JP2012141764A - Spam candidate extraction device and method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and a method that are capable of extracting a spam candidate of multi-word query.SOLUTION: A spam candidate extraction device 1 comprises: an original query extraction part 11 for extracting a multi-word query whose appearance frequency is equal to or greater than a predetermined frequency, as an original query, from a plurality of multi-word queries stored in query log storage means; a modified query generation part 12 for generating a modified query by adding a predetermined change to multiple queries constituting the original query; and a spam candidate extraction part 13 for extracting the original query as a spam candidate when a proportion of appearance frequency of the modified query to that of the original query is equal to or less than a predetermined proportion.

Description

  The present invention relates to a spam candidate extraction apparatus and method for extracting a plurality of word queries that are highly likely to be spam.

  In recent years, a very large number of web pages have been released in document systems such as WWW (World Wide Web) via a network. The user inputs an appropriate query at the search site, selects a URL (Uniform Resource Locator) of the web page that matches the search condition from a plurality of candidates displayed as search results, and follows the link to the selected URL. Thus, a desired web page can be browsed.

  By the way, for a company, whether or not its own web page is listed at the top of the search result page (especially the first page) is extremely important because it greatly affects the number of accesses to the company's web page. Therefore, each company has devised various methods so that their web page is listed above the search result page. Examples of these ideas include selecting the target query appropriately, bringing the appropriate keyword corresponding to the query near the top of the title or web page, and the keyword corresponding to the query in the web page. Techniques such as properly using and linking to sites related to more queries are widely known.

  However, there are also fraudsters that automatically and repeatedly send spam queries to a search site for the purpose of ranking a specific web page higher in the search result page. The fraudulent trader automatically transmits a combination of a specific product name or service name and a specific company name to the search site as a spam query. This is intended to display a web page of a specific company at a higher position on the search result page in response to a query including the specific product name or service name input by a general user.

  On the other hand, for an operator who operates a search site, if the above-mentioned fraud is performed, an appropriate search result page corresponding to the content of the original web page may be returned to a query input by a normal user. become unable. Therefore, search site operators need to properly detect fraudsters and eliminate the effects of spam queries sent by fraudsters when determining the display order of each web page for a query. There is.

  In detecting fraudsters, it is conceivable to appropriately determine spam queries, but even spam queries that are automatically sent are often indistinguishable from regular queries. In addition, there is a clever trick of mixing regular queries that are not spam queries, making spam queries more difficult to distinguish.

  Patent Document 1 discloses a link destination evaluation unit that evaluates a website linked from a blog, a blog evaluation unit that evaluates a blog using the evaluation result of the website, and a blog based on the blog evaluation. There is disclosed a spam blog detection device comprising a determination means for determining whether or not a spam blog.

JP 2010-066980 A

  However, with the techniques for evaluating the contents of a query or website, including the technique of Patent Document 1, as described above, the query itself is indistinguishable from a normal query even if the meaning of the query is analyzed. Cannot be extracted. Therefore, there is a need for a technique that can extract spam query candidates without performing semantic analysis of the query itself. Thus, the inventors have conducted intensive research and found a method for extracting spam query candidates by focusing on the statistics of a variation of a query including a plurality of words, and completed the present invention.

  Therefore, an object of the present invention is to provide a spam candidate extraction apparatus and method capable of extracting spam candidates of a multiple word query.

  The present invention provides the following solutions.

  (1) A spam candidate extraction device that extracts a plurality of word queries that are highly likely to be spam, and a query log that stores a plurality of query logs including a plurality of word queries configured by a combination of a plurality of queries A storage means, a plurality of word queries stored in the query log storage means, a plurality of word queries having an occurrence frequency of a plurality of word queries having a predetermined frequency or more as a source query extraction means, and the source query extraction means A change query generation means for generating a change query by applying a predetermined change to a plurality of queries constituting the original query, and the change query generation means for the appearance frequency of the original query extracted by the original query extraction means Is generated, and the ratio of the appearance frequencies of the change queries stored in the query log storage means is predetermined. If it is lower, spam candidate extracting device and a spam candidate extraction means for extracting the original query as spam candidate.

  According to the invention described in (1), the query log storage unit stores a log of a query including a plurality of word queries configured by a combination of a plurality of queries, and the original query extraction unit stores the query log storage unit in the query log storage unit. Among the plurality of stored multiple word queries, a multiple word query having an appearance frequency equal to or higher than a predetermined value is extracted as an original query. Then, the modified query generation unit generates a modified query by adding a predetermined change to the plurality of queries constituting the original query extracted by the original query extraction unit, and the spam candidate extraction unit includes the original query extraction unit The change query generated by the change query generation unit with respect to the appearance frequency of the extracted original query, and the ratio of the appearance frequency of the change query stored in the query log storage unit is less than or equal to the predetermined query, Extract as spam candidates. Thereby, without analyzing the meaning of the original query, a change query is generated by making a predetermined change to a plurality of queries constituting the original query, and among the query logs stored in the query log storage unit, By simply calculating the ratio of the appearance frequency of the change query to the appearance frequency of the original query, it is possible to extract the original query below the predetermined ratio as a spam candidate and present it to the administrator. As a result, spam candidates for multi-word queries by fraudsters, including clever tricks such as mixing unknown spam queries that are not stored in the spam query database, or adding normal non-spam queries to spam queries. Can be extracted effectively and efficiently.

  (2) The spam candidate extracting device according to (1), wherein the predetermined change includes a change in which a predetermined particle is added between a plurality of queries constituting the original query extracted by the original query extracting unit. Candidate extraction device.

  According to the invention described in (2), even if the meaning of the original query is not analyzed, a change query is generated by adding a predetermined particle between a plurality of queries constituting the original query, and stored in the query log storage unit. The query of the original query that is less than or equal to the predetermined ratio can be obtained by simply calculating the ratio of the appearance frequency of the modified query generated by the modified query generation means to the appearance frequency of the original query extracted by the original query extracting means. Can be extracted as spam candidates and presented to the administrator. As a result, spam candidates for multi-word queries by fraudsters, including clever tricks such as mixing unknown spam queries that are not stored in the spam query database, or adding normal non-spam queries to spam queries. Can be extracted effectively and efficiently.

  (3) The spam candidate extraction device according to (1) or (2), wherein the predetermined change includes a change for changing an order of a plurality of queries constituting the original query extracted by the original query extraction unit. Spam candidate extraction device.

  According to the invention described in (3), the query stored in the query log storage unit can be generated by changing the order of a plurality of queries constituting the original query without analyzing the meaning of the original query. Of these logs, just calculate the ratio of the appearance frequency of the change query generated by the change query generation means to the appearance frequency of the original query extracted by the original query extraction means, and the original query below the predetermined ratio can be selected as a spam candidate. Can be extracted and presented to the administrator. As a result, spam candidates for multi-word queries by fraudsters, including clever tricks such as mixing unknown spam queries that are not stored in the spam query database, or adding normal non-spam queries to spam queries. Can be extracted effectively and efficiently.

  (4) The spam candidate extracting device according to any one of (1) to (3), wherein the predetermined change is a change that combines a plurality of queries constituting the original query extracted by the original query extraction unit. Including spam candidate extraction device.

  According to the invention described in (4), a query log stored in the query log storage unit is generated by combining a plurality of queries constituting the original query and generating a change query without analyzing the meaning of the original query. Of these, by simply calculating the ratio of the appearance frequency of the change query generated by the change query generation means to the appearance frequency of the original query extracted by the original query extraction means, the original queries below the predetermined ratio are extracted as spam candidates. And can be presented to the administrator. As a result, spam candidates for multi-word queries by fraudsters, including clever tricks such as mixing unknown spam queries that are not stored in the spam query database, or adding normal non-spam queries to spam queries. Can be extracted effectively and efficiently.

  (5) The spam candidate extraction device according to any one of (1) to (4), wherein the predetermined change is performed by changing an order of a plurality of queries constituting the original query extracted by the original query extraction unit. Spam candidate extraction device that includes changes to be combined.

  According to the invention described in (5), without changing the meaning of the original query, a change query is generated by changing the order of a plurality of queries constituting the original query and combining them, and the query log storage means Of the stored query logs, it is only necessary to calculate the ratio of the appearance frequency of the change query generated by the change query generation means to the appearance frequency of the original query extracted by the original query extraction means. Queries can be extracted as spam candidates and presented to the administrator. As a result, spam candidates for multi-word queries by fraudsters, including clever tricks such as mixing unknown spam queries that are not stored in the spam query database, or adding normal non-spam queries to spam queries. Can be extracted effectively and efficiently.

  (6) A spam candidate extraction method in which a computer including a control unit and a storage unit extracts a plurality of word queries that are likely to be spam. The storage unit includes a combination of a plurality of queries. Query log storage means storing a query log including a plurality of multiple word queries is provided, and the control means is a plurality of appearance frequencies of the plurality of word queries stored in the query log storage means that are not less than a predetermined number. An original query extracting step for extracting a word query as an original query, and a modified query generating step for generating a modified query by making a predetermined change to a plurality of queries constituting the original query extracted in the original query extracting step; , The variation query generated in the variation query generation step with respect to the appearance frequency of the original query extracted in the original query extraction step. A Li, how to perform a spam candidate extraction step the ratio of frequency of occurrence of the variations queries stored in the query log storage means to extract when the predetermined or less, the original query as spam candidate.

  According to the invention described in (6), the same effect as in (1) can be expected by using the method.

  ADVANTAGE OF THE INVENTION According to this invention, the spam candidate extraction apparatus and its method which can extract a spam query effectively and efficiently can be provided.

It is a figure which shows the whole structure for demonstrating the spam candidate extraction apparatus 1 which concerns on embodiment of this invention. It is a figure which shows the query log database 21 which concerns on embodiment of this invention. It is a flowchart of the spam candidate extraction device 1 according to the embodiment of the present invention. It is a figure which shows the Example of the spam candidate extraction apparatus 1 which concerns on embodiment of this invention.

[overall structure]
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a diagram showing an overall configuration for explaining a spam candidate extraction apparatus 1 according to an embodiment of the present invention. The spam candidate extraction device 1 is connected to the search engine 2 via a network, and includes a control unit 10 and a storage unit 20. The control unit 10 is a part that comprehensively controls each function related to the spam candidate extraction apparatus 1, and includes an original query extraction unit 11, a modified query generation unit, and a spam candidate extraction unit 13. On the other hand, the storage unit 20 is a part that stores various programs (not shown) and a database for functioning as the spam candidate extraction apparatus 1, and includes a query log database 21.

  First, the query log database 21 will be described with reference to FIG. A terminal user using a terminal (not shown) uses the search engine 2 to input a query related to an item to be searched from the terminal. The terminal accepts this and transmits a query to the search engine 2. The query log database 21 is a database in which past queries transmitted from the terminal to the search engine 2 are stored. The query log database 21 functions as a query log storage unit, and the query log database 21 stores a log of a query including a plurality of word queries configured by a combination of a plurality of queries. Although not shown, the query log database 21 normally also stores a search log that associates the IP address of the terminal 2 that transmitted the query with the date and time when the query was received from the terminal 2.

  Returning to FIG. 1, the original query extraction unit 11 functions as an original query extraction unit, and extracts, as an original query, a plurality of word queries having an appearance frequency that is greater than or equal to a predetermined value among a plurality of word queries stored in the query log database 21. . “Appearance frequency” refers to the appearance frequency of each multiple word query in the query log database 21. The “predetermined” value can be set as appropriate. The smaller the value, the more original queries are extracted. For example, the predetermined value may be 1,000, 10,000, or 100,000.

  The variation query generation unit 12 functions as a variation query generation unit, and generates a variation query by applying a predetermined change to a plurality of queries constituting the original query extracted by the original query extraction unit 11.

  As a mode of the predetermined change, (1) a predetermined particle such as “no”, “de”, “ha”, “wa”, etc. is provided between the plurality of queries constituting the original query extracted by the original query extraction unit 11. Changes to be added, (2) Changes to change the order of a plurality of queries constituting the original query extracted by the original query extraction unit 11, (3) Changes to combine a plurality of queries constituting the original query extracted by the original query extraction unit 11 And (4) a change in which the order of a plurality of queries constituting the original query extracted by the original query extraction unit 11 is changed and combined. A change query may be generated using only one type of these changes, or a plurality of change queries may be generated using a plurality of types.

  The spam candidate extraction unit 13 functions as a spam candidate extraction unit, and is a modified query generated by the modified query generation unit 12 for the appearance frequency of the original query extracted by the original query extraction unit 11, and is stored in the query log database 21. The original query is extracted as a spam candidate when the ratio of the appearance frequencies of the changed query is less than or equal to a predetermined value. The predetermined value can be set as appropriate. If this value is small, many original queries can be extracted as spam candidates, but the possibility that even non-spam is actually extracted as spam candidates increases. On the other hand, when this value is large, there is a high possibility that only spam is extracted as a spam candidate, but there is a high possibility that it is not extracted as a spam candidate even though it is actually spam. The administrator of the spam candidate extracting apparatus 1 may set a predetermined ratio with this in mind. For example, it is set to 1/100, 1 / 1,000, or 10,000 minutes. It is conceivable to set it to 1.

[Hardware and Software Configuration of this Embodiment]
The hardware and software configurations of this embodiment will be described. The hardware of the spam candidate extraction device 1 includes a CPU as a control device (control unit), a storage device (storage unit), a communication device (communication unit), a display device (display unit), and an input device (input unit). Is included. Examples of the storage device (storage unit) include a memory (RAM, ROM, etc.), a hard disk drive (HDD), an optical disk (CD, DVD, etc.) drive, and the like. Examples of the communication device (communication unit) include various wired or wireless interface devices. Examples of the display device (display unit) include various displays such as a liquid crystal display or a plasma display. Examples of the input device (input unit) include a keyboard, a pointing device (such as a mouse), a remote control device, and the like.

  The software of the spam candidate extraction apparatus 1 includes programs and data for controlling the hardware. Programs and data are stored in a storage device (storage unit), and are appropriately executed and referenced by a control device (control unit). Further, the computer program and data can be distributed by an electromagnetic method, or can be recorded and distributed on a computer-readable medium such as a CD-ROM.

[flowchart]
FIG. 3 is a flowchart of the spam candidate extraction apparatus 1 according to the embodiment of the present invention.

  In step S1, the original query extraction unit 11 executes an original query extraction process. In this process, the original query extraction unit 11 performs a process of extracting, as an original query, a plurality of word queries having an appearance frequency that is equal to or higher than a predetermined value among a plurality of word queries stored in the query log database 21.

  In step S2, the change query generation unit 12 executes a change query generation process. In this process, the modified query generation unit 12 executes a process of generating a modified query by making a predetermined change to a plurality of queries constituting the original query extracted by the original query extracting unit 11. In the present embodiment, (1) a predetermined particle such as “no”, “de”, “ha”, “wa” is added between a plurality of queries constituting the original query extracted by the original query extraction unit 11. A first variation query is generated, (2) a second variation query is generated by changing the order of a plurality of queries constituting the original query extracted by the original query extraction unit 11, and (3) the original query extraction unit 11 A third change query is generated by combining a plurality of queries constituting the original query extracted by (4), and (4) the order of the plurality of queries constituting the original query extracted by the original query extraction unit 11 is changed. The description will be made assuming that the fourth variation query is generated by combining a plurality of queries. However, the present invention is not limited to the aspect of the present embodiment, and only a part of the variation query may be generated. Even if the modified query is generated with other modifications There.

  In step S3, the spam candidate extraction unit 13 executes a spam candidate extraction process. In this process, the spam candidate extraction unit 13 is a variation query generated by the variation query generation unit 12 with respect to the appearance frequency of the original query extracted by the original query extraction unit 11 and is stored in the query log database 21. When the ratio of the appearance frequencies of the shaped queries is equal to or less than a predetermined value, a process of extracting the original query as a spam candidate is executed. When this process ends, the control unit 10 ends the series of processes.

[Example of spam candidate extraction apparatus 1 according to this embodiment]
FIG. 4 shows an example of the spam candidate extraction device 1 described in the present embodiment.

  First, the query log database 21 stores a log of a query including a plurality of word queries configured by a combination of a plurality of queries. For example, in the example of FIG. 2, a multi-word query composed of a combination of “cheap airline ticket abc tours”, a multi-word query composed of a combination of “esthetic beauty def”, and a combination of “ghi earthen house” A log of a query including a multiple word query configured and a multiple word query configured in combination with the “study abroad jkl agency” is stored.

  Then, the original query extraction unit 11 extracts, as an original query, a plurality of word queries whose appearance frequency is greater than or equal to a predetermined value from among the plurality of word queries stored in the query log database 21. For example, when the appearance frequency of a multiple word query configured by a combination of “cheap airline ticket abc tours” is equal to or greater than a predetermined number, the original query extraction unit 11 is configured by a combination of “cheap airline ticket abc tours”. Extract multiple word queries as original queries. In the upper part of FIG. 4, it is shown that “cheap airline ticket abc tours” which is a multiple word query is an original query.

Subsequently, the variation query generation unit 12 generates a variation query by applying a predetermined change to a plurality of queries constituting the original query extracted by the original query extraction unit 11. FIG. 4 discloses a modification.
First, (1) a first modified form by adding a predetermined particle such as “no”, “de”, “ha”, “wa” between multiple queries “cheap airline ticket abc tours” constituting the original query Generate a query. In this case, “abc tours of cheap flights”, “abc tours with cheap flights”, “abc tours of cheap flights”, “abc tours of cheap flights”, etc. are the first variation queries.
Subsequently, (2) the second variation query is generated by changing the order of the multiple queries “cheap airline ticket abc tours” constituting the original query. In this case, “abc tours cheap air ticket” is the second variation query.
Subsequently, (3) a plurality of queries “cheap airline ticket abc tours” constituting the original query are combined to generate a third variation query. In this case, “cheap airline ticket abc tours” excluding the space between the multiple queries is the third variation query.
Subsequently, (4) the order of the multiple queries “cheap airline ticket abc tours” constituting the original query is changed, and the changed multiple queries are combined to generate a fourth changed query. In this case, the “abc tours cheap air ticket” in which the space between the multiple queries is further excluded from the “abc tours cheap air ticket” whose order has been changed is the fourth variation query.

  The spam candidate extraction unit 15 then generates a modified query generated by the modified query generation unit 12 for the appearance frequency of the original query extracted by the original query extraction unit 11 among the plurality of word queries stored in the query log database 21. The original query is extracted as a spam candidate. In the example of FIG. 4, the ratio of x and y1, the ratio of x and y2, the ratio of x and y3, the ratio of x and y4, and the ratio of x and y5 are calculated, respectively, and the calculated values are all below a predetermined value. In some cases, the original query is extracted as a spam candidate, but this is not a limitation.

  According to the invention described in the present embodiment, even if the meaning of the original query is not analyzed, a change query is generated by making a predetermined change to a plurality of queries constituting the original query, and is stored in the query log database 21. In addition, by calculating the ratio of the appearance frequency of the modified query to the appearance frequency of the original query among the multiple word queries, it is possible to extract the original query below the predetermined ratio as a spam candidate and present it to the administrator. it can. As a result, spam candidates for multi-word queries by fraudsters, including clever tricks such as mixing unknown spam queries that are not stored in the spam query database, or adding normal non-spam queries to spam queries. It is possible to provide a spam candidate extracting apparatus 1 and a method for extracting the spam candidate effectively and efficiently.

  In addition, the effect described in this embodiment only enumerated the most suitable effect which arises from this invention, and the effect by this invention is not limited to what was described in this embodiment.

DESCRIPTION OF SYMBOLS 1 Spam candidate extraction apparatus 10 Control part 11 Original query extraction part 12 Variant query production | generation part 13 Spam candidate extraction part 20 Storage part 21 Query log database

Claims (6)

A spam candidate extraction device that extracts a plurality of word queries that are likely to be spam,
A query log storage means in which a log of a query including a plurality of word queries configured by a combination of a plurality of queries is stored;
Original query extraction means for extracting, as an original query, a plurality of word queries having an appearance frequency of a predetermined value or more among a plurality of word queries stored in the query log storage means;
Change query generation means for generating a change query by making a predetermined change to a plurality of queries constituting the original query extracted by the original query extraction means;
The change query generated by the change query generation means with respect to the appearance frequency of the original query extracted by the original query extraction means, and the ratio of the appearance frequency of the change query stored in the query log storage means is predetermined. A spam candidate extraction device comprising: spam candidate extraction means for extracting the original query as a spam candidate when: The spam candidate extraction device according to claim 1,
The spam candidate extracting apparatus, wherein the predetermined change includes a change in which a predetermined particle is added between a plurality of queries constituting the original query extracted by the original query extracting unit. The spam candidate extracting device according to claim 1 or 2,
The spam candidate extracting apparatus, wherein the predetermined change includes a change for changing an order of a plurality of queries constituting the original query extracted by the original query extracting unit. The spam candidate extracting device according to any one of claims 1 to 3,
The spam candidate extracting apparatus, wherein the predetermined change includes a change that combines a plurality of queries constituting the original query extracted by the original query extraction unit. The spam candidate extracting device according to any one of claims 1 to 4,
The spam candidate extracting apparatus, wherein the predetermined change includes a change in which the order of a plurality of queries constituting the original query extracted by the original query extraction unit is changed and combined. A spam candidate extraction method in which a computer including a control unit and a storage unit extracts a plurality of word queries that are highly likely to be spam,
The storage means is provided with a query log storage means in which a log of a query including a plurality of word queries configured by a combination of a plurality of queries is stored,
The control means includes
An original query extraction step of extracting, as an original query, a plurality of word queries having an appearance frequency of a predetermined value or more among the plurality of word queries stored in the query log storage unit;
A change query generation step of generating a change query by making a predetermined change to a plurality of queries constituting the original query extracted in the original query extraction step;
The change query generated in the change query generation step with respect to the appearance frequency of the original query extracted in the original query extraction step, wherein the ratio of the appearance frequency of the change query stored in the query log storage unit is predetermined. A spam candidate extraction step of extracting the original query as a spam candidate when:

Images