JP2011100208A - Action estimation device, action estimation method, and action estimation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a minute and highly precise estimation result concerning the action of a user. <P>SOLUTION: An action estimation device 10 includes: a tree structure generating part 121 for generating tree structure data 112 where each WEB page referred to by the user is defined as a node based on reference history data 111 indicating the reference order of the WEB pages by the user; a function analyzing part 122 for analyzing the function of the WEB page corresponding to each node included in the tree structure data 112; and an action estimating part 123 for estimating the action of the user based on a state that each node included in the tree structure data 112 is associated with one another, and also the analysis result of the function of the WEB page corresponding to each node. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a behavior estimation device, a behavior estimation method, and a behavior estimation program.

  In recent years, a technique for estimating a user's action referring to a WEB page has been actively researched and actually used in various scenes. For example, in an online shopping site, what is the user trying to purchase based on the product sold on the WEB page that the user is currently referring to and the past information on the purchase of the product by the user in the past? A technique is used that estimates and displays recommended products selected based on the estimation result in an empty space on the WEB page.

  In addition, the browsing behavior is classified into categories from the record of the web browsing behavior performed by the user on the PC, and “new search behavior”, “redisplay of previous search behavior”, “WEB browsing by clicking from search behavior” In addition, a technique for estimating user behavior such as “WEB browsing started from other than search” is also known (see Patent Document 1).

JP 2009-128937 A

Lee, Uichin, “Automatic identification of user goals in Web search”, Proceedings of the 14th international conference on World Wide Web, ACM, 2005, pp391-400

  However, in the above-described conventional technology, the approximate “content” of the user's current actions can be estimated, but the “purpose” cannot be estimated, so the estimation result may not be effectively used. .

  For example, in an online shopping site, when a user is trying to reference a WEB page of a product, if the purpose of the action is “to purchase the product or the same type of product”, the same type of product is displayed as a recommended product. It is effective to do. However, if the purpose of the action is “search for something you want,” it is less effective to display the same type of product as a recommended product. Rather, recommend a product that matches the user's preference. It is more effective to display as a product.

  This invention is made | formed in view of the above, Comprising: It aims at obtaining a detailed and highly accurate estimation result about a user's action.

  In order to solve the above-described problems and achieve the object, the present invention provides a storage means for storing reference history data indicating an order in which a user refers to a plurality of WEB pages, and based on the reference history data, Tree structure data having each WEB page referred to by a user as a node, and a second node corresponding to a second WEB page referred to next to the first WEB page is the first WEB Tree structure generation means for generating tree structure data associated with each node so as to be a child node of the first node corresponding to the page, and each node included in the tree structure data generated by the tree structure generation means Function analysis means for analyzing the function of the WEB page corresponding to each node, and each node included in the tree structure data generated by the tree structure generation means The situation are associated, on the basis by functional analysis means and the analysis result of the function of the WEB page corresponding to each node, characterized by comprising a behavior estimation means for estimating the behavior of the user.

  The behavior estimation apparatus, the behavior estimation method, and the behavior estimation program according to the present invention have an effect that a detailed and highly accurate estimation result can be obtained for the user's behavior.

FIG. 1 is a diagram illustrating the configuration of the behavior estimation apparatus according to the first embodiment. FIG. 2 is a diagram illustrating an example of reference history data. FIG. 3 is a diagram illustrating an example of tree structure data. FIG. 4 is a diagram illustrating an example of a tree structure. FIG. 5 is a diagram illustrating an example of the function analysis rule. FIG. 6 is a diagram illustrating an example of the tree structure data after the function analysis. FIG. 7 is a diagram illustrating an example of a tree structure after function analysis. FIG. 8 is a diagram illustrating an example of a behavior estimation rule. FIG. 9 is a diagram illustrating an example of a subtree. FIG. 10 is a diagram illustrating another example of a partial tree. FIG. 11 is a diagram illustrating a processing procedure of behavior estimation processing executed by the behavior estimation device. FIG. 12 is a diagram illustrating the configuration of the behavior estimation apparatus according to the second embodiment. FIG. 13 is a diagram illustrating an example of reference history data. FIG. 14 is a diagram illustrating an example of tree structure data. FIG. 15 is a diagram illustrating a processing procedure of tree structure generation processing. FIG. 16A is a diagram illustrating a state after the initial processing is completed. FIG. 16B is a diagram illustrating a state after the processing of the first row is completed. FIG. 16C is a diagram illustrating a state after the processing on the second row is completed. FIG. 16D is a diagram illustrating a state after the processing on the third row is completed. FIG. 16-5 is a diagram illustrating a state after the processing on the fourth row is completed. FIG. 17 is a diagram illustrating an example of the tree structure data after the function analysis. FIG. 18 is a diagram illustrating a processing procedure of the function analysis processing. FIG. 19 is a diagram illustrating an example of a subtree extraction rule. FIG. 20 is a diagram illustrating an example of partial tree extraction data. FIG. 21 is a diagram illustrating a processing procedure of the partial tree extraction processing. FIG. 22 is a diagram illustrating an example of a behavior estimation rule. FIG. 23 is a diagram illustrating a processing procedure of the subtree matching process. FIG. 24 is a diagram illustrating another processing procedure of the subtree matching process. FIG. 25 is a diagram illustrating an example of user response data. FIG. 26 is a diagram illustrating a processing procedure of behavior estimation rule setting processing. FIG. 27 is a diagram illustrating an example of a subtree extraction rule. FIG. 28 is a diagram illustrating an example of an estimation result. FIG. 29 is a diagram illustrating an example of tree structure data. FIG. 30 is a diagram illustrating an example of a partial tree extraction rule. FIG. 31 is a diagram illustrating an example of an estimation result. FIG. 32 is a diagram illustrating an example of a subtree extraction rule. FIG. 33 is a diagram illustrating an example of an estimation result. FIG. 34 is a diagram illustrating a configuration of a computer.

  Embodiments of a behavior estimation device, a behavior estimation method, and a behavior estimation program according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments.

  First, the configuration of the behavior estimation apparatus 10 according to the present embodiment will be described with reference to FIG. The behavior estimation device 10 is a device that estimates a user's behavior from a history of referring to a WEB page by executing a behavior estimation method according to the present embodiment. As illustrated in FIG. 1, the behavior estimation apparatus 10 includes a storage unit 110 and a control unit 120.

  The storage unit 110 is, for example, a hard disk device or a semiconductor storage device, and stores reference history data 111, tree structure data 112, a function analysis rule 113, and an action estimation rule 114.

  The reference history data 111 is history information indicating in what order the WEB page is referred to by the user. The content of the reference history data 111 may be, for example, an access log recorded every time the WEB server receives a request message, or a log recorded every time a WEB browser requests a WEB page from the WEB server. Also good.

  The tree structure data 112 is data having a tree structure that indicates which WEB page the user has referred to and then which other WEB page is referred to. In the tree structure data 112, each WEB page referred to by the user is held as one node, the node corresponding to the transition source WEB page is a parent node, and the node corresponding to the forward transition destination WEB page is the node. The nodes are connected so as to be child nodes. For example, when a user traces a WEB page called B by tracing a link of a WEB page called A, and returns to A and traces a link and references a WEB page called C, the node corresponding to A corresponds to B And the parent node of the node corresponding to C.

  The function analysis rule 113 is information for specifying the function of the WEB page corresponding to each node included in the tree structure data 112. The behavior estimation rule 114 is based on the tree structure of the node in the tree structure data 112 and the function of the WEB page corresponding to each node in the tree structure data 112 specified using the function analysis rule 113. This is information for estimating an action.

  Note that part or all of the data illustrated as being stored in the storage unit 110 in FIG. 1 can be accessed from the behavior estimation device 10 to other devices configured to be communicable with the behavior estimation device 10. It may be stored in a state.

  The control unit 120 is a control unit that controls the entire behavior estimation apparatus 10, and includes a tree structure generation unit 121, a function analysis unit 122, and a behavior estimation unit 123.

  The tree structure generation unit 121 extracts information on a user whose behavior is to be estimated from the reference history data 111, associates the nodes corresponding to the referenced WEB pages based on the reference order, and sets the tree structure data 112. Generate. The function analysis unit 122 specifies the function of the WEB page corresponding to each node included in the tree structure data 112 generated by the tree structure generation unit 121 based on the function analysis rule 113.

  The behavior estimation unit 123 uses the tree structure of the nodes in the tree structure data 112 representing the reference order of the WEB pages and the function of the WEB page corresponding to each node in the tree structure data 112 specified using the function analysis rule 113. Is collated with the behavior estimation rule 114 to estimate the user's behavior. That is, the behavior estimation unit 123 does not estimate a user's behavior by paying attention to a single WEB page referred to by the user, but in what order the WEB page of which function is used by the user. The user's behavior is estimated based on whether it is referred to.

  In general, the order in which the user refers to the WEB page of which function depends on the purpose of the user. For example, if you are searching for a specific product to purchase after deciding which type of product to purchase, the user will use the search WEB page to display a list of that type of product. From there, it is often referred to a WEB page on which detailed information on individual products is posted. In other words, if you are searching for a specific product to purchase after deciding what kind of product to purchase, after referring to the WEB page that has the function of “product search”, the function of “product information” It often refers to the WEB page that it has.

  On the other hand, if you are looking for a product that seems to be interesting, the user browsed the WEB page that lists the top-selling products, and then detailed information about each product was posted there. Often refers to a WEB page. That is, when a product that looks interesting is vaguely searched, a WEB page having a function “product information” is often referred to after referring to a WEB page having a function “ranking”.

  In this way, even if the function of the WEB page that the user finally refers is the same, the WEB that the user refers to before reaching the WEB page depends on what purpose the user has. The page functions are different. Further, even if the function of the WEB page that the user first refers to is the same, the function of the WEB page that the user finally refers to depends on what purpose the user has.

  In the behavior estimation apparatus 10 according to the present embodiment, as described above, the behavior estimation unit 123 estimates the user's behavior based on what function the WEB page is referred to by the user in the order. Therefore, it is possible to estimate the user's action by grasping what purpose the user has, and as a result, a detailed and highly accurate estimation result can be obtained.

  Next, processing executed by each unit included in the control unit 120 illustrated in FIG. 1 will be described in more detail.

  The tree structure generation unit 121 reads, from the reference history data 111, data related to the WEB page referred to by the user whose behavior is to be estimated, in the order of reference. If the WEB page indicated by the read data is not yet reflected as a node in the tree structure generation unit 121, a node corresponding to the WEB page is generated, and the data read immediately before the generated node is displayed. The node corresponding to the indicated WEB page is associated as a child node.

  On the other hand, if the WEB page indicated by the read data is already reflected as a node in the tree structure generation unit 121, the WEB page indicated by the next read data is generated without generating a node corresponding to the WEB page. Is processed as a child node to the node corresponding to the same WEB page as the WEB page indicated by the data read this time.

  A specific example of processing executed by the tree structure generation unit 121 will be described with reference to FIGS. FIG. 2 is a diagram illustrating an example of the reference history data 111. As shown in the example of FIG. 2, the reference history data 111 includes items such as a user ID and a URL (Uniform Resource Locator). The user ID is an identifier for identifying the user who referred to the WEB page, and the URL is the address of the WEB page referred to by the user.

  The value of the user ID may be, for example, a user ID used when the user logs in to the WEB site, or a session ID issued when the user logs in to the WEB site. Alternatively, it may be a network address assigned to a device used by a user to access a WEB page.

  A row is added to the reference history data 111 every time the user refers to the WEB page. For this reason, if a row in which the same user ID as the user ID of the target person whose behavior is to be estimated is read in order from the reference history data 111, which web page the user referred to in which order Can be grasped.

For example, lines 1 to 4 of the reference history data 111 shown in FIG. 2 indicate that the user corresponding to the user ID “U03”
http://www.foobar.com/
After referring to the WEB page with the address
http://ranking.foobar.com/
Refer to the WEB page with the address and press the back button etc.
http://www.foobar.com/
Return to the WEB page with the address, and then follow the link.
http://search.foobar.com/srch?name=xxx
It is shown that the WEB page at the address is referred to.

  3 is a diagram illustrating an example of the tree structure data 112 generated by the tree structure generation unit 121 from the reference history data 111 illustrated in FIG. 2, and FIG. 4 is included in the tree structure data 112 illustrated in FIG. It is the figure which represented the logical structure of the node typically.

  As shown in the example of FIG. 3, the tree structure data 112 has items such as a user ID, a node ID, a parent node ID, a URL, and a page function, for each node, that is, for each WEB page referred to by the user. A line is added.

  The user ID is an identifier for identifying the user who referred to the WEB page, and corresponds to the user ID of the reference history data 111. The node ID is an identifier for identifying a node corresponding to the WEB page referred to by the user. The parent node ID is the node ID of the parent node of the node corresponding to the WEB page referenced by the user. The tree structure data 112 logically represents a tree structure by a combination of a node ID and a parent node ID.

  The URL is an address of a WEB page referred to by the user. The page function is a value indicating the function of the WEB page referred to by the user, and the value is not set when the tree structure data 112 is generated by the tree structure generation unit 121.

  For example, when estimating the user behavior corresponding to the user ID “U03”, the tree structure generation unit 121 first reads the first row of the reference history data 111 illustrated in FIG. 2. Since the tree structure generation unit 121 has not yet generated a node corresponding to the URL of the read line, the tree structure generation unit 121 generates a node corresponding to the WEB page indicated by the read line. At this stage, since there is no line read immediately before, the association with the parent node is not performed.

  Specifically, the tree structure generation unit 121 adds a line to the tree structure data 112, and transcribes the user ID and URL of the line read from the reference history data 111. Then, a newly issued unique value (“N01”) is set as the node ID, and the parent node ID is left unset. By this processing, the first line of the tree structure data 112 shown in FIG. 3 is generated.

  Subsequently, the tree structure generation unit 121 reads the second row of the reference history data 111 illustrated in FIG. Since the node corresponding to the URL of this line has not yet been generated, the tree structure generation unit 121 generates a node corresponding to the WEB page indicated by the read line, and corresponds to the WEB page indicated by the line read immediately before. Associate the node as a child node.

  Specifically, the tree structure generation unit 121 adds a line to the tree structure data 112, and transcribes the user ID and URL of the line read from the reference history data 111. Then, the newly issued unique value (“N02”) is set as the node ID, and the node ID (“N01”) of the immediately preceding row is set as the parent node ID. By this process, the second line of the tree structure data 112 shown in FIG. 3 is generated.

  Subsequently, the tree structure generation unit 121 reads the third row of the reference history data 111 illustrated in FIG. Since the URL of this line is the same as the URL of the node added to the first line of the tree structure data 112, the tree structure generation unit 121 does not generate a node corresponding to the WEB page indicated by the read line.

  Subsequently, the tree structure generation unit 121 reads the fourth line of the reference history data 111 illustrated in FIG. Since the node corresponding to the URL of this line has not yet been generated, the tree structure generating unit 121 generates a node corresponding to the WEB page indicated by the read line. Since the URL of the line read immediately before has already created a node, the generated node is associated as a child node with the node corresponding to the same WEB page as the WEB page indicated by the line read immediately before.

  Specifically, the tree structure generation unit 121 adds a line to the tree structure data 112, and transcribes the user ID and URL of the line read from the reference history data 111. Then, a newly issued unique value (“N03”) is set as the node ID, and the node ID (“N01”) in the first line in which the same URL is set is set as the parent node ID. By this processing, the third line of the tree structure data 112 shown in FIG. 3 is generated.

  Subsequently, the tree structure generation unit 121 reads the fourth line of the reference history data 111 illustrated in FIG. Since the node corresponding to the URL of this line has not yet been generated, the tree structure generation unit 121 generates a node corresponding to the WEB page indicated by the read line, and corresponds to the WEB page indicated by the line read immediately before. Associate the node as a child node.

  Specifically, the tree structure generation unit 121 adds a line to the tree structure data 112, and transcribes the user ID and URL of the line read from the reference history data 111. Then, a newly issued unique value (“N04”) is set as the node ID, and the node ID (“N03”) of the immediately preceding row is set as the parent node ID. By this processing, the fourth line of the tree structure data 112 shown in FIG. 3 is generated.

  As a result of the above processing, as shown in FIG. 4, a node with a node ID of “N02” and a node with a node ID of “N03” exist as child nodes of the node with a node ID of “N01”. As a child node of the node with the node ID “”, a tree structure in which a node with the node ID “N04” exists is formed. Thereafter, the tree structure generation unit 121 reads the fifth and subsequent lines of the reference history data 111 shown in FIG. 2 and continues the processing, thereby generating the fifth to sixth lines of the tree structure data 112 shown in FIG. As a result, a tree structure as shown in FIG. 4 is formed.

  As is clear from the above description, in the behavior estimation method according to the present embodiment, when there is a transition back to the already-referenced WEB page, the node corresponding to the transition-destination WEB page is set as a child node. Instead of being associated with a node corresponding to a WEB page, a tree structure is formed such that a node corresponding to a WEB page referred to thereafter branches off from a node corresponding to the returned WEB page.

  When a user refers to a WEB page for any purpose, the transition to the referenced WEB page is performed in order to refer to a new WEB page from the returned WEB page and return to the referenced WEB page. Often it does not make sense. In the behavior estimation method according to the present embodiment, a node is added only when a new WEB page is referred to and associated with a node corresponding to the transition source WEB page. A structure can be obtained.

  Next, a specific example of processing executed by the function analysis unit 122 will be described with reference to FIGS. FIG. 5 is a diagram illustrating an example of the function analysis rule 113. As shown in the example of FIG. 5, the function analysis rule 113 has items such as a URL pattern and a page function. The URL pattern is a character string used for matching with the URL of the WEB page, and is set in a predetermined format such as a regular expression. The page function is a name indicating the function of the WEB page whose URL matches the URL pattern.

  The function analysis unit 122 matches the URL with the URL pattern of the function analysis rule 113 for each row of the tree structure data 112, and sets the page function corresponding to the matched URL pattern as the page function of the tree structure data 112. As a result, as shown in the example of FIG. 6, a value is set for the page function of the tree structure data 112, and the function of each node included in the tree structure data 112 is specified as shown in the example of FIG. It becomes.

  Note that a default URL pattern that matches a URL when it does not match any other URL pattern may be provided in the function analysis rule 113 so that the URL of the WEB page does not match any URL pattern. Alternatively, when the URL of the WEB page does not match any URL pattern, the function of the WEB page may be specified by a default rule embedded in the function analysis unit 122. In addition, the function of the WEB page is not specified by matching the URL of the WEB page and the URL pattern, but an HTML file that is the actual state of the WEB page is acquired based on the URL of the WEB page, and the acquired HTML file is The function of the WEB page may be specified by analyzing whether a predetermined keyword is included.

  Next, a specific example of processing executed by the behavior estimation unit 123 will be described with reference to FIGS. FIG. 8 is a diagram illustrating an example of the behavior estimation rule 114. As shown in the example of FIG. 8, the behavior estimation rule 114 has items such as a subtree pattern formula and user behavior. The subtree pattern expression is an expression representing a subtree in which nodes having specific functions are combined in a specific structure. The user behavior is a user behavior estimated when the subtree represented by the subtree pattern formula is included in the tree structure data 112.

Here, the format of the subtree pattern formula in this embodiment will be described. The basic format of the subtree pattern formula in this embodiment is as follows.
Parent node function-(list of child node functions)

  “Product details— (review, store information)”, which is the subtree pattern expression in the first row of the behavior estimation rule 114 shown in FIG. 8, is the parent node of the function “product details” as shown in FIG. This represents a subtree in which a node with a function “review” and a node with a function “store information” exist as child nodes. When such a partial tree is included in the behavior estimation rule 114, the behavior estimation unit 123 is considering “at which store to buy,” set as the user behavior on the first line of the behavior estimation rule 114. Is output as an estimation result of the user's behavior.

  Note that the parent-child structure of the node can be nested in the part of the function list of the child node. As shown in FIG. 10, “site top- (store top-store information)”, which is the partial tree pattern expression in the second row of the behavior estimation rule 114 shown in FIG. This represents a subtree in which a node having a function “shop top” exists as a parent node, and a node having a function “site map” exists as a parent node of the node having a function “shop top”. When such a partial tree is included in the behavior estimation rule 114, the behavior estimation unit 123 determines “what kind of store is present” set as the user behavior on the second line of the behavior estimation rule 114. Output as a result of user behavior estimation.

  Next, a processing procedure of behavior estimation processing executed by the behavior estimation device 10 shown in FIG. 1 will be described with reference to the flowchart of FIG. As shown in FIG. 11, when an estimation of a user's behavior is requested, the tree structure generation unit 121 acquires the user information from the reference history data 111 and generates the tree structure data 112 (steps). S101).

  Subsequently, the function analysis unit 122 identifies the function of each node included in the tree structure data 112 (step S102). Then, the behavior estimation unit 123 collates the subtree included in the behavior estimation rule 114 with the tree structure data 112, and outputs a user behavior estimation result (step S103).

  As described above, the behavior estimation apparatus 10 according to the present embodiment generates a tree structure indicating transition of reference from the history information of the WEB page referred to by the user, and the user's behavior based on the generated tree structure. Therefore, a detailed and accurate estimation result can be obtained.

  In the present embodiment, a modified example of the behavior estimation apparatus 10 described in the first embodiment is shown. First, the configuration of the behavior estimation apparatus 20 according to the present embodiment will be described with reference to FIG. As illustrated in FIG. 12, the behavior estimation apparatus 20 includes a storage unit 210 and a control unit 220.

  The storage unit 210 is, for example, a hard disk device or a semiconductor storage device, and stores reference history data 211, tree structure data 212, function analysis rules 213, subtree extraction rules 214, and behavior estimation rules 215.

  The partial tree extraction rule 214 is information for extracting a partial tree having a specific structure from the tree structure data 212. The reference history data 211, the tree structure data 212, the function analysis rule 213, and the behavior estimation rule 215 are respectively equivalent to the reference history data 111, the tree structure data 112, the function analysis rule 113, and the behavior estimation rule 114 shown in the first embodiment. To do.

  The function analysis rule 213 is edited by the function analysis rule input device 1, the subtree extraction rule 214 is edited by the subtree extraction rule input device 2, and the behavior estimation rule 215 is edited by the behavior estimation rule input device 3. . The function analysis rule input device 1, the subtree extraction rule input device 2, and the behavior estimation rule input device 3 may be configured integrally with the behavior estimation device 20.

  Note that part or all of the data illustrated as being stored in the storage unit 210 in FIG. 12 can be accessed from the behavior estimation device 20 to other devices configured to be communicable with the behavior estimation device 20. It may be stored in a state.

  The control unit 220 is a control unit that controls the entire behavior estimation apparatus 20, and includes a tree structure generation unit 221, a function analysis unit 222, and a behavior estimation unit 223. The tree structure generation unit 221, the function analysis unit 222, and the behavior estimation unit 223 correspond to the tree structure generation unit 121, the function analysis unit 122, and the behavior estimation unit 123 described in the first embodiment, respectively.

  The behavior estimation unit 223 includes a partial tree extraction unit 223a and a partial tree matching unit 223b. The partial tree extraction unit 223 a extracts a partial tree having a specific structure from the tree structure data 212 based on the partial tree extraction rule 214. The partial tree matching unit 223b estimates the user's behavior by matching the partial tree extracted by the partial tree extracting unit 223a with the behavior estimation rule 215.

  In this way, by extracting a subtree having a specific structure and collating it with the behavior estimation rule 215, even if the tree structure included in the tree structure data 212 is large, the subtree included in the behavior estimation rule 215 The part to be collated can be narrowed down in advance, and the time required for collation is shortened.

  Next, processing executed by each unit included in the control unit 220 illustrated in FIG. 12 will be described in more detail.

  A specific example of the processing executed by the tree structure generation unit 221 will be described with reference to FIGS. The tree structure generation unit 221 is different from the tree structure generation unit 121 shown in the first embodiment in the operation when the WEB page indicated by the data read from the reference history data 211 has already created a node.

  Specifically, even if the WEB page indicated by the data read from the reference history data 211 has already created a node, the tree structure generation unit 221 has read the node corresponding to the WEB page immediately before If there is no path on the tree structure from the node corresponding to the WEB page indicated by the data to the root node, a node corresponding to the WEB page indicated by the read data is generated, and the generated node is read immediately before A node corresponding to the WEB page indicated by the data is associated as a child node.

  By such processing, a series of transitions of a WEB page referred to by a user for a specific purpose can be made into a different tree structure from a series of transitions of a WEB page previously referred to for another purpose. , The estimation accuracy of the user's behavior is improved.

  FIG. 13 is a diagram illustrating an example of the reference history data 211. As illustrated in the example of FIG. 13, the reference history data 211 includes an item called a history number in addition to the items included in the reference history data 111 illustrated in FIG. 2. The history number is an identifier for identifying each row of the reference history data 211.

  FIG. 14 is a diagram illustrating an example of the tree structure data 212. As shown in the example of FIG. 14, the tree structure data 212 has an item of a history number list in addition to the items of the tree structure data 112 shown in FIG. The history number list is a list of history numbers indicating rows of the reference history data 211 corresponding to the nodes.

  FIG. 15 is a flowchart illustrating a processing procedure of tree structure generation processing executed by the tree structure generation unit 221. The processing procedure shown in FIG. 15 is based on the MF algorithm (see Non-Patent Document 1), but may be a processing procedure using another algorithm as long as the same result can be achieved.

  As illustrated in FIG. 15, the tree structure generation unit 221 first initializes “a”, which is a variable indicating the row of the reference history data 211 to be processed, to a value (for example, “1”) indicating the first row. (Step S201). Subsequently, the tree structure generation unit 221 initializes n, which is a variable representing the node ID of the next node to be created, to a value representing the first node (for example, “N01”) (step S202).

  In addition, the tree structure generation unit 221 empties the list N and the list Y (step S203). The list N is a list of node IDs of nodes existing on the path traced from the root node to the last added node. The list Y is a list of row numbers of the rows of the reference history data 211 corresponding to the nodes included in the list N.

  Then, the tree structure generation unit 221 acquires a row corresponding to the variable a from the reference history data 211 (Step S204). Here, if the number of the line including the same URL as the acquired line is not included in the list Y (No in step S205), the tree structure generation unit 221 adds a line to the tree structure data 212 and performs the following process. A value is set to (step S206). That is, the user ID, URL, and history number of the acquired row are set in the user ID, URL, and history number list, the value of the variable n is set as the node ID, and the parent node ID is set as the parent node ID. , The value of the last element of the list N is set.

  Subsequently, the tree structure generation unit 221 adds an element to the end of the list Y and sets the value of the variable a (step S207), adds an element to the end of the list N and sets the value of the variable n ( In step S208, the variable n is updated to a value representing the next node (step S209). Then, the tree structure generation unit 221 adds 1 to the variable a (step S214), and if there is a row corresponding to the variable a in the reference history data 211 (Yes in step S215), the processing procedure is restarted from step S204.

  On the other hand, if the list Y includes a line number including the same URL as the line acquired from the reference history data 211 (Yes in step S205), the tree structure generation unit 221 uses the line number including the same URL. The position of the included element in the list Y is set to the variable j (step S210). Then, the tree structure generation unit 221 deletes elements after the position where 1 is added to the value of the variable j from the list Y and the list N (step S211).

  Subsequently, the tree structure generation unit 221 acquires a value included in the last element of the list N (step S212), and the history number list in the row of the tree structure data 212 in which the value is set as the node ID, The history number of the row of the reference history data 211 corresponding to the variable a is added (step S213). Then, the tree structure generation unit 221 adds 1 to the variable a (step S214), and if there is a row corresponding to the variable a in the reference history data 211 (Yes in step S215), the processing procedure is restarted from step S204.

  The above processing procedure is repeatedly executed until all the rows are acquired from the reference history data 211, and if there is no row corresponding to the variable a in the reference history data 211 (No in step S215), the tree structure generation unit 221 The tree structure generation process ends.

  FIG. 16A shows the state of various data at the stage when step S203 is completed. The tree structure data 212, the list Y and the list N are empty, and the variable a and the variable n are “1”, respectively. ”And“ N01 ”.

  FIG. 16-2 shows the state of various data in which the first row is acquired from the reference history data 211 and steps S206 to S209 and step S214 are executed. A row is added to the tree structure data 212. Elements are added to the list Y and the list N. Further, the values of the variable a and the variable n are updated.

  FIG. 16C shows the state of various data in which the second row is acquired from the reference history data 211 and steps S206 to S209 and S214 are executed. Further rows are added to the tree structure data 212. Further, elements are added to the list Y and the list N. Further, the values of the variable a and the variable n are updated.

  FIG. 16-4 shows the state of various data in which the third line including the same URL as the first line is acquired from the reference history data 211 and steps S210 to S214 are executed. The history numbers of the list Y and the list N are deleted, and the elements after the element corresponding to the first line of the reference history data 211 are deleted. The value of variable a is updated.

  FIG. 16-5 shows the state of various data in which the first row is acquired from the reference history data 211 and steps S206 to S209 and step S214 are executed. A row is added to the tree structure data 212. Elements are added to the list Y and the list N. Further, the values of the variable a and the variable n are updated. In the stage shown in FIG. 16D, since the element is deleted from the list Y and the list N, the parent node ID of the row added to the tree structure data 212 is the node ID of the node added immediately before. Instead of “N02”, “N01” which is the node ID of the node corresponding to the returned WEB page is set.

  Next, a specific example of processing executed by the function analysis unit 222 will be described with reference to FIGS. For each row of the tree structure data 212, the function analysis unit 222 matches the URL with the URL pattern of the function analysis rule 213 similar to the function analysis rule 113 shown in FIG. 5, and displays a page function corresponding to the matched URL pattern. The page function of the tree structure data 212 is set. As a result, as shown in the example of FIG. 17, a value is set in the page function of the tree structure data 212.

  FIG. 18 is a flowchart illustrating the processing procedure of the function analysis process of the function analysis unit 222. As shown in FIG. 18, the function analysis unit 222 selects the first row of the tree structure data 212 (step S301). When the row of the tree structure data 212 can be selected (Yes at Step S302), the function analysis unit 222 selects the first row of the function analysis rule 213 (Step S303).

  Subsequently, when the line of the function analysis rule 213 can be selected (Yes at Step S304), the function analysis unit 222 uses the URL of the line selected from the tree structure data 212 as the URL pattern of the line selected from the function analysis rule 213. (Step S305).

  If the URL matches the URL pattern (Yes at Step S306), the function analysis unit 222 sets the page function value of the row selected from the function analysis rule 213 to the row selected from the tree structure data 212 (Step S306). In step S307, the next row of the tree structure data 212 is selected (step S309), and the processing procedure is restarted from step S302.

  On the other hand, if the URL does not match the URL pattern (No at Step S306), the function analysis unit 222 selects the next line from the function analysis rule 213 (Step S308), and restarts the processing procedure from Step S304. If there is no line to be selected in the function analysis rule 213 (No at Step S304), the function analysis unit 222 selects the next line of the tree structure data 212 (Step S309), and the processing from Step S302 is performed. Resume procedure.

  The above processing procedure is repeatedly executed until all the rows of the tree structure data 212 are selected. When there are no more rows to be selected (No at Step S302), the function analysis unit 222 ends the processing procedure.

  Next, a specific example of processing executed by the partial tree extraction unit 223a will be described with reference to FIGS.

  FIG. 19 is a diagram illustrating an example of the subtree extraction rule 214. As shown in the example of FIG. 19, the subtree extraction rule 214 includes items such as a root node selection condition, a descendant node selection condition, and a pattern format. The root node selection condition is a condition for selecting the root node of the subtree. The descendant node selection condition is a condition for selecting a descendant node of the root node. The pattern format is the format of the output format of the selected subtree. The subtree extraction rule 214 is set in advance so that subtrees that may match the subtree included in the behavior estimation rule 215 are extracted from the tree structure data 212 without omission and with as few nodes as possible. .

  The subtree extraction unit 223a extracts all subtrees satisfying the root node selection condition and the descendant node selection conditions from the tree structure data 212, and outputs them as subtree extraction data according to the pattern format. As shown in the example of FIG. 19, the root node selection condition is “having two or more children”, the descendant node selection condition is “the depth from the root node is 1 or less”, and the pattern format is “root node” Page function- (list of page functions of child nodes), but the duplication of page functions of child nodes is removed ", the subtree extracting unit 223a uses the tree structure data 212 shown in FIG. Extract subtree extraction data as shown in FIG.

  As shown in the example of FIG. 20, the partial tree extraction data includes items such as a user ID, a node ID, a history number list, and a partial tree pattern expression, and a row is added for each extracted partial tree. The user ID is an identifier for identifying the user. The node ID is an identifier for identifying the root node of the extracted subtree. The history number list is a list of identifiers of rows of the reference history data 211 corresponding to each node of the extracted subtree. The subtree pattern expression is a subtree output according to the pattern format.

  FIG. 21 is a flowchart illustrating the processing procedure of the partial tree extraction process of the partial tree extraction unit 223a. As shown in FIG. 21, the subtree extraction unit 223a first initializes a, which is a variable indicating a row of the tree structure data 212 to be processed, to a value (for example, “1”) indicating the first row. (Step S401). Subsequently, the partial tree extraction unit 223a selects a row of the tree structure data 212 corresponding to the variable a (Step S402).

  If the row of the tree structure data 212 corresponding to the variable a can be selected (Yes at Step S403), the subtree extraction unit 223a verifies whether the node indicated by the selected row satisfies the root node selection condition ( Step S404). If the root node selection condition is not satisfied (No at Step S405), the subtree extraction unit 223a adds 1 to the variable a (Step S414), and restarts the processing procedure from Step S402.

  On the other hand, when the root node selection condition is satisfied (Yes at Step S405), the subtree extraction unit 223a empties L, which is a list of row numbers of child nodes (Step S406). Subsequently, the partial tree extraction unit 223a selects one of the rows in which the node ID of the row of the tree structure data 212 corresponding to the variable a is set as the parent node ID from the tree structure data 212 (step S407).

  If a row can be selected (Yes at step S408), the subtree extraction unit 223a determines whether the node indicated by the selected row (or the node indicated by the selected row and its descendant node) satisfies the descendant node selection condition. Verification is performed (step S409). If the node indicated by the selected row (or the node indicated by the selected row and its descendant node) satisfies the descendant node selection condition (Yes at step S410), the subtree extraction unit 223a selects the row of the selected row. The number is added to the list L (step S411), and the process proceeds to step S412. On the other hand, when the node indicated by the selected row (or the node indicated by the selected row and its descendant node) does not satisfy the descendant node selection condition (No at step S410), the subtree extracting unit 223a selects the row of the selected row. Without adding the number to the list L, the process proceeds to step S412.

  In step S412, the subtree extraction unit 223a selects from the tree structure data 212 another row in which the node ID of the row of the tree structure data 212 corresponding to the variable a is set as the parent node ID (step S412). The processing procedure is restarted from step S408.

  When all the rows in which the node ID of the row of the tree structure data 212 corresponding to the variable a is set as the parent node ID have been selected from the tree structure data 212 (No at step S408), the subtree extraction unit 223a converts the page function of the row of the tree structure data 212 corresponding to the variable a and the page function of the row of the tree structure data 212 corresponding to each element of the list L into the partial tree extraction data in a format based on the pattern format. Output (step S413). Subsequently, the partial tree extraction unit 223a adds 1 to the variable a (step S414), and restarts the processing procedure from step S402.

  The above processing procedure is repeatedly executed until all the rows of the tree structure data 212 are selected. When there are no more rows to be selected (No at Step S403), the subtree extraction unit 223a ends the processing procedure.

  Next, a specific example of processing executed by the subtree matching unit 223b will be described with reference to FIGS. Although an example of a method for outputting matching user behavior based on a reference table is shown here, matching user behavior can also be output using a prediction method such as machine learning.

  FIG. 22 is a diagram illustrating an example of the behavior estimation rule 215. As shown in the example of FIG. 22, the behavior estimation rule 215 has a score set for each of a plurality of types of user behaviors and for each subtree pattern expression. The subtree matching unit 223b selects, from the behavior estimation rule 215, a row in which a subtree pattern expression that matches each subtree extracted by the subtree extraction unit 223a is selected, and the highest score is set in that row. Are output as an estimation result of the behavior indicated by the partial tree extracted by the partial tree extraction unit 223a. Note that the match between each subtree extracted by the subtree extraction unit 223a and the subtree pattern expression is not necessarily a complete match, and may be a partial match.

  For example, the subtree represented by the subtree pattern expression “site map— (ranking, search result, review)” in the first line of the subtree extraction data shown in FIG. 20 is the behavior estimation rule shown in FIG. In the second line of the behavior estimation rule 215, which matches the subtree pattern formula in the second row of 215, the score of the user action “considering which store to buy” is the highest, so the subtree matching unit 223 b Outputs “being considered at which store” as an estimation result of the behavior indicated by the partial tree in the first row of the partial tree extraction data.

  Further, the subtree represented by the subtree pattern expression “search result— (product details)” in the second row of the subtree extraction data shown in FIG. 20 is one row of the behavior estimation rule 215 shown in FIG. In the first line of the behavior estimation rule 215, which matches the subtree pattern expression of the eye, the subtree matching unit 223 b performs subtree extraction because the user action score “selecting which product to buy” is the highest. “Selecting which product to buy” is output as an estimation result of the behavior indicated by the subtree in the second row of data.

  FIG. 23 is a flowchart showing a processing procedure of the subtree matching process of the subtree matching unit 223b. As shown in FIG. 23, the subtree matching unit 223b first initializes a, which is a variable indicating the row of the subtree extraction data to be processed, to a value (for example, “1”) representing the first row. (Step S501). Subsequently, the subtree matching unit 223b selects a row of the subtree extraction data corresponding to the variable a (Step S502).

  If the row of the subtree extraction data corresponding to the variable a can be selected (Yes at Step S503), the subtree matching unit 223b selects a row that matches the subtree pattern formula from the behavior estimation rule 215 ( Step S504). If there is a corresponding row (Yes at Step S505), the subtree matching unit 223b outputs the user action having the largest numerical value as the estimation result (Step S506), and adds 1 to the variable a. (Step S507), the processing procedure is restarted from Step S502.

  On the other hand, when there is no corresponding row (No at Step S505), the subtree matching unit 223b adds 1 to the variable a without outputting anything (Step S507), and restarts the processing procedure from Step S502. . In this case, nothing may be output, and it may be output that the estimated action is not found, such as “not applicable”. Or it is good also as outputting a user action with the largest value which totaled the numerical value of each user action of action presumption rule 215 for every row (each user action) as an estimation result.

  The above processing procedure is repeatedly executed until all the rows of the subtree extraction data are selected. When there are no more rows to be selected (No in step S503), the subtree matching unit 223b ends the processing procedure.

  In addition, it is good also as outputting only one estimation result instead of outputting an estimation result for every partial tree extracted by the partial tree extraction part 223a. FIG. 24 is a flowchart illustrating a processing procedure when only one estimation result is output. As shown in FIG. 24, the subtree matching unit 223b first initializes a, which is a variable indicating the row of the subtree extraction data to be processed, to a value (for example, “1”) representing the first row. (Step S601).

  Subsequently, the subtree matching unit 223b initializes a variable corresponding to each user behavior of the behavior estimation rule 215 to 0 (step S602). Then, the subtree matching unit 223b selects a row of the subtree extraction data corresponding to the variable a (Step S603).

  If the row of the subtree extraction data corresponding to the variable a can be selected (Yes at step S604), the subtree matching unit 223b selects a row that matches the subtree pattern expression from the behavior estimation rule 215 ( Step S605). If there is a corresponding row (Yes at Step S606), the subtree matching unit 223b adds the numerical value of each user action in the row to the corresponding variable (Step S607), and adds 1 to the variable a. (Step S608), the processing procedure is restarted from Step S603.

  On the other hand, if there is no corresponding row (No at Step S606), the subtree matching unit 223b adds 1 to the variable a without adding anything to the variable (Step S608), and the processing procedure starts from Step S603. Resume.

  The above processing procedure is repeatedly executed until all rows of the subtree extraction data are selected, and if there are no more rows to be selected (No in step S604), the subtree matching unit 223b The user action corresponding to the variable with the largest numerical value is output as an estimation result (step S609), and the processing procedure is terminated. In this method, for example, when the partial tree extraction data is as shown in FIG. 20 and the behavior estimation rule 215 is as shown in FIG. 22, it corresponds to the user behavior of “selecting which product to buy”. The value of the variable is 1 + 10 = 11, and the value of the variable corresponding to the user action “considering at which store to buy” is 3 + 00 = 3, and the user is “looking for what you want” The value of the variable corresponding to the action is 0 + 2 = 2, and the user action corresponding to the variable having the largest value is “selecting which product to buy” is output as the estimation result.

  Here, an example of how to set the behavior estimation rule 215 will be described. In the example shown here, as shown in FIG. 25, user response data is prepared in advance by adding the result of listening to the behavior of the user or the expert from each row of the reference history data as a user action. In addition, it is assumed that each numerical value of the behavior estimation rule 215 is initialized to 0.

  As shown in FIG. 26, the behavior estimation rule input device 3 generates partial tree extraction data from user response data by the method already described (step S701). Then, the behavior estimation rule input device 3 selects the top row of the user response data (step S702).

  When a row can be selected (Yes at Step S703), the behavior estimation rule input device 3 selects all rows including the same history number in the history number list from the partial tree extraction data (Step S704). Then, the behavior estimation rule 215 selects a subtree pattern expression that matches that of the row selected from the subtree extraction data (step S705), and selects the user behavior value of each selected row. The value of the user response data corresponding to the user action is increased by 1 (step S706).

  Then, the behavior estimation rule input device 3 selects the next line of the user answer data (step S707), and restarts the processing procedure from step S703. In this way, when all the rows of the user answer data have been selected and there are no more rows to be selected (No in step S703), the behavior estimation rule input device 3 ends the processing procedure.

  As shown in FIG. 27, the subtree extraction rule 214 can be set so as to extract subtrees of three or more layers. When the subtree extraction rule 214 is set as shown in FIG. 27, the subtree matching unit 223b outputs an estimation result as shown in FIG.

  By the way, the structure of the action estimation apparatus in each Example mentioned above can be changed in the range which does not deviate from the summary of invention, and can be implement | achieved combining in the range which does not deviate from the summary of invention.

  For example, as shown in FIG. 29, the function analysis unit 222 can be configured so that two or more page functions can be specified for each row of the tree structure data 212. In this case, the combination of the subtree extraction rule 214 and the estimation result may be such that all the page functions are listed as shown in FIG. 30 and FIG. 31, or as shown in FIG. 32 and FIG. The two or more page functions may be output as different subtree pattern expressions.

  In the second embodiment, after the function analysis unit 222 identifies the function of the node included in the tree structure data 212, the partial tree extraction unit 223a extracts a partial tree having a specific structure from the tree structure data 212. However, the partial tree extraction unit 223a may extract a partial tree having a specific structure from the tree structure data 212, and then the function analysis unit 222 may specify the function of the node. In this case, for example, the subtree extraction unit 223a extracts a subtree pattern expression in which the URL is formatted according to the pattern format, and the function analysis unit 222 replaces the URL in the subtree pattern expression with the page function, and is duplicated after the replacement. If there is a subtree pattern expression to be deleted, leave one and delete it.

  In addition, the functions of the control unit 120 of the behavior estimation device 10 and the control unit 220 of the behavior estimation device 20 are implemented as software, and the same functions as the behavior estimation device 10 and the behavior estimation device 20 are executed by a computer. It can also be realized. Below, an example of the computer which performs the action estimation program 1071 which mounted the function of the control part 120 as software is shown.

  FIG. 34 is a functional block diagram showing a computer 1000 that executes the behavior estimation program 1071. The computer 1000 includes a CPU (Central Processing Unit) 1010 that executes various arithmetic processes, an input device 1020 that receives input of data from a user, a monitor 1030 that displays various information, and a medium that reads a program from a recording medium. A reading device 1040, a network interface device 1050 that exchanges data with other computers via a network, a RAM (Random Access Memory) 1060 that temporarily stores various information, and a hard disk device 1070 are connected via a bus. Configured.

  The hard disk device 1070 includes an action estimation program 1071 having the same function as that of the control unit 120 shown in FIG. 1 and action estimation data 1072 corresponding to various data stored in the storage unit 110 shown in FIG. Is memorized. Note that the behavior estimation data 1072 may be appropriately distributed and stored in another computer connected via a network.

  Then, the CPU 1010 reads out the behavior estimation program 1071 from the hard disk device 1070 and develops it in the RAM 1060, whereby the behavior estimation program 1071 functions as the behavior estimation process 1061. Then, the behavior estimation process 1061 appropriately expands information read from the behavior estimation data 1072 in an area allocated to itself on the RAM 1060, and executes various data processing based on the expanded data.

  Note that the behavior estimation program 1071 is not necessarily stored in the hard disk device 1070, and the computer 1000 may read and execute the program stored in a storage medium such as a CD-ROM. . The computer 1000 stores the program in another computer (or server) connected to the computer 1000 via a public line, the Internet, a LAN (Local Area Network), a WAN (Wide Area Network), or the like. You may make it read and run a program from these.

  In the present invention, for example, when the user behavior is “looking for something to want”, a product that matches the user's preference profile is presented, and the user behavior is “selecting which product to buy”. In this case, it can be used for various site navigations such as presenting a product similar to the product that the user is currently browsing. By using the present invention for site navigation at an online shopping site, for example, the purchase probability can be increased as compared with a system that simply recommends based on user preferences or presents similar products.

DESCRIPTION OF SYMBOLS 1 Function analysis rule input device 2 Subtree extraction rule input device 3 Behavior estimation rule input device 10 Behavior estimation device 110 Memory | storage part 111 Reference history data 112 Tree structure data 113 Function analysis rule 114 Action estimation rule 120 Control part 121 Tree structure generation part 122 function analysis unit 123 behavior estimation unit 20 behavior estimation device 210 storage unit 211 reference history data 212 tree structure data 213 function analysis rule 214 subtree extraction rule 215 behavior estimation rule 220 control unit 221 tree structure generation unit 222 function analysis unit 223 behavior Estimation unit 223a Subtree extraction unit 223b Subtree verification unit 1000 Computer 1010 CPU
1020 Input device 1030 Monitor 1040 Medium reader 1050 Network interface device 1060 RAM
1061 Action estimation process 1070 Hard disk device 1071 Action estimation program 1072 Action estimation data

Claims (5)

Storage means for storing reference history data indicating an order in which a user refers to a plurality of WEB pages;
Based on the reference history data, tree-structured data having each WEB page referred to by the user as a node, and corresponding to a second WEB page referenced next to the first WEB page Tree structure generation means for generating tree structure data in which each node is associated with each other so that the node becomes a child node of the first node corresponding to the first WEB page;
Function analysis means for analyzing the function of the WEB page corresponding to each node included in the tree structure data generated by the tree structure generation means;
Based on the situation in which each node included in the tree structure data generated by the tree structure generation unit is associated and the analysis result of the function of the WEB page corresponding to each node by the function analysis unit, the user A behavior estimation device comprising: behavior estimation means for estimating the behavior of the behavior.   The behavior estimation unit extracts a subtree of a predetermined pattern from the tree structure data after the function analysis unit analyzes the function of the WEB page corresponding to each node, and each node included in the extracted subtree 2. The behavior according to claim 1, wherein the behavior of the user is estimated based on a situation in which the user is associated and a result of analysis of a function of a WEB page corresponding to each node by the function analysis unit. Estimating device.   The behavior estimation means is configured to analyze an action designation rule in which a user's action is defined for each subtree in which the function of each node is specified, and after the function of the WEB page corresponding to each node is analyzed by the function analysis means. The behavior estimation apparatus according to claim 1, wherein the behavior of the user is estimated by collating with the tree structure data. A storage step of storing reference history data indicating an order in which a user refers to a plurality of WEB pages;
Based on the reference history data, tree-structured data having each WEB page referred to by the user as a node, and corresponding to a second WEB page referenced next to the first WEB page A tree structure generation step of generating tree structure data in which each node is associated with each other so that the node becomes a child node of the first node corresponding to the first WEB page;
A function analysis step of analyzing the function of the WEB page corresponding to each node included in the tree structure data generated in the tree structure generation step;
Based on the situation in which each node included in the tree structure data generated in the tree structure generation step is associated, and the analysis result of the function of the WEB page corresponding to each node in the function analysis step, the user An action estimation method comprising: an action estimation step for estimating an action of   The computer program for functioning a computer as an action estimation apparatus as described in any one of Claims 1 thru | or 3.

Images