there is an ambiguous trend data.

Eater (606) is more similar to “social event” trend data (602) than “political party” trend data (604). this

From this, users who entered ambiguous queries containing “party” were more interested in social events than political parties.

It can be inferred that there are, for example, results related to political parties before those related to social events.

Search results can be presented by presenting the results.

[0055] The similarity between two trend data sets can be considered and used to establish a level of confidence in the inferred intent.

It should be noted that there is. In the example shown in Figure 6, ambiguous trend data 606 is the "social event" trend data.

Since it matches more closely with data 602, the level of confidence in the inference increases. In other cases, the confidence level is

It may be lower or higher depending on user behavior. These levels of confidence ensure that responses to queries are clearly

However, there may be other factors. For example, given a sufficiently high level of trust, social events may initially be

Only relevant results are presented, in which case the user may be given the option to search for results related to a political party.

These examples are illustrative and modifications and changes are possible. For example, a number of criteria (e.g., time of day and use

Trends based on age of the user can be considered together in determining similarity. Trend comparison shows actual frequency values, derivatives

based on any suitable measure of the similarity of the two curves, such as (rate of change), higher-order derivative, or any combination thereof.

can do.

[0057] As a fourth example, consider the term “poison,” which could be either poison or a heavy metal band. Also, any

The query may include other terms that resolve ambiguity, such as "rat poison" or "poison control" and

The same query relates to poison, while "poison lyrics" or "poison mp3" relates to a band. But everything

Other queries will simply be ambiguous.

[0058] Analysis such as that described in the “party” example can be used to resolve ambiguity. Another approach is the "poison" band

belongs to the Music Players cluster, which includes other performers whose names are relatively unambiguous.

involves recognizing that (For example, virtually anyone entering "Britney Spears" as a query would

The user is interested in famous singers.) Queries containing unambiguous performer names serve as the "control" group.

can be used to draw inferences about the behavior of users seeking information about performers.

You can. This behavior refers to the performer, in order to estimate the likelihood that an ambiguous term is intended to refer to the performer.

This can be compared to the behavior of users typing ambiguous terms that may or may not be possible (e.g. "poison").

[0059] Numerous comparisons between control groups and queries containing ambiguous terms are possible. For example, ambiguous terms

Fractions and modifiers in queries that contain "poison" and also include terms clearly related to music (e.g., "lyrics").

The corresponding "control" fraction of queries containing unambiguous names (e.g., "Britney Spears") may be computed.

The control fraction estimates the likelihood that music-interested users will enter potentially ambiguous queries.

can be considered. By comparing this to the frequency of explicitly music-oriented queries containing "poison", we find that queries

We can infer the likelihood that a user who enters "poison" is interested in a band.

[0060] To make this example more specific, the fraction of queries that contain the term “poison” and also contain the term “lyrics” is

Assume that the fraction of queries that contain the term "Britney Spears" and that also contain the term "lyrics" is 01, and that the fraction of queries that contain the term "lyrics" is 02.

From this, we can infer that 1/2 of the users who typed "poison" probably did not mean to refer to the band.

You can. In this case, both band-related and poison-related results are displayed prominently. searched for poison

To estimate the likelihood that users will fail to include disambiguating terms,

Similar analyzes can be done using control fractions based on clusters (e.g., drug-related clusters).

You can. If the likelihood of an ambiguous query for a term in a drug-related cluster is small, the ambiguous query "poison" is banned.

It is suggested that it is easy to say de.

[0061] Additionally, receiving potentially ambiguous queries from users interested in drugs as opposed to users interested in music

User intent can be inferred by comparing the likelihood of doing something. For example, from users interested in music

29% of queries are single-unit queries containing only performer names, while 29% of queries are from users interested in drugs.

Assume that only 2% of queries contain a single unit. The user who enters the single unit query "poison" is

It can be inferred that there will be more interest in .

[0062] It will be appreciated that more sophisticated variations of this analysis may be used. For example, only one performer

Rather than calculating control fractions using

Similarly, rather than considering a single disambiguation term (as opposed to "matchbox twenty", "Johnny Cash", etc.)

Multiple terms may be used (e.g. "lyrics" as well as "mp3", "tour", "album", etc.) and other ambiguous terms.

Similar analysis can be applied to terms and other clusters.

[0063] As a fifth example, assume that the user enters a query called “digital cameras.” User uses a digital camera

You may be looking for a site that sells digital cameras, or you may be looking for information about digital cameras (product reviews, etc.). (example

For example, trend analysis of previous queries based on various criteria (including user profiles and/or longitudinal criteria).

It can be used to determine which one has more appropriate results.

[0064] As a sixth example, in some embodiments, search server system 160 may suggest related searches based on trend information.

there is. For example, if a user known to be under the age of 21 enters a query that includes a singer's name, the search server system

In addition to displaying a list of sites related to the singer, it analyzes trends among users under the age of 21.

suggests searches related to one or more other singers (e.g. "Avril Lavgne") recognized as popular on

You may. For a 35-year-old user, a different singer might be suggested (eg, “Celine Dion”). different view

Well, when a query for “new york city” is entered on a weekday, “jobs” suggestions are more prominent than “hotel” suggestions.

On the other hand, if this matches the trend data, the same query would be entered on a weekend and vice versa.

You can.

[0065] In some embodiments, the user's likely gender, age, or geographic location is identified by the search server system 160.

Trend data may also be used to infer information about this particular user. This inference is a trend

Matches user behavior against different criteria by comparing multiple queries entered by the same user regarding data

It may include ordering. These inferences are not guaranteed to be accurate, and they do not depend on how users act.

It should be understood that it reflects whether the user is active or not and is not necessarily the user. Anything made about a specific user

Inferences can also be made using trend data as an indication of possible user intent input by that user.

It can be used to tailor responses to additional queries. In this way, trend data is specific to the type of query entered.

It can be used to personalize the search server's response to the user.

[0066] Trend data may include advertising results, such as the ad(s) displayed on a particular occasion and/or the fees charged for a particular ad placement.

It may also be used by the provider of the search server system 160 in connection with the search server system 160.

[0067] Although the invention has been described with respect to specific embodiments, those skilled in the art will recognize that numerous variations are possible.

For example, the number and limitation of query criteria and subsets used in trend analysis may vary, and all received

Not all queries are used for trend analysis. Cluster, signature and trend information are dynamically defined;

Trend analysis may be performed over time (e.g., daily or weekly) to update trend information in response to changes in user behavior. In another embodiment, cluster, signature and trend data is updated in near real time.

Queries can be processed as they are received so that they can be dated. The automated systems and methods described herein are

Resulting clusters, including clusters, signatures, and trend information related to some or all of the units, clusters, and signatures.

The above dictionary may be augmented or supplemented by human review of all or part of it.

[0068] Embodiments described herein allow websites, links, and the World Wide Web (or a subset thereof) to serve as searchable resources.

In some cases, reference may be made to other specified technologies. The systems and processes described herein are compatible with other search resources (electronic

can be adapted for use in any database (database, document repository, etc.), with the result being that the content can be discovered as well as the content itself.

It should be understood that links or references to other locations may also be included.

[0069] Although the invention has thus been described with respect to specific embodiments, the invention is intended to cover all modifications and variations within the scope of the following claims.

It will be recognized as covering equivalents.

Claims (5)

As a method of processing a query,
Receiving a set of previous queries each containing one or more units;
Classifying the queries into subsets according to criteria;
generating one or more subset-specific concept networks for each subset of the queries; and
Trends for a unit by comparing corresponding subset-specific concept networks from at least two of the subsets
A query processing method, including the step of generating de-information. A query processing method, wherein the standard is a time standard. A query processing method wherein the criteria are defined with reference to one or more demographic characteristics of users.
law. According to claim 1,
Receiving the following query;
parsing the next query into one or more structural units; and
and using the trend information to clarify a response to the next query.
Query processing method using Jing. According to claim 6,
A query processing method, wherein the trend information is used to resolve ambiguous terms in the query.