JP2010157151A - System and method for analyzing access - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an access analysis system and an access analysis method, for analyzing an access trend of a user terminal based on contents of a content and recording data including a recorded access record running into conversion by detecting and recording access of the user terminal to a Web server opening the content. <P>SOLUTION: In the access analysis system 10, access analysis is executed by use of: a Web structure DB 14 for recording a page structure of a Web site; a page category DB 18 for classifying the page structure into a category and recording it; a path definition DB 17 for defining a path along which each user terminal wanders each page based on the record of an access log DB 11; a customer class DB 16 for defining a class of a customer in order of running into the conversion; a property pattern DB 15 for recording a result obtained by analyzing a property of the customer class by customer class distinction processing; and a control part 20 connected to a display terminal 19. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an access analysis system that detects and records access of a user terminal to a web server that publishes content by a plurality of pages, and analyzes an access trend of the user terminal based on the content of the content and the recorded access record. And an access analysis method.

  With the rapid spread of the Internet in recent years, as e-commerce (electronic commerce) websites that sell various products and services through the Internet increase, consumers who are visitors to the website Opportunities for customers to purchase these products and services are increasing.

  Under such circumstances, companies that operate e-commerce also pay attention to customer behavior on the Internet, and e-marketing, which is different from traditional marketing, has become more important. For example, virtual malls・ Customer actions (hereinafter referred to as “conversions”) are becoming increasingly important for businesses such as purchasing products on the site and membership registration on the community site. In particular, e-marketing aims to improve the conversion rate, and collects and analyzes customer actions in order to provide products and information according to customer preferences.

  In general, in e-commerce, where repetitive purchases are the mainstream, it is easy to determine good customers who frequently purchase products based on abundant customer purchase histories and product characteristics by accumulating POS data. On the other hand, in e-commerce, which mainly sells durable consumer goods, the customer purchases once every few years due to the fact that the assumed useful life is one year or more and the purchase price is relatively high. Customer purchase history information does not exist, and the most important information is the referrer (information of the site immediately before visiting the site), landing page, page browsing route, etc. You must rely on the access records that come from.

  As a technique for solving such a problem, by combining many pieces of information in e-commerce where user profiles, which are user information, are frequently updated as a result of access by many users and products are frequently replaced, There is something to solve. In Patent Document 1, in order to recommend the presentation of products and the like according to the user's preference in an electronic store or the like, a static filter using collaborative filtering that extracts the user's preference from the user profile storing customer information is disclosed. A recommendation system that combines general recommendation and dynamic recommendation using data mining based on a user's access history from an access log is disclosed.

JP 2002-366838 A

  Although the above-mentioned patent document 1 is processed using a user profile storing customer information and an access log, such a technique is used for e-commerce mainly for one-time purchase where customer information is scarce. Analyzing it is necessary to create analysis software specialized for each product and each field, which is not easy.

  Therefore, the present invention detects and records the access of the user terminal to the web server that publishes the content by a plurality of pages, and based on the content of the content and the recorded data including the recorded access record that resulted in the conversion, An object of the present invention is to provide an access analysis system for analyzing access trends of user terminals by using general-purpose analysis software by reducing the rate of specialization for each product and each field.

  In order to achieve the above object, the access analysis system according to the present invention includes a web server that publishes content to a user terminal by a plurality of pages, records the access of the user terminal to the web server, and In an access analysis system that analyzes access trends of user terminals based on recorded access records, category classification means for classifying each page of a web server related to business into a predetermined number of categories, and important for business Path definition means for defining user terminal access that leads to conversion, which is an access result of the user terminal that increases in nature, as a continuous path leading to conversion based on the classified category, and incidental recorded in the access record By combining information further, the user terminal The terminal class classification means that classifies the tendency of the access pattern, which is a pattern of walking a number of pages, into terminal classes for each category, the characteristics of the terminal class based on the defined continuous route and the classified category, the correlation rule Terminal characteristic calculation means for calculating a support value of a user terminal, which is a ratio of the number of access patterns caused by a certain terminal class, out of the total number of access patterns caused by a user terminal corresponding to the terminal class. In order to obtain the contribution degree of the access pattern, terminal contribution degree calculating means for calculating a confidence value for the entire access of the user terminal, which is a ratio of the number of times the access pattern has occurred among the total number of access patterns, and calculation Support values and confidence values for the user terminals It has a narrowing means Filter access pattern based on a threshold, and based on information from the narrowing means and to predict the terminal class to a conversion.

  Further, in the access analysis system according to the present invention, the narrowing-down means includes a dividing means for dividing a continuous access pattern of the user terminal extracted from the access record into a plurality of short access patterns, and a support level and A first filter means for discriminating access of a terminal class having a low degree of confidence from an access record based on a predetermined threshold value, and a terminal class before the conversion from the access record obtained by the first filter means Focused on the second filter means that extracts the threshold value of the support value and the confidence value, and the terminal class that led to conversion from the remaining access records that were distinguished and extracted by the first and second filter means Depending on support and confidence value thresholds Third filter means for extracting, and output means for outputting information about the terminal class are extracted, characterized in that it has a.

  The access analysis system according to the present invention further includes an access record collecting means for collecting two access records, a learning record and a test record, and threshold values of support values and confidence values obtained from the learning records. And threshold value verification means for verifying the threshold value by processing the test record based on the above.

  Further, the access analysis method according to the present invention is based on the contents of the content and the recorded access record by the web server publishing the content to the user terminal by a plurality of pages and recording the access of the user terminal to the web server. In an access analysis method for analyzing access trends of user terminals, a category classification step of classifying each page of a web server related to a business into a predetermined number of categories, and access results of user terminals that are increasingly important for business The user terminal further combines the route definition step of defining the user terminal access that leads to the conversion as a continuous route leading to the conversion based on the classified category and the incidental information recorded in the access record. It is a pattern of walking across multiple pages To determine the characteristics of terminal classes based on the terminal class classification process for classifying access pattern trends into terminal classes for each category and the defined continuous routes and the classified categories, using association rules. The terminal characteristic calculation step of calculating the support value of the user terminal, which is the ratio of the number of access patterns generated by a certain terminal class, out of the total number of access patterns generated by the corresponding user terminal, and the contribution degree of the access pattern In order to obtain, a terminal contribution calculation step of calculating a confidence value for the entire access of the user terminal, which is a ratio of the number of times that the access pattern has occurred among all the access patterns, a calculated support value of the user terminal, and Access pattern based on a predetermined threshold It includes a narrowing step to narrow, and based on information from the narrowing process, characterized in that to predict the terminal class to a conversion.

  Further, in the access analysis method according to the present invention, the narrowing-down step includes a dividing step of dividing a continuous access pattern of the user terminal extracted from the access record into a plurality of short access patterns, and the support level and the short access pattern based on the short access pattern. A first filter step for discriminating access of a terminal class having a low degree of confidence from an access record based on a predetermined threshold, and a terminal class before the conversion from the access record obtained in the first filter step We focused on the second filter process that extracts the threshold value of the support value and the confidence value focused on, and the terminal class that led to conversion from the remaining access records that were discriminated and extracted by the first and second filter processes Extract by threshold value of support value and confidence value And having a third filter process for an output step of outputting information related to the extracted terminal class, a.

  Further, in the access analysis method according to the present invention, an access record collecting step for collecting two access records of a learning record and a test record, and a support value and a confidence value threshold obtained from the learning record And a threshold value verification step for verifying the threshold value by processing the test record based on the above.

  With the present invention, in e-commerce mainly for one-time purchases where customer information is scarce, the access of a user terminal to a web server that publishes content by a plurality of pages is detected and recorded, and the contents are recorded. It is possible to analyze the access trend of the user terminal based on the record data including the access record that has led to the conversion.

Hereinafter, the best mode for carrying out the present invention (hereinafter referred to as an embodiment) will be described with reference to the drawings.
(Embodiment)

  FIG. 1 shows the overall configuration of the access analysis system 10. The website home page has a plurality of pages, and the web server 1 that publishes the website to the user terminals (31 to 33) provides contents to the user terminals (31 to 33) via the Internet 2. The access analysis system 10 detects information of each page accessed by a click operation of a customer who operates the user terminal 31 and stores the information in an access log DB (database, hereinafter abbreviated as DB) 11.

  The access analysis system 10 is a computer system, and a web structure DB 14 that records the page structure of a website, a page category DB 18 that classifies and records the page structure into page categories, and a user based on the records in the access log DB 11. A route definition DB 17 that defines the route the terminal has traveled across each page, a customer class DB 16 that defines customer classes in the order of conversion, and a property that records the result of analyzing the characteristics of the customer class by a customer class discrimination process described later Access analysis is executed using the pattern DB 15 and the control unit 20 connected to the display terminal 19. The access log DB 11 includes a learning log DB 12 and a test log DB 13 that are used in a customer class determination process described later.

  FIG. 2 shows a website of “Shibira (registered trademark)” for which access analysis is performed in the present embodiment. “Shibira (registered trademark)” is software for access analysis, and since it corresponds to a single purchase, there is usually no customer purchase history, and as a powerful information element, referrer (site page Link page information when you click a link to go to another page, etc.), landing page (web page linked to from Internet advertisements or search engine search results), page browsing route, page Only access log data provided by the customer, such as browsing time and search keywords.

  FIG. 3 shows the flow of access analysis processing in this embodiment. In this access analysis process, in order to identify the good customers who are expected to upgrade to conversion, various parameters necessary for the customer class determination process are specified using the learning log and the test log, and then the customer class determination is performed. Is to do.

  When the access analysis process is executed, the loop count is determined (step S10). If it is the first time, the learning log is read (step S12). Next, a page category definition (step S14) for classifying and defining a plurality of pages provided on the website, and a customer class definition (step S16) for defining a customer class related to the page category. Execute. Based on the learning log, page category, and customer class, the unique access pattern is extracted (step S18), and the customer class characterized by the unique access pattern is specified (step S20). If the parameter is specified by the parameter used when specifying the customer class (step S22), the customer class determination (step S24) is executed, and the evaluation of the determined customer class is executed in step S26. It returns to step S10 which is the head.

  In the loop number determination (step S10), when it is determined to be the second time, the test log is read (step S30). Next, based on the test log, the page category, and the customer class, the unique access pattern is extracted (step S18), and the customer class characterized by the unique access pattern is specified (step S20). The parameter is specified by the parameter used when specifying the customer class (step S22), and the customer class determination (step S24) is executed. In step S26, the determined customer class is evaluated and the parameters are optimized. When continuing, it returns to step S10 which is a loop head.

  In the loop number determination (step S10), when it is determined that the number of times is 3 or more, the access log is read (step S32), and then the customer class determination (step S34) is performed to reach the conversion. Anticipate potential customers. In step S36, when it is determined not to continue, the process ends.

  FIG. 4 shows an example of the website page category. The contents of the page category that leads to the conversion of a one-time purchase site are: (1) price list, (2) gift application, (3) member registration, (4) inquiry / document request (5) trial, (6) It can be classified into six types, such as business negotiation reservations. Here, the importance for business increases as the number increases.

  In general, since a website is composed of a large number of pages divided into a plurality of layers (P01-P0101 etc.) as shown in FIG. 4, a customer is used as a unit for accessing each page (hereinafter abbreviated as PV: page view). When trying to analyze the access route, the histogram regarding the number of visits to each page becomes extremely flat, making it difficult to capture the behavioral characteristics of the customer. Therefore, it is necessary to categorize the pages while maintaining the details that are summarized in the contents and can grasp the behavioral characteristics of the customers. In the present embodiment, there are six page categories.

  FIG. 5 shows an example of the customer class of the website. After the categorization of the six pages, it is classified into N customer classes (C1 to Cn) based on the page category path viewed by the customer. In the embodiment, four customer classes C1 to C4 are defined from the page category path. Specifically, in the case of “Shibira (registered trademark)”, the customer class C1 is P01, P04, P05 using (1) price list, (4) inquiry / material request, (5) trial, etc. Customers who left the site without reaching. Customer class C2 browses P01 but does not reach P04 and P05. Customer class C3 viewed P04 and P05, but did not complete transmission. The customer class C4 browses P04 and P05 and classifies them as customers who have completed transmission. By classifying customer classes in this way, it is possible to represent the flow of upgrade and access from the path and start point of the page category to the goal point.

  The access pattern has a continuous kPV (page view) as a basic unit, and in “Sybilla (registered trademark)”, k = 3 and K = 6 from the analysis of the learning log. PV discrimination is 1 to kPV, 2 to (k + 1 to) PV,..., (Kk + 1) to KPV), that is, 1 to 3 PV (1, 2, 3), 2 to 4 PV (2, 3, 4). , 3 to 5 PV (3, 4, 5), 4 to 6 PV (4, 5, 6) as a standard, the class was determined within 6 PV at maximum.

  In order to extract the access pattern, values of support value and confidence value are used as statistical meaning of the correlation rule. The support value for the access pattern p of C1 to C4 is a ratio of the number of times that the access pattern p has occurred among the total number of access patterns that have been caused by the customer of Ci. The confidence value is the ratio of the number of times that Ci has caused the access pattern p in the total number of access patterns p.

  A high support value for an access pattern in a customer class means that there is a high probability that a customer belonging to that class will generate the access pattern. It is impossible to express characteristics.

  For example, if the support value is high but the confidence value is low, the probability that the customer who caused the access pattern belongs to another customer class is higher. Therefore, in order to characterize a certain pattern as a characteristic of a certain customer class, it is necessary that both the support value and the confidence value have high values. For this reason, when the threshold value α or β is exceeded for the Ci access pattern p, the access pattern is set as the Ci characteristic pattern.

  FIG. 6 shows an example of the customer class discrimination process. Customer class discrimination processing is to identify the characteristics of a customer class by extracting access patterns specific to each customer class from the access log and obtaining the characteristics that are likely to cause that pattern to occur for customers belonging to that class. It is. Further, as a specific example, there is shown a case where there are only four access patterns in two customer classes (C1, C2) and page views (A) P01, (B) P04, (C) P05.

  7A shows an example of access pattern extraction, FIG. 7B shows a support value calculated from FIG. 7A, FIG. 7C shows a confidence value calculated from FIG. FIG. 7D shows the final characteristic pattern.

  Here, a procedure for selecting threshold values α and β for identifying page view pattern 1 that has browsed P01, P04, and P05 (ABC) in C2 customer class and has completed transmission will be described. FIG. 7D shows that the support value satisfying the characteristic pattern ABC in the C2 customer class is 0.375 and the confidence value is 0.75, and threshold values (α, If β) = (0.375, 0.75) or higher, 0.445 C1 customer class characteristic pattern A-C-B exceeding the support value 0.375 is applicable, but the confidence value is 0.44. Yes, since it does not exceed 0.75 of the C2 customer class, the characteristic pattern A-C-B is not selected in the C2 customer class. Further, the threshold value obtained from the learning log from FIGS. 7B and 7C is, for example, the characteristic pattern A of the C2 customer class even if (α, β) = (0.35, 0.5). -B-C will be selected.

  As described above, in the analysis of the one-time purchase site, the information elements constituting the characteristic pattern depend on the access log brought by the customer such as a refiner, landing page, page browsing route, page browsing time, search keyword, etc. . Whether or not the access pattern should be configured by a combination of these cannot be generally determined by the page structure of the website or the access behavior of the customer. Therefore, in the present embodiment, a plurality of combinations are evaluated using the learning log and the test log, and the combination that gives the best result is determined by the evaluation of the customer class determination process described later.

  FIG. 8 shows the flow of the customer class discrimination process. In this embodiment, in the determination of the customer class to be upgraded from C1-P (C1 final customer; stage 1) in FIG. 5, when the customer walks across six pages, attention is paid to 1 to 3 PV (page view) at T = 1. Then, non-excellent customers are selected, and whether T2 is 2 to 4 PV, T = 3 is 3 to 5 PV, and T = 4 is 4 to 6 PV, and whether or not the customer is upgraded is selected. Here, k = 3, K = 3, and i are subscripts of Ci.

  When the customer class discrimination process starts, T = 1 is substituted first (step S40). Next, it is determined whether it is outside the loop of 1 to 3 PV (step S42). If it is within the loop condition, the process proceeds to step S46. If an upgrade has been performed, a determination is made at step S44. In step S46, it is determined whether the customer who is in stage 1 (stages 1, 2, and 3) at the time T upgrades or stays in the current customer class. If it is determined that the upgrade is performed at this time, the process proceeds to step S58 to end the determination. If conversion has been achieved at time T, the customer is subsequently excluded from consideration (step S58). If it is determined not to upgrade, the process proceeds to step S48 in order to continue the determination even after the time (T + 1).

  In step S48, it is determined whether it is a U1 (unique characteristic pattern) of C1, and it is determined in step S52 that it matches the characteristic pattern of the C1 customer class based on the above-described threshold values α and β. If the C2 UCP is included by upgrading to a non-C1 UCP, the process moves to step S50 and waits for a plurality of classes of UCPs. If a plurality of classes of UCPs are received as a result of the upgrade, if they match the characteristic pattern common to the two customer classes in step S54, they are distributed to each customer class according to the ratio of the support values of the two customer classes. Further, when the distribution cannot be performed according to the support ratio, the distribution is performed based on the number of customers ratio (step S56). When these processes are completed, the process proceeds to step S60, 1 is added to T, and it is determined whether or not the process returns to the loop in step S62. If the process returns, the remaining T = 2, 3, and 4 are sequentially executed.

  In step S50, a plurality of classes of UCP are awaited, referrer, landing page, Dwell Time, which is the page browsing time, and the like are taken into the access pattern, and reference data for characteristic pattern discrimination accuracy is accumulated. This reference data evaluates the combination that gives the best results.

  In the customer class determination process described above, the learning log is used to obtain threshold values (α, β) for determining the characteristic pattern of each customer class by combining information elements expressing the characteristic pattern, and the best result is obtained. The resulting combination was evaluated.

  Generally, since these parameters have an allowable range, the customer class determination process is further evaluated using a test log. In performing this evaluation, values of Recall and Precision applied to the binomial distribution characteristics are calculated.

  FIG. 9 and FIG. 10 show examples of selecting good customers who will be upgraded to conversions. In four stages from T = 1 (1 to 3 PV) to T = 4 (4 to 6 PV) in stage 1 of C1, Recall and Precision values are shown, respectively. FIG. 9A shows a non-excellent customer selection result that discriminates customers who leave immediately after access at T = 1 (1 to 3 PV). Here, the precision of C1-P is high (72.64%), and it can be seen that non-excellent customers can be selected with high accuracy. FIG. 9 (B) The first refinement at T = 2 is the Precision of the customer who upgrades to C2 (34.37%), while the Precision is also at T = 3,4 (34.25%, 34.67%) In addition, it was found that Recall was also successfully raised and functioned as a sieve.

  Furthermore, when viewing the Precision of the upgrade at T = 4, 355/1024 = 34.67%. This tends to seem low accuracy at first glance, but the probability that 1024 people will be indiscriminately extracted from the test log and 355 or more will be correctly identified when each of them randomly determines whether or not an upgrade is possible. And about 0.02%. From this fact, it will be understood whether the accuracy of the good customer judgment in the customer class judgment processing is high.

  In addition, the calculation method of the probability at the time of determining at random sets the correct answer rate to X, and the correct answer rate X follows the binomial distribution of N = 1024, θ = 0.2922 which is the original value of the upgrade, and N is sufficiently large Therefore, when the probability that X is 355 or more is obtained using normal approximation, the value is about 0.00002 (0.002%).

  As described above, when the access analysis system according to the present embodiment is used, the user terminal access to the web server that publishes the content by a plurality of pages in the e-commerce mainly for one-time purchase where customer information is scarce. Can be detected and recorded, and the access trend of the user terminal can be analyzed based on the content of the content and the recorded data including the recorded access record that has led to the conversion.

(Addendum)
In the present invention, the access analysis system and the access analysis method have been mainly described. However, the access analysis program is also important, and the contents will be described below.

  According to the access analysis program of the present invention, a web server publishes content to a user terminal by a plurality of pages, records the access of the user terminal to the web server, and records the content of the content and the recorded access record. In the access analysis program for analyzing the access trend of the user, there are a category classification step for classifying each page of the web server related to the business into a predetermined number of categories, and the access result of the user terminal increasing in importance to the business. By further combining the route definition step of defining user terminal access that leads to conversion as a continuous route leading to conversion based on the classified category, and additional information recorded in the access record, the user terminal can Walk across pages To determine the characteristics of the terminal class based on the terminal class classification step for classifying the access pattern tendency that is a turn into a terminal class for each category and the defined continuous route and the classified category, using the correlation rule, A terminal characteristic calculation step for calculating a support value of the user terminal, which is a ratio of the number of access patterns generated by a certain terminal class, out of the total number of access patterns generated by the user terminal corresponding to the terminal class, In order to obtain the contribution degree, a terminal contribution degree calculation step for calculating a confidence value for the entire access of the user terminal, which is a ratio of the number of times that the access pattern has occurred among the total number of access patterns, and the calculated user terminal Predetermined thresholds for support and confidence values Based executed and refinement steps Filter access pattern, by the computer, based on information from the narrowing step, characterized in that to predict the terminal class to a conversion.

  In the access analysis program according to the present invention, the narrowing-down step includes a dividing step of dividing a continuous access pattern of the user terminal extracted from the access record into a plurality of short access patterns, and a support level and a short access pattern based on the short access patterns. A first filter step for discriminating access of a terminal class having a low degree of confidence from an access record according to a predetermined threshold, and a terminal class before the conversion from the access record obtained in the first filter step Focused on the second filter step that is extracted based on the support value and the confidence value threshold value, and the terminal class that has been converted from the remaining access records that have been distinguished and extracted by the first and second filter steps Support value and control And having a third filter extracting the threshold Idensu value, and an output step of outputting information about the terminal class extracted.

  The access analysis program according to the present invention further includes an access record collecting step for collecting two access records, a learning record and a test record, and threshold values of support values and confidence values obtained from the learning records. And a threshold verification step of verifying the threshold by processing the test record based on the above.

  The present invention relates to an access analysis system that detects and records access of a user terminal to a web server that publishes content by a plurality of pages, and analyzes an access trend of the user terminal based on the content of the content and the recorded access record. It is possible to use it.

It is a lineblock diagram showing the whole access analysis system composition concerning an embodiment of the present invention. It is an example of the website which performed the access analysis in this embodiment. It is a flowchart figure which shows the flow of the access analysis process which concerns on this embodiment. It is explanatory drawing explaining an example of the page category which concerns on this embodiment. It is explanatory drawing explaining the customer class of the website which concerns on this embodiment. It is explanatory drawing explaining an example of the customer class discrimination | determination process which concerns on this embodiment. It is a schematic diagram which shows the example of extraction of the access pattern which concerns on this embodiment. It is a flowchart figure which shows the flow of the customer class discrimination | determination process concerning this embodiment. It is a schematic diagram which shows the example of selection of the good customer who will lead to the conversion which concerns on this embodiment. It is a schematic diagram which shows the example of selection of the good customer who will lead to the conversion which concerns on this embodiment.

Explanation of symbols

  1 Web server, 2 Internet, 10 access analysis system, 11 access log DB, 12 learning log DB, 13 test log DB, 14 web structure DB, 15 characteristic pattern DB, 16 customer class DB, 17 route definition DB, 18 Page category DB, 19 display terminal, 20 control unit, 31, 32, 33 user terminal.

Claims (6)

An access analysis system in which a web server publishes content to a user terminal by a plurality of pages, records the access of the user terminal to the web server, and analyzes the access trend of the user terminal based on the contents of the content and the recorded access record In
Category classification means for classifying each page of a web server related to business into a predetermined number of categories;
A route definition means for defining user terminal access that leads to conversion, which is a result of user terminal access that is increasingly important to business, as a continuous path leading to conversion based on the classified category;
Further combining the incidental information recorded in the access record, the terminal class classification means for classifying the tendency of the access pattern, which is a pattern in which the user terminal walks across a plurality of pages, into a terminal class for each category;
In order to determine the characteristics of the terminal class based on the defined continuous route and the classified category using the correlation rule, a certain terminal class arose out of the total number of access patterns caused by the user terminal corresponding to the terminal class. Terminal characteristic calculation means for calculating a support value of the user terminal, which is a ratio of the number of access patterns,
In order to obtain the contribution degree of the access pattern, a terminal contribution degree calculation unit that calculates a confidence value for the entire access of the user terminal, which is a ratio of the number of times the access pattern has occurred among the total number of access patterns,
Narrowing means for narrowing down the access pattern based on a predetermined threshold for the calculated support value and confidence value of the user terminal;
Have
An access analysis system that predicts a terminal class leading to conversion based on information from a narrowing-down means. The access analysis system according to claim 1,
Narrowing means
A dividing means for dividing a continuous access pattern of the user terminal extracted from the access record into a plurality of short access patterns;
A first filter means for discriminating access of a terminal class having a low degree of support and confidence based on a short access pattern from an access record according to a predetermined threshold;
A second filter means for extracting from the access record obtained by the first filter means based on a support value and a confidence value threshold value focusing on a terminal class before conversion;
A third filter means for extracting from the remaining access records that are distinguished and extracted by the first and second filter means by a support value and a threshold value of the confidence value focusing on the terminal class that has led to conversion;
An output means for outputting information on the extracted terminal class;
An access analysis system comprising: The access analysis system according to claim 1 or 2, further comprising:
An access record collecting means for collecting two access records, a learning record and a test record;
Threshold verification means for verifying the threshold by processing the test record based on the support value and the confidence value threshold obtained from the learning record;
An access analysis system comprising: An access analysis method in which a web server publishes content to a user terminal by a plurality of pages, records the access of the user terminal to the web server, and analyzes the access trend of the user terminal based on the contents of the content and the recorded access record In
A category classification step for classifying each page of a web server related to business into a predetermined number of categories;
A path definition step for defining user terminal access that leads to conversion, which is a result of user terminal access increasing in importance to the business, as a continuous path to conversion based on the classified category;
Further combining the incidental information recorded in the access record, a terminal class classification step of classifying the access pattern tendency, which is a pattern in which the user terminal moves across a plurality of pages, into terminal classes for each category;
In order to determine the characteristics of the terminal class based on the defined continuous route and the classified category using the correlation rule, a certain terminal class arose out of the total number of access patterns caused by the user terminal corresponding to the terminal class. A terminal characteristic calculation step of calculating a support value of the user terminal, which is a ratio of the number of access patterns,
In order to determine the contribution degree of the access pattern, a terminal contribution degree calculation step of calculating a confidence value for the entire access of the user terminal, which is a ratio of the number of times the access pattern has occurred among all access pattern numbers,
A narrowing-down process for narrowing down the access pattern based on a predetermined threshold value for the calculated support value and confidence value of the user terminal;
Including
An access analysis method for predicting a terminal class leading to a conversion based on information from a narrowing-down process. The access analysis method according to claim 4,
The refinement process is
A division step of dividing a continuous access pattern of the user terminal extracted from the access record into a plurality of short access patterns;
A first filtering step for discriminating access of a terminal class having a low degree of support and confidence based on a short access pattern from an access record according to a predetermined threshold;
A second filter step of extracting from the access record obtained in the first filter step by a support value and a threshold value of the confidence value focusing on the terminal class before conversion;
A third filtering step for extracting from the remaining access records that have been distinguished and extracted by the first and second filtering steps, based on the support value and the confidence value threshold value focusing on the terminal class that has led to conversion;
An output process for outputting information on the extracted terminal class;
An access analysis method characterized by comprising: The access analysis method according to claim 4 or 5, further comprising:
An access record collecting process for collecting two access records, a learning record and a test record;
A threshold verification process for verifying the threshold by processing the test record based on the support value and the confidence value threshold obtained from the learning record;
An access analysis method characterized by comprising: