JP3705330B2 - Hypertext structure change support device and method, and storage medium storing hypertext structure change support program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention provides support for changing a hypertext structure based on the degree of similarity between a user's access tendency and content distribution tendency with respect to content (hypertext information) published on the website in a website that provides information on the Internet. The present invention relates to an apparatus and method, and a storage medium recording a support program for realizing the support apparatus by a computer.
[0002]
[Prior art]
In recent years, there has been an increasing demand for providing information on the World Wide Web (hereinafter abbreviated as Web). Businesses also have Web sites for providing information on the Web, and buying and selling of goods and services by means of advertising activities of their products or means such as electronic commerce are being actively performed on the Web sites. Under such circumstances, there is an increasing demand for technology for knowing the trends of users who have accessed the Web and quickly changing the structure of information to be provided in accordance with the user trends.
[0003]
Information provided on a website is called content and is expressed in hypertext (for example, HTML). Individual hypertext information (hereinafter referred to as nodes) is connected by links. A user who wants to access a website can access a node in the website by following a link and obtain various information. The Web server can record a history of accessing the Web server by the user. The recorded content is hereinafter referred to as a log. For example, the log includes the IP address of the computer that has accessed the information (the computer that the user is using to access the Web server, or the computer that is relaying the user's usage information), the time of access, and the information A command sent to the Web server for access is recorded. The command includes an identifier (for example, URL) on the server of the node accessed by the user.
[0004]
As a conventional technique for knowing a user's trend using an access log, for example, Ming-Syan et al., “Data Mining for Path Traversal Patterns on the Web”, IEEE Trans. on Knowledge and Data Engineering, 1998, and the like. The technique described in this document is a technique for taking out a series called MF (Maximum Forward references). FIG. 15 is an explanatory diagram of an example of a Web site structure and a user access route. In the example shown in FIG. 15, it is assumed that the nodes A to H are linked as illustrated. At this time, it is assumed that a certain user browses nodes in the order of A → B → C → D → E → F → G as indicated by the bold line in the figure. In this case, MFs [ABCD], [ABE], and [AFG] are obtained. The user's access pattern can be obtained by obtaining the MF for all users who have accessed and counting the number of appearances of the MF subsequence.
[0005]
With this technique, it is possible to obtain a linear sequence of nodes that a user has followed a link. However, the sequence in which the user has traced back from the node returned from a certain node is obtained as a different sequence. Therefore, it is difficult to obtain a group of nodes accessed by the user as a lump.
[0006]
As another conventional technique, for example, J. Org. Borges et al., “Minning Association Rules in Hypertext Databases”, In Proc. There is a technique described in of KDD-98. In this technique, the number of user transitions between nodes connected by a link is extracted from a log, and a correlation rule for a node sequence accessed by the user is obtained. In this technique, since only the number of user transitions between nodes is observed, the same user does not always follow the sequence. FIG. 16 is an explanatory diagram of an example of the number of user transitions between the website structure and the nodes. In the figure, the direction following the link is indicated by an arrow line, and the number of times the link is followed is indicated by a numerical value. According to this prior art, in the example shown in FIG. 16, for example, a correlation that a user who follows a link of B → A tends to follow a link of A → F is obtained. However, this does not indicate that the same user is visiting all of the nodes B, A, and F, and cannot accurately reflect the access trend of the user.
[0007]
In addition, in the above two conventional technologies, it is possible to know the frequent access route by obtaining the user access pattern from the log, but since the processing is not integrated with the content, the access pattern is based on the access pattern. It has been difficult to provide support related to the maintenance and management of Web sites, such as which content should be changed.
[0008]
[Problems to be solved by the invention]
The present invention has been made in view of the above-described circumstances. By comparing the user's access tendency and the similarity between content distribution trends, the user's trend is grasped and the user's trend is matched. It is an object of the present invention to provide a hypertext structure change support device and method capable of supporting the change of Web site content, and a storage medium storing a support program for realizing the support device on a computer. .
[0009]
[Means for Solving the Problems]
The present invention generates a hierarchical log cluster from a hypertext set based on access frequency information by a user to a hypertext information group managed by an information system, and the degree of similarity between individual hypertexts in the hypertext information group. The content cluster is generated from the hypertext set based on the similarity, the partial structure that cannot be matched between the log cluster and the content cluster is specified, and the elements in the content cluster that could not be matched It presents hypertext. If the log cluster cannot be associated with the content cluster, for example, the frequency of reference from the user is clearly different for hypertext having the same content, the website structure is inappropriate, or There is a possibility that the link is not provided properly. By presenting such hypertext, a change in the structure of the hypertext can be promoted and the structure change can be supported.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 is a block diagram showing an embodiment of the hypertext structure change support device of the present invention. In the figure, 1 is a Web server, 2 is a log clustering unit, 3 is a content clustering unit, 4 is an intentional error detecting unit, 5 is an intentional error presenting unit, 11 is a log recording unit, and 12 is a content providing unit.
[0011]
The Web server 1 stores information (content) that the user wants to provide to the user and transmits the information on the network. The Web server 1 includes a log recording unit 11 and a content providing unit 12. The content providing unit 12 provides content according to user access. In addition, each time the user accesses the log recording unit 11, the user identifier (IP address) for identifying the user, the access time, and the hypertext accessed by the user (hereinafter referred to as a node) Record (URL). The IP address recorded by the Web server 1 may be the address of a proxy server acting as a proxy for access from the client computer in addition to the client computer actually used by the user. In the latter case, a plurality of users use the same IP address. In the following description, it is assumed that the user is identified by the IP address and date of the computer that has accessed the Web server 1. That is, even if accesses are made from the same IP address, if the access dates are different, they are identified as different users.
[0012]
The log used in the present invention is not limited to the log described above, and may be a log identifying a user by using a proxy server log as described in, for example, Japanese Patent Laid-Open No. 10-224349. Further, for example, even a log recorded by a mechanism having an access notification mechanism on the client side such as a Java applet described in Japanese Patent Laid-Open No. 10-207838 can be identified by the user, the location of the accessed information, and the access order Anything that can be understood.
[0013]
The log clustering unit 2 acquires a link structure formed by nodes and links and a log from the Web server 1, and clusters the nodes based on the log to form a log cluster. Details of the log clustering unit 2 will be described later.
[0014]
The content clustering unit 3 acquires content from the Web server 1 and performs clustering based on, for example, content similarity to form a content cluster. The content cluster is formed without referring to the link structure.
[0015]
The intention shift detection unit 4 associates the content cluster obtained by the content clustering unit 3 with the log cluster obtained by the log clustering unit 2. At this time, the content cluster that could not be matched is detected as a possibility that the intention is shifted.
[0016]
The intention disagreement presentation unit 5 displays the hypertext, which is an element in the content cluster, detected as a possibility that the intention disagreement is not possible in the intention disagreement detection unit 4 without being associated with each other. To present.
[0017]
FIG. 2 is a block diagram showing an example of a log clustering unit in the embodiment of the hypertext structure change support device of the present invention. In the figure, 21 is an access information storage unit, 22 is an access number counting unit, 23 is a cluster configuration unit, 24 is a cluster generation acceleration unit, and 25 is a cluster generation restriction unit.
[0018]
The access information storage unit 21 stores an identifier and information in which the nodes accessed by the user having the identifier are recorded in chronological order from the log information. The identifier here is an identifier that is identified and assigned as a different user when the access date is different even when accessing from the same IP address as described above.
[0019]
The access number counting unit 22 calculates the number of identifiers having a history of access that satisfies the order constraint given to all given hypertext sets using the information stored in the access information storage unit 21.
[0020]
The cluster configuration unit 23 has a cluster composed only of a predetermined starting node as a vertex, and nodes that are reachable by following one link from the starting node and that are extracted one by one from the node set from which logs are acquired A cluster set consisting of a set of starting nodes is generated as a new cluster candidate one layer below. When it is calculated that the generated cluster candidate has an access count equal to or greater than a threshold given in advance by the access count counter 22, a new cluster is generated as a child cluster of the starting node. Each of the generated child clusters is also included in the node set that can be reached by following one link from the node set included in the child cluster, and the log has been acquired and is not included in the child cluster. Are extracted one by one and added to the node set included in the child cluster to generate a cluster candidate at a lower hierarchy. Furthermore, when it is calculated that the cluster candidate has the number of accesses equal to or greater than a predetermined threshold value, the process of generating a new child cluster with the child cluster as the parent cluster is performed until no new child cluster is generated. repeat. In this way, a hierarchical cluster is generated.
[0021]
The cluster generation acceleration unit 24, when generating a hierarchical cluster in the cluster configuration unit 23, in a higher-level cluster having a subset of node sets included in a newly generated cluster candidate, The cluster candidate is not generated when the number of accesses from the user is equal to or less than a predetermined threshold. Thereby, it is possible to shorten the processing time by stopping the cluster generation after the cluster candidates that are not generated.
[0022]
When generating a new cluster candidate for a predetermined node, the cluster generation restriction unit 25 is used for generating a candidate for a node set that can be reached by following one link only from that node. Not to be. Thereby, the number of child cluster candidates to be generated can be limited.
[0023]
Next, an example of operation | movement in one Embodiment of the hypertext structure change assistance apparatus of this invention is demonstrated. FIG. 3 is a flowchart showing an example of the operation in the log clustering unit 2. FIG. 4 is an explanatory diagram of a specific example of the structure of content stored in the Web server 1. In the following description, it is assumed that the content as shown in FIG. 4 is associated by the link as shown in the figure, and the description will be made with reference to FIG. The main processing shown in FIG. 3 is performed in the cluster configuration unit 23.
[0024]
First, in S31, a threshold value for the number of accesses relating to the number of users who visited all the nodes included in the cluster is read. In S32, a non-inherited node set in which nodes that do not follow a link are generated when a cluster is generated is read. In this example, the threshold of the number of accesses is 20 and the non-inherited node set is given in advance as [site entrance]. It should be noted that the set is indicated by being surrounded by '[' and ']'.
[0025]
In S33, the link table in which the link relation of the nodes on the Web site is extracted in advance is read from the Web server 1. When this link table is generated, links to nodes of other sites that cannot obtain logs at the Web site are removed.
[0026]
At the same time, in step S34, the starting node given in advance is read. In this example, the top page “site entrance”, which is the starting node, is used in the Web site having the structure shown in FIG. From the top page, it can be accessed by following links to all nodes published by the website.
[0027]
Next, in S35, a cluster consisting only of the origin node is generated as an initial cluster and added to the cluster set. In the example shown in FIG. 4, the cluster set is [site entrance].
[0028]
In S36, it is determined whether or not the cluster set is empty. If the cluster set is empty, it indicates that the processing has been completed, and thus the process proceeds to S39, where the cluster set generated so far is output as a log cluster, and the processing of the log clustering unit 2 is completed. If the cluster set is not empty, the process proceeds to S37 to generate a cluster candidate. In the initial state, since an initial cluster has been generated, the process proceeds to cluster candidate generation processing in S37.
[0029]
In S37, a cluster is extracted from the cluster set, and a cluster candidate is generated by a combination of the cluster and a link destination node from a node included in the cluster. In the example shown in FIG. 4, since the node “site entrance” is linked to the nodes “printer” and “copy”, the node set is a child cluster consisting of [site entrance, printer] as a new cluster candidate. , [Site Entrance, Copy] are generated as cluster candidates.
[0030]
Next, in S38, it is checked whether or not the generated cluster candidate satisfies a criterion that can be a cluster, and one that satisfies the cluster criterion is selected. FIG. 5 is a flowchart illustrating an example of processing for determining whether or not a cluster candidate satisfies a cluster criterion. In S41, it is determined whether or not there is a cluster having the same node set. If there is a cluster having the same node set, it is excluded from the candidates.
[0031]
Next, in S42, it is determined whether or not a cluster having a subset of cluster candidates has been excluded as not satisfying the cluster criteria in the past, and if a cluster having a subset of cluster candidates has been excluded in the past, The cluster candidate is excluded from the candidates. This process is performed in the cluster generation acceleration unit 24.
[0032]
Next, in S43, the number of users (the number of user identifiers) accessing all the node sets included in the cluster and accessing the starting node first is counted using the access number counting unit 22. If the number is less than the threshold value of the number of accesses read in S31, it is excluded from the cluster candidates. As a method of counting the number of users, for example, a method of counting only those in which the order of accessing each node matches can be performed. In this case, the node sets included in the cluster also have order information, and are regarded as the same cluster only when the order information matches.
[0033]
In S38, cluster candidates that pass the determination criteria shown in FIG. 5 are added to the cluster set. Then, the process returns to S36.
[0034]
Returning to S36, if the cluster set is not empty, the processes of S37 and S38 are repeated. At this time, if there are a plurality of clusters in the newly generated cluster set, the cluster with the smallest number of clusters is taken out and cluster candidates are generated in the same manner. Thereafter, cluster candidates are generated using nodes that are accessible by following one link from the nodes included in the cluster. This is because when the user accesses, it may be possible to return to the node using the back function of the browser and access another page.
[0035]
However, the nodes that are indexes in the Web server contents can be restricted by the cluster generation restriction unit 25 so that the link destination is not used for generation of child cluster candidates, thereby preventing an increase in the number of combinations. . In the example shown in FIG. 4, assuming that the node “site entrance” is an index, when a subsequent cluster is generated, the expansion from the cluster [site entrance] to the cluster [copy, printer] Not done. With respect to this index node, it is possible not to generate a cluster by an explicit input from the user. Further, here, since the correspondence with the clustering based on the similarity of the contents is performed, if the similarity between the contents is low, the corresponding content cluster is not assigned in the intention deviation detection unit 4, and therefore a cluster including those nodes is generated. It does not have to be done. Therefore, calculate the similarity between nodes connected by links from a certain node in advance, detect the nodes whose average similarity is below the threshold, and limit the generation of clusters using the detected nodes as index nodes It is possible.
[0036]
Further, when generating a cluster candidate in S37, if a child cluster having the same node set has already been generated from another parent cluster, an attempt is made to generate the child cluster that has already been generated. It is possible to generate a relationship as a child cluster of the parent cluster.
[0037]
FIG. 6 is an explanatory diagram of an example of a log cluster. By the processing in the log clustering unit 2 as described above, a log cluster as shown in FIG. 6 is output from the content structure shown in FIG. That is, after cluster 1 [site entrance] is generated as an initial cluster, nodes 2 “printer” and “copy” that follow one link from node [site entrance], cluster 2 [site entrance, printer] Then, cluster 3 [site entrance, copy] is generated. Similarly, cluster 4 and cluster 5 are generated from cluster 2, and cluster 6 is generated from cluster 3. Cluster candidate 1 [site entrance, copy, model D], which is a cluster candidate based on cluster 3, is not a cluster because the number of accesses is less than 20.
[0038]
From cluster 4, cluster 8 and cluster 9 are generated. At this time, the same cluster as the cluster 8 is generated from the cluster 5 as a child cluster candidate. In this case, as described above, a relationship as a child cluster of the cluster 5 is also generated for the already generated cluster 8. Similarly, a cluster 10 is generated from the cluster 8, but a relationship as a child class of the cluster 9 is also generated.
[0039]
From cluster 6, cluster candidate 2 having a node set [site entrance, copy, model C, model D] can be generated, but cluster candidate 1 having a subset of this node set has not previously met the criteria, It is removed by the condition of S42 in FIG.
[0040]
The clusters generated in this way have a directed acyclic graph (DAG) structure as shown in FIG. 6, and there is a relationship between which cluster has a child cluster and which cluster has a parent cluster. Has been generated. Since the generated log cluster has a tree structure having a parent-child relationship, the structure shown in FIG. 6 is similar to that shown in FIG. 4, but the log cluster is formed as a structure different from the content structure. Note that this is done.
[0041]
Next, processing for generating a content cluster by the content clustering unit 3 will be described. The content clustering unit 3 generates a profile of each node for each node using, for example, the method disclosed in Japanese Patent Application No. 9-153387, calculates the similarity for all two node combinations, Remember. Based on the calculated similarity, clustering is performed using a known method. As a clustering technique, for example, Ellen M. et al. Vorhees, “IMPLEMENTING AGGLOMERATIVE HIERARCHIC CLUSTERING ALGORITHMS FOR USE IN DOCUMENT RETRIVAL”, Information Processing & Management, Vol. 22, no. 6, pp. Clustering can be performed using a complete link of Hierarchical Clustering Algorithm described in 465-476, 1986 and the like.
[0042]
FIG. 7 is an explanatory diagram of an example of a content cluster. When contents as shown in FIG. 4 are stored in the Web server 1, the contents cluster structure as shown in FIG. 7 is obtained by hierarchically clustering these contents. That is, since the contents of the node “model A” and the node “function of model A” are similar, a cluster [model A, function A of model A] is obtained. Further, since these nodes are similar to the node “model B”, a cluster [model A, function A of model A, model B] is obtained. On the other hand, since the contents of the node “model C” and the node “model D” are similar, a cluster [model C, model D] is obtained. Finally, since these nodes are similar, a cluster [model A, function A of model A, model B, model C, model D] is obtained. In this way, [Model A, Model A Function A], [Model A, Model A Function A, Model B], [Model C, Model D], [Model A, Model A Function A, Model B , Model C, model D] are obtained.
[0043]
Note that the processing of the content clustering unit 3 does not consider the links between the nodes, and performs clustering based on, for example, the similarity of the contents. For this reason, in the example shown in FIG. 4, content that is similar in content is connected by a link, but nodes that are distant from each other in the link relationship may be combined into one cluster by content clustering. Alternatively, a node that has no link for some reason can also be included in the content cluster.
[0044]
FIG. 8 is a flowchart showing an example of operations in the intention deviation detection unit 4 and the intention deviation presentation unit 5. First, in S51 and S52, the log cluster and the content cluster generated as described above are read. Here, it is assumed that the log cluster shown in FIG. 6 and the content cluster shown in FIG. 7 are read.
[0045]
Next, in S53, a content cluster included in a leaf cluster of the log cluster (a cluster in which a cluster in which a node is newly added to the node set of the cluster has not been generated) is detected. The leaf clusters of the log cluster shown in FIG. Content clusters [model A, function A of model A] and [model A, function A of model A, model B] are detected from the cluster 10 as content clusters included in these log clusters. Hereinafter, the content cluster [model A, function A of model A, model B] is referred to as content cluster 1, and the content cluster [model A, function A of model A] is referred to as content cluster 2. Note that the content cluster included in the cluster 6 is not detected.
[0046]
In S54, the leaf cluster of the log cluster including the content cluster detected in S53 is added to the set 1. Content cluster 1 and content cluster 2 are included in leaf cluster 10 of the log clusters. Therefore, cluster 10 is added to set 1 among the log clusters.
[0047]
In S55, it is determined whether or not the set 1 is empty. If the set 1 is empty, the process ends. If set 1 is not empty, the following processing is performed.
[0048]
In S56, one of the log clusters is extracted from the set 1. In step S57, the content cluster is associated with the parent log cluster from the extracted log cluster. FIG. 9 is an explanatory diagram of the correspondence between log clusters and content clusters. As described above, since the set 1 includes the clusters 10 of the log clusters, the cluster 10 is extracted from the set 1. Here, the content cluster included in the cluster 10 is shown as the content cluster 1 in FIG. Then, as indicated by the arrow in FIG. 9, the content cluster 1 is correlated from the cluster 10 to the clusters 8 and 9 which are the parent log clusters.
[0049]
At this time, in S58, it is determined whether or not a node not included in the parent log cluster exists in the content cluster. If a node not included in the parent log cluster exists in the content cluster, the process proceeds to the determination in S59. FIG. 10 is an explanatory diagram of an example of how content clusters and log clusters overlap. In this example, when the content cluster 1 is associated with the cluster 9 that is the parent log cluster from the cluster 10, the node “model B” existing in the content cluster 1 is not included in the cluster 9 as shown in FIG. 10. Therefore, it progresses to S59. In the determination in S58, a shift between the log cluster and the content cluster can be detected. That is, it is possible to detect the presence of a portion where the user's access to the node does not match the similarity of the contents.
[0050]
In S59, it is determined whether or not the relationship between the content cluster and the parent log cluster is in a predetermined format. FIG. 11 and FIG. 12 are explanatory diagrams of an example of the difference between the content cluster and the log cluster. As the predetermined format, for example, the formats shown in FIGS. 11 and 12 can be determined. The format shown in FIG. 11 is an example in which a part of the contents of the node 2 is described in detail in the node 3, and the side that created the site expects access from the node 2 to the node 3 further. It is a link format. However, it indicates that the node 3 is not actually accessed. In the format shown in FIG. 12, the contents of node 1 are explained in detail in node 2 and node 3, and the side that created the site expects the same number of accesses to node 1 and node 2. This is a link format that is important to know the user's selection point depending on whether or not the user accesses. The format shown in FIG. 12 indicates that node 2 is being accessed, but node 3 is hardly accessed. Here, since the example shown in FIG. 10 is the link format shown in FIG. 12, the parent log cluster and the content cluster are determined to be in a predetermined format.
[0051]
If it is determined in S59 that the format is predetermined, an evaluation value is calculated and presented to an administrator or the like in S61, assuming that an intentional deviation has occurred. The evaluation value can be represented by, for example, the number of accesses classified into the parent class / the number of accesses classified into the child clusters. The larger the evaluation value, the higher the possibility that the number of user accesses can be increased by changing the content, or the clearer the user's selection. As shown in FIG. 6 or FIG. 9, since the number of accesses whose access tendency is classified into the cluster 9 is 50 and the number of accesses whose access tendency is classified as the class 10 is 30, the evaluation value is 1.7. It becomes.
[0052]
FIG. 13 is an explanatory diagram of an example of the presented evaluation result. Using the evaluation value calculated as described above, for example, it can be presented to the administrator or the like in the format shown in FIG. As points to change the site, highlight the nodes in the content cluster that are out of the log cluster and the link source node to the node, and the node structure included in the log cluster and not included in the log cluster Indicates the relationship of the nodes. At this time, each node presents the contents of the node in the same format as the Web browser, or presents only the page title when the display size is limited. In FIG. 13, for the convenience of illustration, the highlighted display is hatched. Further, the display method is arbitrary, and the display can be made easy to understand without being limited to the form shown in FIG.
[0053]
In the above description, the case where the content cluster is associated with the cluster 9 as the parent log cluster of the cluster 10 in S57 has been described. A case will be described in which a content cluster is associated with cluster 8, which is another parent log cluster, as a parent log cluster of cluster 10. In this case, since the node “function A of model A” included in the content cluster 1 is not included in the cluster 8, the process proceeds from S58 to S59. Since the relationship between the cluster 8 and the content cluster 1 corresponds to the link format shown in FIG. 11, an evaluation value is calculated in S61 and an evaluation result is presented. FIG. 14 is an explanatory diagram of another example of the presented evaluation result. In this case, the node “model A” and the node “function A of model A” are highlighted, and the evaluation value is displayed on the link between them. Thereby, the change point of the site can be shown to the administrator or the like. Also in this case, the display form is arbitrary.
[0054]
If all the nodes of the content cluster are included in the parent cluster in S58, or if the relationship between the parent cluster and the content cluster is not in a predetermined format in S59, the evaluation value is not calculated and the evaluation result is not displayed. , Go to S60. Also, after calculating the evaluation value and displaying the evaluation result in S61, the process proceeds to S60.
[0055]
In S60, it is determined whether or not the parent cluster still includes the content cluster. If so, the parent cluster is added to the set 1 in S62. Since the cluster 9 which is the parent cluster of the cluster 10 includes the content cluster 2 [function A of model A, model A], the cluster 9 is added to the set 1. Note that the cluster 8 that is the parent cluster of the other cluster 10 does not include the content cluster, and thus is not added to the set 1.
[0056]
Then, the process returns to S55. In this example, cluster 1 still exists in set 1. Therefore, the processing after S56 is repeated. The content cluster 2 is associated with the cluster 4 that is the parent log cluster of the cluster 9, but the node “model A function A” is not included in the cluster 4. Since the relationship between the parent log cluster and the content cluster in this case is in the format shown in FIG. 11, the evaluation value is calculated in S61 and the evaluation result is displayed.
[0057]
Since this cluster 4 does not include a content cluster, it is not added to the set 1, and the process returns to S55 where the set 1 becomes empty and the processing is terminated.
[0058]
In this way, it is possible to detect and display a shift between a node intended for the same access as the site creator from the content and a node that is actually accessed. For example, by referring to the display as shown in FIG. 13, the administrator can know that the model A is popular. Alternatively, it is possible to take measures to increase access to the model B. Further, for example, by referring to the display as shown in FIG. 14, it is possible to take a measure that allows the function A of the model A to be accessed more.
[0059]
The above-described embodiment can also be realized by a computer program. In that case, the program, data used by the program, and the like can be stored in a computer-readable storage medium. A storage medium is a signal format that causes a state of change in energy such as magnetism, light, electricity, etc. according to the description of a program to a reader provided in the hardware resources of a computer. Thus, the description content of the program can be transmitted to the reading device. For example, a magnetic disk, an optical disk, a CD-ROM, a memory built in a computer, and the like.
[0060]
【The invention's effect】
As is clear from the above description, according to the present invention, it is possible to detect a deviation between the user's access tendency and the intention of the content creator of the website and provide support when changing the content of the website. It is possible to maintain and manage the website promptly according to the user's trend.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an embodiment of a hypertext structure change support device of the present invention.
FIG. 2 is a block diagram showing an example of a log clustering unit in the embodiment of the hypertext structure change support device of the present invention.
FIG. 3 is a flowchart illustrating an example of an operation in the log clustering unit 2;
FIG. 4 is an explanatory diagram of a specific example of the structure of content stored in the Web server 1;
FIG. 5 is a flowchart illustrating an example of processing for determining whether or not a cluster candidate satisfies a cluster criterion.
FIG. 6 is an explanatory diagram of an example of a log cluster.
FIG. 7 is an explanatory diagram of an example of a content cluster.
FIG. 8 is a flowchart showing an example of operations in the intention deviation detection unit 4 and the intention deviation presentation unit 5;
FIG. 9 is an explanatory diagram of a correspondence relationship between a log cluster and a content cluster.
FIG. 10 is an explanatory diagram of an example of how content clusters and log clusters overlap.
FIG. 11 is an explanatory diagram of an example of a difference between a content cluster and a log cluster.
FIG. 12 is an explanatory diagram of another example of a shift between a content cluster and a log cluster.
FIG. 13 is an explanatory diagram of an example of the presented evaluation result.
FIG. 14 is an explanatory diagram of another example of presented evaluation results.
FIG. 15 is an explanatory diagram of an example of a Web site structure and a user access route;
FIG. 16 is an explanatory diagram of an example of the number of user transitions between a website structure and nodes;
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Web server, 2 ... Log clustering part, 3 ... Content clustering part, 4 ... Intentional deviation detection part, 5 ... Intentional deviation presentation part, 11 ... Log recording part, 12 ... Content provision part, 21 ... Access information storage part, 22 ... Access count section, 23 ... Cluster configuration section, 24 ... Cluster generation acceleration section, 25 ... Cluster generation restriction section.

Claims (13)

  Log clustering means for generating a hierarchical log cluster from a hypertext set based on access frequency information by a user to a hypertext information group managed by an information system, and similarity of individual hypertexts in the hypertext information group Corresponding between content clustering means for calculating and generating content clusters from a hypertext set based on the similarity, log clustering results obtained by the log clustering means and content clustering results obtained by the content clustering means Intentional deviation detection means for identifying a partial structure that cannot be attached, and intentional intention presentation means for presenting hypertext that is an element in a content cluster that could not be matched by the intentional deviation detection means Hypertext structure changing support device according to claim Rukoto.   The log clustering means includes one user identifier for accessing the hypertext information group, one for a user having the same user identifier who has accessed for a predetermined period from an access history in which the time and location of the accessed information are recorded. An access information storage means for storing information that is recorded in a time-series order without duplication and information that is accessed by a user having the identifier by giving an identifier and storing the identifier as a set, and given to all the given hypertext sets The access number counting means for calculating the number of identifiers having a history of access satisfying the order restriction, and the hypertext based on the number of identifiers calculated by the access number counting means Cluster configuration means for generating a hierarchical cluster from the set, and the cluster The star composing means has a cluster consisting only of a predetermined starting hypertext as a vertex, one from a hypertext set that can be reached by following one link from the starting hypertext and whose access history is acquired. A cluster set composed of a pair of hypertext and starting hypertext taken out is generated as a new cluster candidate in a lower hierarchy, and the cluster candidate has an access count equal to or greater than a threshold given in advance by the access count counting means. If it is calculated, it is generated as a child cluster of the starting hypertext as a new cluster, and each child cluster is reached by following one link from the hypertext set included in the child cluster. Included in possible hypertext collections One hypertext that has been acquired in one access history and not included in the child cluster is extracted one by one and added to the hypertext set included in the child cluster to generate a cluster candidate in the next lower hierarchy If it is calculated that the cluster candidate has the number of accesses equal to or greater than the predetermined threshold, the new child cluster generates a process of generating a new child cluster having the child cluster as a parent cluster. The hypertext structure change support device according to claim 1, wherein a hierarchical cluster obtained by repeating until it is not generated is generated.   The log clustering means further includes: a cluster for a higher layer having a subset of a hypertext set included in a cluster candidate newly generated when forming a hierarchical cluster; A cluster generation speed-up unit that does not generate the cluster candidate when the number of accesses is equal to or less than a predetermined threshold, and a link from only the hypertext when generating a new cluster candidate for a predetermined hypertext. The hypertext set that can be reached by tracing one by one is provided with cluster generation limiting means for limiting the number of child cluster candidates to be generated by not using the hypertext set when generating candidates. Hypertext structure change support device.   When the candidate of the child cluster generated from the parent cluster has already been generated as a child cluster of another parent cluster, the cluster configuration means has already been generated with the parent cluster that is attempting to newly generate the child cluster. 3. The hypertext structure change support device according to claim 2, wherein a parent-child relationship is generated with the child cluster.   The intention shift detecting means is based on a similarity between hypertexts obtained by the content clustering means for a log cluster set having no child cluster in a cluster having a hierarchical structure obtained by the log clustering means. A content cluster included in each cluster of the log cluster set is selected from the content cluster set, and the content cluster is associated with the parent log cluster of the log cluster associated with the selected content cluster. Whether or not the hypertext is included in the parent log cluster, there is a hypertext that is not included, and the hypertext includes a hypertext set included in the higher-level log cluster and a predetermined link relationship And The hypertext is detected when the number of accesses to the upper log cluster and the change in the number of accesses to the corresponding log cluster are in a predetermined relationship. The hypertext having a predetermined relationship is presented to the user, and the intention deviation detection unit and the intention deviation presentation unit repeat the above processing when a content cluster included in the parent log cluster exists. The hypertext structure change support device according to claim 1, wherein detection / presentation of intention shift is repeated until there is no content cluster included in the parent log cluster or a predetermined stop condition is satisfied.   The intention shift detection means uses a cluster having a hierarchical structure based on the similarity between hypertexts, and if even one element of a lower layer cluster is not included, The hypertext structure change support device according to claim 5, wherein a cluster of a hierarchy is excluded.   The cluster generation restriction unit calculates a similarity between hypertexts connected by links from the same link source hypertext, and the link source hypertext whose average value of the similarities is equal to or less than a predetermined threshold value 4. The hypertext structure change support device according to claim 3, wherein the detected hypertext is used as the predetermined hypertext.   The log clustering means generates a hierarchical log cluster from the hypertext set based on the access frequency information by the user to the hypertext information group managed by the information system, and the similarity of each hypertext in the hypertext information group is determined. Based on the calculated similarity, the content clustering means generates a content cluster from the hypertext set, and the intentional deviation detecting means can identify and associate a partial structure that cannot be matched between the log cluster and the content cluster. A hypertext structure change support method, characterized in that an unintentional presentation means presents hypertext that is an element in a content cluster that has not been present.   The process of generating the log cluster in the log clustering means is the same user who has accessed for a predetermined period from the access history in which the user identifier accessing the hypertext information group, the time, and the location of the accessed information are recorded One identifier is given to a user having an identifier, information accessed by the user having the identifier is recorded in a time-series order without duplication, and the identifier is stored as a set of access information. The number of identifiers having a history of access that satisfies the order constraints given to all text sets is calculated as the number of accesses using the access information, and a cluster consisting only of a predetermined starting hypertext is used as a vertex, Reached by following one link from the origin hypertext A cluster set composed of a combination of hypertext and starting hypertext extracted one by one from the hypertext set for which access history has been acquired is created as a new cluster candidate in the next lower hierarchy, and the cluster candidate Is calculated as a child cluster of the origin hypertext as a new cluster, and each of the child clusters is also included in the child cluster. One hypertext that is included in the hypertext set that can be reached by following one link from the hypertext set and that has an access history and is not included in the child cluster is taken out one by one, Add one more to the included hypertext set If a cluster candidate of a higher rank is generated and it is calculated that the cluster candidate has the number of accesses equal to or greater than the predetermined threshold value, a new child cluster is generated with the child cluster as a parent cluster. 9. The hypertext structure change support method according to claim 8, wherein a hierarchical cluster obtained by repeating the process until no new child cluster is generated is generated.   A process for identifying a partial structure that cannot be associated between the log cluster and the content cluster in the intention disagreement detection unit, and presenting the hypertext that is an element in the content cluster that could not be associated with the intention disagreement presentation unit. In the cluster having a hierarchical structure obtained by the log clustering process, the log cluster set having no child cluster is compared with the log from the content cluster set obtained by the content clustering process based on the similarity between hypertexts. Hypertext in the content cluster is selected by selecting a content cluster included in each cluster of the cluster set and associating the content cluster with the parent log cluster of the log cluster associated with the selected content cluster Calculating whether it is included in the parent log cluster, there is a hypertext that is not included, and the hypertext is in a predetermined link relationship with a hypertext set included in the upper log cluster, and When the number of accesses to the higher-level log cluster and the change in the number of accesses to the corresponding log cluster are in a predetermined relationship, the hypertext is detected as being out of intention, and the hypertext is determined in advance. If there is a content cluster that is included in the parent log cluster and the content is included in the parent log cluster, the content that is included in the parent log cluster is detected and presented as an intentional error by repeating the above process. A cluster no longer exists or meets a predefined outage condition Hypertext structure changing support method according to claim 8, characterized in that repeated.   Log clustering processing for generating a hierarchical log cluster from a hypertext set based on access frequency information by a user to a hypertext information group managed by an information system, and similarity of individual hypertexts in the hypertext information group Corresponding between content clustering processing that calculates and generates content clusters from hypertext sets based on the similarity, log clustering results obtained by the log clustering processing, and content clustering results obtained by the content clustering processing Unintentional detection processing that identifies partial structures that cannot be associated, and intention-intention presentation processing that presents hypertext that is an element in the content cluster that could not be associated A storage medium storing hypertext structure change support program for.   The log clustering process includes one identifier for a user having the same user identifier that has been accessed for a predetermined period from an access history in which the user identifier, time, and location of the accessed information are recorded. The information accessed by the user having the identifier is recorded as access information by combining the information recorded without allowing duplication in time-series order and the identifier, and given to all the given hypertext sets. The number of identifiers having a history of access satisfying the order restriction is calculated as the number of accesses using the access information, and a cluster consisting only of a predetermined starting hypertext is used as a vertex, and a link from the starting hypertext is 1 It is reachable by tracing and access history is acquired Generating a cluster set consisting of a hypertext taken from the hypertext set one by one and a starting hypertext as a new cluster candidate at a lower hierarchy, and the cluster candidate is accessed in a predetermined threshold or more. If it is calculated to have a number, a new cluster is generated as a child cluster of the starting hypertext, and each child cluster is also followed by one link from the hypertext set included in the child cluster. The hypertext that is included in the hypertext set that can be reached and the access history has been acquired and that is not included in the child cluster is extracted one by one and added to the hypertext set included in the child cluster. Generate a cluster candidate in the next lower hierarchy, and If it is calculated that the star candidate has the number of accesses equal to or greater than the predetermined threshold, the process of generating a new child cluster with the child cluster as a parent cluster is performed until no new child cluster is generated. 12. The storage medium storing a hypertext structure change support program according to claim 11, wherein a hierarchical cluster obtained by repeating is generated.   The intention shift detection process is obtained by the content clustering process based on the similarity between hypertexts for a set of log clusters having no child cluster in a cluster having a hierarchical structure obtained by the log clustering process. A content cluster included in each cluster of the log cluster set is selected from the content cluster set, and the content cluster is associated with the parent log cluster of the log cluster associated with the selected content cluster. Whether or not the hypertext is included in the parent log cluster, there is a hypertext that is not included, and the hypertext includes a hypertext set included in the higher-level log cluster and a predetermined link relationship And The hypertext is detected when the number of accesses to the higher-level log cluster and the change in the number of accesses to the corresponding log cluster are in a predetermined relationship, and the intention deviation presentation process is predetermined as the hypertext. Hypertext having the same relationship is presented to the user, and the intentional deviation detection process and the intentional deviation presentation process detect the intentional deviation by repeating the above process when there is a content cluster included in the parent log cluster. 12. The storage medium storing the hypertext structure change support program according to claim 11, wherein the presentation is repeated until there is no content cluster included in the parent log cluster or a predetermined stop condition is satisfied.

Images