JP2007249600A - Method for classifying objective data to category - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for automatically classifying objective data to categories with accuracy close to that in manual operation, and continuously acquiring such a reliable category sample (learning case) to be usable for subsequent classification of new objective data. <P>SOLUTION: In a terminal device 20, a user follows, when searching objective data such as a commodity stored in a server 10, a link categorized in a tree structure and browses an objective data group of a desired kind. The server 10 stores the association of the objective data with the category. The server 10 assigns categories as candidates of classification when data for learning a category for classification is received. The user of the terminal device 20 selects any one of the assigned categories and reaches the objective data. The server 10 stores this selected category. The server 10 classifies, upon receipt of new objective data, the data to the learnt category. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method, a server, and a program for classifying target data to be browsed by a user of a terminal device.

  Conventionally, various methods for searching information (target data) on the Internet have been provided, and one of them is to search for target information (following a tree-structured category). There is a search engine called a directory type. This method searches for target data that has been classified into categories by hand, and can browse the same type of information collectively. For example, it is used for searching for product information.

  In order to realize this method, it is necessary to store category information for information providing items such as products in advance. Furthermore, in order for the search by the user to be performed effectively, high accuracy is required when classifying into categories. Under such circumstances, techniques for improving the accuracy of information classification are known.

For example, in Patent Document 1, in order to reduce misclassification due to overlapping information between classes (categories), features that appear in the class of interest for each class but are difficult to appear in other classes, and appear in other classes However, a method is disclosed in which a feature that is difficult to appear in the class of interest is obtained, and the similarity is corrected using the data.
JP 2003-256441 A

  However, even with the above method, the accuracy of classification depends on the accuracy of the learning examples (a combination of keywords and other characteristics that characterize the target data and the category) or the accuracy of the similarity evaluation at the time of classification. The possibility of being left still remains. For this reason, if category information is automatically assigned to each objective data that is normally performed manually, the classification accuracy will gradually deteriorate due to the accumulation of misclassifications included in automatically classified learning cases. Is concerned.

  Therefore, the present invention can automatically classify target data into categories with an accuracy close to that performed manually, and continuously obtain such highly reliable classification samples (learning examples). It is an object of the present invention to provide a method that can be used to classify new target data.

  Specifically, the following are provided.

(1) A method in which a server connected to a terminal device via a communication network classifies target data to be browsed by a user of the terminal device into a category,
Storing the target data in association with a plurality of candidate category data;
Transmitting the target data associated with the selected candidate category data in response to receiving data representing one selection of the plurality of candidate category data by the user from the terminal device;
In response to receiving data representing selection of the target data by the user from the terminal device, the user counts up the number of selections of candidate category data selected by the user, and the selected target data and Storing in association with the selected candidate category data;
Determining the category for classifying the target data based on the selection frequency data in a predetermined period;
Including methods.

  According to the invention of (1), the server stores the target data to be browsed by the user of the terminal device in association with a plurality of candidate category data, and from the terminal device, one of the plurality of candidate category data by the user is stored. In response to receiving the data indicating the selection of the user, the target data associated with the selected candidate category data is transmitted, and the data indicating the selection of the target data by the user is received from the terminal device. Accordingly, the selection count data of the candidate category data selected by the user is counted up, stored in association with the selected target data and the selected candidate category data, and the selection count in a predetermined period. The category for classifying the target data is determined based on the data.

  As a result, the server can store the actual user access history (from which category the target data has been reached) for a plurality of candidate categories, so that by selecting a category based on the access history Even if there are unsuitable candidate categories, such categories are automatically excluded and automatically classified into categories that are close to the user's expectations (high accuracy). There is a possibility.

  (2) The method according to (1), wherein the category to be determined has a number of selections of the candidate category data equal to or greater than a predetermined number.

  According to the invention of (2), the server classifies the target data into categories in which the number of selections is a predetermined number or more.

As a result, the server classifies the category into a category with a track record of selection (more than a predetermined number of times), so that there is a possibility that it can be automatically classified into a category that can be easily selected by an actual user.
(3) The method according to (1), wherein the category to be determined has a number of selections of the candidate category data equal to or higher than a predetermined rank.

  According to the invention of (3), the server classifies the target data into categories corresponding to those with the number of selections equal to or higher than a predetermined order.

  Accordingly, since the server selects one or a plurality of categories that are frequently selected, there is a possibility that the server can automatically classify the categories that many users think (easy to come up with).

  (4) The method according to (1), wherein the category to be determined is determined by relative evaluation of the number of selections of the candidate category data.

  According to the invention of (4), the server determines a category for classifying the target data by relative evaluation of the number of selections of the category data.

  As a result, the server can automatically select a category that is closer to the user's orientation because it can select a category that has a higher proportion of selection than other categories. There is sex. In addition, a category can be appropriately selected even when no candidate category has reached the predetermined number of selections in a predetermined period or when there is a large difference in the number of selections between candidate categories within a predetermined rank.

  (5) The method according to any one of (1) to (4), further including a step of extracting and storing keyword data included in the target data in association with the determined category data.

  According to the invention of (5), the server extracts and stores the keyword data included in the target data in association with the determined category data.

  Thus, the server stores the association between the category data and the keyword data, and can be used later as a classification example.

(6) calculating a similarity between the new target data and the keyword data in response to accepting registration of new target data different from the target data;
Determining a category for classifying the new target data in accordance with a category evaluation based on the calculated similarity;
The method according to (5), further comprising:

  According to the invention of (6), the server calculates the similarity between the new target data and the keyword data in response to receiving registration of new target data different from the target data. Then, based on the calculated similarity, a category for classifying the new target data is determined.

  Accordingly, the server can automatically classify the target data having similar keywords into corresponding categories based on the association between the stored keyword data and the category data.

  (7) The step of storing the target data in association with a plurality of the category data calculates a similarity between the target data and the keyword data, and the candidate is evaluated according to a category evaluation based on the calculated similarity. The method according to (5) or (6), wherein category data is selected.

  According to the invention of (7), the server calculates the similarity between the target data and the keyword data, and selects the candidate category data based on the calculated similarity.

  As a result, the server performs learning based on the association between the stored keyword data and category data (determining the category to be classified based on the access history) as candidates for classifying a plurality of categories having similar keywords. It can be associated with the target data. Therefore, it is possible to automatically extract a category for which a user access history is acquired.

(8) The step of extracting and storing keyword data included in the target data in association with the determined category data further stores selection probability data based on the number of selections of the category,
The method according to (6) or (7), wherein an evaluation of a category associated with the keyword data is based on the similarity and the selection probability data.

  According to the invention of (8), the server further stores selection probability data based on the number of selections of the category in association with the determined category data, and evaluates the category associated with the keyword data with the similarity And based on the selection probability data.

  Thereby, for example, the server can preferentially classify a category having a high selection probability by the user from categories that are candidates for classification extracted based on the degree of similarity. Therefore, there is a possibility that more natural classification can be performed according to the user's intention.

(9) A server that is connected to a terminal device via a communication network and classifies target data to be browsed by a user of the terminal device,
Means for storing the target data in association with a plurality of candidate category data;
Means for transmitting target data associated with the selected candidate category data in response to receiving data representing one selection of the plurality of candidate category data by the user from the terminal device;
In response to receiving data representing selection of the target data by the user from the terminal device, the user counts up the number of selections of candidate category data selected by the user, and the selected target data and Means for storing in association with the selected candidate category data;
Means for determining the category for classifying the target data based on the selection frequency data in a predetermined period;
A server comprising

  According to the invention of (9), the same effect as that of (1) can be expected by operating the server.

  (10) The server according to (9), wherein the category to be determined is one in which the candidate category data is selected a predetermined number of times or more.

  According to the invention of (10), the same effect as in (2) can be expected by operating the server.

  (11) The server according to (9), wherein the category to be determined has a number of selections of the candidate category data equal to or higher than a predetermined order.

  According to the invention of (11), the same effect as that of (3) can be expected by operating the server.

  (12) The server according to (9), wherein the category to be determined is determined by relative evaluation of the number of selections of the candidate category data.

  According to the invention of (12), the same effect as in (4) can be expected by operating the server.

  (13) The server according to any one of (9) to (12), further including means for extracting and storing keyword data included in the target data in association with the determined category data.

  According to the invention of (13), the same effect as in (5) can be expected by operating the server.

(14) means for calculating a similarity between the new target data and the keyword data in response to accepting registration of new target data different from the target data;
Means for determining a category for classifying the new target data according to the evaluation of the category based on the calculated similarity;
The server according to (13), further comprising:

  According to the invention of (14), the same effect as in (6) can be expected by operating the server.

  (15) A means for storing the target data in association with a plurality of the category data calculates a similarity between the target data and the keyword data, and the candidate is evaluated according to a category evaluation based on the calculated similarity. The server according to (13) or (14), wherein category data is selected.

  According to the invention of (15), the same effect as that of (7) can be expected by operating the server.

(16) The means for extracting and storing the keyword data included in the target data in association with the determined category data further stores the selection probability data based on the number of selections of the category,
The server according to (14) or (15), wherein an evaluation of a category associated with the keyword data is based on the similarity and the selection probability data.

  According to the invention of (16), the same effect as that of (8) can be expected by operating the server.

(17) A program for causing a server connected to a terminal device via a communication network to classify target data to be viewed by a user of the terminal device into a category,
Storing the target data in association with a plurality of candidate category data;
In response to receiving data representing one selection of the plurality of candidate category data by the user from the terminal device, transmitting target data associated with the selected candidate category data;
In response to receiving data representing selection of the target data by the user from the terminal device, the user counts up the number of selections of candidate category data selected by the user, and the selected target data and Storing in association with the selected candidate category data;
Determining the category for classifying the target data based on the selection frequency data in a predetermined period;
A program that executes

  According to the invention of (17), the same effect as in (1) can be expected by executing the program on the server.

  (18) The program according to (17), wherein the category to be determined has a number of selections of the candidate category data equal to or greater than a predetermined number.

  According to the invention of (18), the same effect as in (2) can be expected by executing the program on the server.

  (19) The program according to (17), wherein the category to be determined has a number of selections of the candidate category data equal to or higher than a predetermined order.

  According to the invention of (19), the same effect as in (3) can be expected by executing the program on the server.

  (20) The program according to (17), wherein the category to be determined is determined by relative evaluation of the number of selections of the candidate category data.

  According to the invention of (20), the same effect as in (4) can be expected by executing the program on the server.

  (21) The program according to any one of (17) to (20), further including a step of extracting and storing keyword data included in the target data in association with the determined category data.

  According to the invention of (21), the same effect as in (5) can be expected by executing the program on the server.

(22) a step of calculating a similarity between the new target data and the keyword data in response to accepting registration of new target data different from the target data;
Determining a category for classifying the new target data in accordance with a category evaluation based on the calculated similarity;
The program according to (21), further including:

  According to the invention of (22), the same effect as in (6) can be expected by executing the program on the server.

  (23) The step of storing the target data in association with a plurality of the category data causes the similarity between the target data and the keyword data to be calculated, and the candidate is evaluated according to the category evaluation based on the calculated similarity. The program according to (21) or (22), wherein category data is selected.

  According to the invention of (23), the same effect as in (7) can be expected by executing the program on the server.

(24) The step of extracting and storing the keyword data included in the target data in association with the determined category data further stores the selection probability data based on the number of selections of the category,
The program according to (22) or (23), wherein an evaluation of a category associated with the keyword data is based on the similarity and the selection probability data.

  According to the invention of (24), the same effect as in (8) can be expected by executing the program on the server.

  According to the present invention, there is a possibility that the target data can be automatically classified into categories with an accuracy close to that performed manually. Furthermore, by continuously acquiring such highly reliable classification samples (learning examples) and using them for classification of new target data after the next time, there is a possibility that classification accuracy can be maintained high.

  An example of a preferred embodiment according to the present invention will be described below based on the drawings.

[overall structure]
FIG. 1 is a conceptual diagram of a data classification method according to an example of a preferred embodiment of the present invention.

  The server 10 that performs category classification is connected to the terminal device 20 used by the user for browsing the target data via the communication network 30 and operates an information providing service for the user of the terminal device 20.

  In the terminal device 20, when the user searches for information provision items (purpose data) such as products stored in the server 10, the user sorts the desired category of target data (product) by following the links categorized in a tree shape. Data etc.) Browse the group. The server 10 stores the association between the target data and the category.

  When the server 10 receives data (product A) for learning a category to be classified, the server 10 assigns categories (classification 1 and classification 2) that are candidates for classification. The user of the terminal device 20 selects one of the assigned categories and arrives at the target data (product A). The server 10 stores the category selected at this time.

  By storing such a category selection history for a predetermined period, the server 10 can learn category classification close to the user's idea (details of the process will be described later). As a result, when the server 10 receives new target data (product B), the server 10 classifies it into the learned category (for example, classification 2).

[Computer configuration]
FIG. 2 is a block diagram showing a configuration of each computer (server 10 or terminal device 20) constituting the computer system according to an example of the preferred embodiment of the present invention.

  The control unit 110, the storage unit 120, the input unit 130, the display unit 140, and the communication control unit 150 are connected via a bus 160.

  The control unit 110 is an information processing unit (CPU) that performs calculation and processing of information, and controls the entire computer. The control unit 110 reads and executes various programs stored in the storage unit 120 as appropriate, thereby realizing various functions according to the present invention in cooperation with the above-described hardware.

  The storage unit 120 may include a local memory used for executing a program in combination with the control unit 110, a large-capacity bulk memory, and a cache memory used for efficiently searching the bulk memory. The computer readable medium that implements the storage unit 120 may include an electrical, magnetic, optical, or electromagnetic implementation. More specifically, a semiconductor storage device, a magnetic tape, a floppy (registered trademark) disk, a random access memory (RAM), a read only memory (ROM), a CD-ROM, a CD-R / W, and a DVD Includes optical discs.

  The input unit 130 receives input from the user, and may include a keyboard, a pointing device, and the like. The input unit 130 can be connected to the computer directly or via an intervening I / O controller.

  The display unit 140 displays a screen for accepting data input to the user or displays a calculation result screen by the computer, such as a cathode ray tube display (CRT), a liquid crystal display (LCD), or the like. Including a display device. The display unit 140 can be connected to the computer directly or via an intervening I / O controller.

  The communication control unit 150 is a network adapter that enables the computer to be connected to another arithmetic processing system or a storage device via a dedicated network or a public network. The communication control unit 150 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

[Learning process flow]
FIG. 3 is a diagram showing a data classification learning process flow according to an example of the preferred embodiment of the present invention.

  In step S105, the server 10 receives objective data for learning. The target data may be received from the input unit 130 or may be received from another arithmetic processing system or storage device via the communication control unit 150. For example, in the case of data relating to a product, the target data includes information such as a product name and a product description, and a category to be classified is given through this learning process.

  In step S110, the server 10 assigns a plurality of categories that are candidates for classification to the received target data. The assignment of this category may be done manually, but it is desirable that the server automatically performs the assignment if there are already classified samples (details will be described later with reference to FIG. 4).

  As a result of Steps S105 and S110, for example, in the case of product data, candidate categories, product names, and product descriptions are stored in the product classification table 40 as shown in FIG. Based on these data, the terminal device 20 displays product data in association with the category, and operates an information providing service for the user.

  In step S <b> 115, the server 10 receives data representing category selection from the terminal device 20. Specifically, the server 10 receives the data indicating the category (link) followed by the user of the terminal device 20, and is given in step S110 when the user selects (views) the target data. The category data selected from the candidate categories is determined.

  In step S120, the server 10 counts up and stores the number of times each candidate category has been selected. Specifically, for example, an access history table 50 shown in FIG. 6 is used. For each candidate category (category field) assigned to the target data (product name field) in step S110, the value of the selection count field is counted up every time data representing the category selection received in step S115 is received. And remember.

  In step S125, the server 10 determines whether to determine a category for classifying the target data. Specifically, for example, it is possible to determine a category to be classified when a predetermined period has passed since a candidate category is given to the target data and browsing to the user is started, and the number of selections is predetermined. It may be determined when the number is reached.

  Or the server 10 is good also as determining the category classified based on the frequency | count of selection when new target data is received later, without determining the category classified at this time. In this case, the server 10 can store the user's tendency for a longer period and use it for category classification. At this time, steps S125 and S130 are not necessary.

  In step S130, the server 10 determines a category for classifying the target data. Specifically, for example, it can be determined by the following method.

  (1) Select a category with the largest number of selections. In this case, the server 10 selects the category with the highest number of selections from the candidate categories assigned to the target data and associates it with the target data.

  (2) Select a plurality of categories with the highest number of selections. In this case, the server 10 selects a predetermined number from the candidate categories assigned to the target data in descending order of selection, and associates them with the target data.

  (3) Select categories that have more selections than others. In this case, for example, the server 10 calculates a deviation value of the number of selections for each candidate category, selects a deviation value that is equal to or greater than a predetermined value, and associates it with target data.

  The method for determining the category to be classified is not limited to the above, but according to such a method, the server 10 can effectively classify the target data for a category that tends to be selected by many users. it can. As a result, it is also possible to leave only categories that are easy for the user to trace and exclude categories that are difficult to select (inaccurate).

  If it is determined that a category to be classified cannot be determined even after a predetermined period of time has elapsed, the server 10 may display a message to that effect on the administrator's terminal to prompt resetting of candidate categories.

  In step S135, the server 10 associates with the category determined in step S130 (all candidate categories if step S125 and step S130 are not performed), and based on the keywords included in the target data and the number of category selections. The selection probability data of the category to be calculated is stored. Specifically, for example, the classification learning table 60 shown in FIG. 7 is used.

  In the classification learning table 60, the name of the target data (for example, product name), the keyword included in the target data, the category determined in step S130, and the category selection probability are stored in association with each other. Here, the keyword is a word that characterizes target data. For example, in the case of product data, the product description in the product classification table 40 (FIG. 5) is decomposed into words by morphological analysis or the like, and TF * IDF or probability It can be extracted by using a method such as language model.

  The storage means is not limited to a table, and can be stored as an inverted index file. Furthermore, the keyword to be stored may be stored in association with the appearance location, the number of appearances, or the appearance frequency in combination with other classification samples (learning examples). These can be used in the calculation of similarity described later.

  Note that step S135 is performed after the classification category is determined in this processing flow, but the timing is not limited to this. For example, keywords may be stored when the target data is received in step S105. The selection probability may be stored together with the number of selections in step S120.

[Classification process flow]
FIG. 4 is a diagram showing a classification processing flow into categories according to an example of the preferred embodiment of the present invention. The classification process is a process of classifying new target data based on the category classification learned in the learning process flow (FIG. 3), but it can also be used for assigning candidate categories in step S110 of the learning process flow. it can.

  In step S205, the server 10 receives new target data for performing category classification. The new target data may be received from the input unit 130, or may be received from another arithmetic processing system or storage device via the communication control unit 150.

  In step S210, the server 10 determines the similarity between the new target data and the keyword group for each category based on the classification learning table 60 or the inverted index file stored in step S135 of the learning process flow (FIG. 3). calculate. Specifically, for example, the degree of similarity can be calculated by using a technique such as TF * IDF, kNN method, or Naive Bayes method. At this time, if data such as the appearance frequency is stored in advance in association with the keyword, these can be used. Thus, the server 10 can extract a keyword group having a high similarity with the new target data and a category corresponding to the keyword group.

  In step S215, the server 10 determines a category to be classified based on the similarity calculated in step S210. Specifically, the server 10 evaluates whether or not each corresponding category should be classified using the similarity. At this time, the server 10 determines one or more categories to be classified as follows, for example.

  (1) A category corresponding to a keyword group having the highest similarity, or a predetermined number of categories in descending order of similarity (Naive Bayes method or the like). In this case, for example, the target data including “bag” and “numerous leather” in the product description has a high similarity to the keyword of the learning example in the first row according to the example of the classification learning table 60 (FIG. 7). Thus, the server 10 classifies the target data into the “handbag” category.

  In this way, the server 10 refers to the latest keyword / category combination stored at the time when the new target data is received, and uses it to classify the new target data. Note that this category learning table 60 excludes candidate categories that are considered unnecessary based on the actual selection by the user in step S130 of the learning process flow (FIG. 3), as described above. The deterioration of accuracy due to the accumulation of misclassifications can also be suppressed.

  (2) A large number of categories to be classified in an appropriate number of samples having a high degree of similarity are determined (kNN method). In this case, for example, when there are three learning cases similar to the target data and the keyword, if the corresponding category is two “handbags” and one “tote bag”, the server 10 determines the majority of the target data. Classify as “handbag”.

  (3) After weighting by similarity, a large number of categories to be classified in an appropriate number of samples are determined (kNN method). In this case, in the example of (2) above, the “handbag” has a larger number of learning cases with similar keywords than the “totebag”, but if the “totebag” similarity score is higher, the server 10 May classify the “tote bag” as a result of weighting, with a higher rating.

  (4) A category having the highest average similarity with a keyword group associated with the same category, or a predetermined number of categories in descending order. In this case, the server 10 can classify the learning examples classified into the same category into those that have no bias even if the similarity is not the highest, rather than those that vary in similarity with the target data. There is sex.

  (5) After the selection probabilities are weighted on the similarity, classification is performed by the methods (1) to (4) described above. In this case, assuming that the selection probability is high, it is assumed that the classification accuracy is high. For example, the server 10 evaluates the selection probability of the classification learning table 60 (FIG. 7) by multiplying the similarity with the high accuracy learning. It can be classified into categories based on cases.

  As described above, according to the embodiment of the present invention, it is possible to learn a category for classifying target data by operating an information providing service, and to automatically classify new target data for the learned category. it can.

  Here, with regard to the learning results (learning examples), many and newer ones can be expected to have higher accuracy and reflect the user's opinions. Therefore, it is preferable to periodically perform the learning process. For example, when new target data is automatically classified, a plurality of categories may be assigned and learning may be performed.

  In this way, the server 10 can continuously acquire and use learning examples while operating the information providing service, and can maintain high accuracy of category classification.

  As mentioned above, although embodiment of this invention was described, this invention is not restricted to embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

It is a conceptual diagram of the data classification method which concerns on an example of suitable embodiment of this invention. It is a block diagram which shows the structure of each computer which comprises the computer system which concerns on an example of suitable embodiment of this invention. It is a figure which shows the learning processing flow of the data classification which concerns on an example of suitable embodiment of this invention. It is a figure which shows the classification processing flow to the category which concerns on an example of suitable embodiment of this invention. It is a figure which shows the goods classification table which concerns on an example of suitable embodiment of this invention. It is a figure which shows the access history table which concerns on an example of suitable embodiment of this invention. It is a figure which shows the classification learning table which concerns on an example of suitable embodiment of this invention.

Explanation of symbols

10 server 20 terminal device 30 communication network 40 product classification table 50 access history table 60 classification learning table 110 control unit 120 storage unit 130 input unit 140 display unit 150 communication control unit 160 bus

Claims (24)

A server connected to a terminal device via a communication network is a method for classifying target data to be viewed by a user of the terminal device into a category,
Storing the target data in association with a plurality of candidate category data;
Transmitting the target data associated with the selected candidate category data in response to receiving data representing one selection of the plurality of candidate category data by the user from the terminal device;
In response to receiving data representing selection of the target data by the user from the terminal device, the user counts up the number of selections of candidate category data selected by the user, and the selected target data and Storing in association with the selected candidate category data;
Determining the category for classifying the target data based on the selection frequency data in a predetermined period;
Including methods.   The method according to claim 1, wherein the category to be determined is one in which the candidate category data is selected a predetermined number of times or more.   The method according to claim 1, wherein the category to be determined has a number of selections of the candidate category data equal to or higher than a predetermined rank.   The method according to claim 1, wherein the category to be determined is determined by a relative evaluation of the number of selections of the candidate category data.   5. The method according to claim 1, further comprising a step of extracting and storing keyword data included in the target data in association with the determined category data. Calculating a similarity between the new target data and the keyword data in response to accepting registration of new target data different from the target data;
Determining a category for classifying the new target data in accordance with a category evaluation based on the calculated similarity;
The method of claim 5 further comprising:   The step of storing the target data in association with a plurality of the category data calculates a similarity between the target data and the keyword data, and stores the candidate category data according to a category evaluation based on the calculated similarity. The method according to claim 5 or 6, wherein the method is selected. Extracting and storing keyword data included in the target data in association with the determined category data, further storing selection probability data based on the number of selections of the category,
The method according to claim 6 or 7, wherein an evaluation of a category associated with the keyword data is based on the similarity and the selection probability data. A server that is connected to a terminal device via a communication network and classifies target data to be browsed by a user of the terminal device,
Means for storing the target data in association with a plurality of candidate category data;
Means for transmitting target data associated with the selected candidate category data in response to receiving data representing one selection of the plurality of candidate category data by the user from the terminal device;
In response to receiving data representing selection of the target data by the user from the terminal device, the user counts up the number of selections of candidate category data selected by the user, and the selected target data and Means for storing in association with the selected candidate category data;
Means for determining the category for classifying the target data based on the selection frequency data in a predetermined period;
A server comprising   10. The server according to claim 9, wherein the category to be determined is one in which the candidate category data is selected a predetermined number of times or more.   10. The server according to claim 9, wherein the category to be determined has a number of selections of the candidate category data equal to or higher than a predetermined order.   The server according to claim 9, wherein the category to be determined is determined by relative evaluation of the number of selections of the candidate category data.   The server according to claim 9, further comprising means for extracting and storing keyword data included in the target data in association with the determined category data. Means for calculating the similarity between the new target data and the keyword data in response to accepting registration of new target data different from the target data;
Means for determining a category for classifying the new target data according to the evaluation of the category based on the calculated similarity;
The server according to claim 13, further comprising:   Means for storing the target data in association with a plurality of the category data calculates a similarity between the target data and the keyword data, and stores the candidate category data according to a category evaluation based on the calculated similarity. The server according to claim 13 or 14, wherein the server is selected. Means for extracting and storing keyword data included in the target data in association with the determined category data, and further storing selection probability data based on the number of selections of the category;
The server according to claim 14 or 15, wherein an evaluation of a category associated with the keyword data is based on the similarity and the selection probability data. A program that causes a server connected to a terminal device via a communication network to classify target data to be viewed by a user of the terminal device into a category,
Storing the target data in association with a plurality of candidate category data;
In response to receiving data representing one selection of the plurality of candidate category data by the user from the terminal device, transmitting target data associated with the selected candidate category data;
In response to receiving data representing selection of the target data by the user from the terminal device, the user counts up the number of selections of candidate category data selected by the user, and the selected target data and Storing in association with the selected candidate category data;
Determining the category for classifying the target data based on the selection frequency data in a predetermined period;
A program that executes   18. The program according to claim 17, wherein the category to be determined is one in which the candidate category data is selected a predetermined number of times or more.   18. The program according to claim 17, wherein the category to be determined has a number of selections of the candidate category data equal to or higher than a predetermined order.   The program according to claim 17, wherein the category to be determined is determined by relative evaluation of the number of selections of the candidate category data.   21. The program according to claim 17, further comprising a step of extracting and storing keyword data included in the target data in association with the determined category data. In response to accepting registration of new target data different from the target data, calculating the similarity between the new target data and the keyword data;
Determining a category for classifying the new target data in accordance with a category evaluation based on the calculated similarity;
The program according to claim 21, further comprising:   The step of storing the target data in association with a plurality of the category data causes the similarity between the target data and the keyword data to be calculated, and the candidate category data is determined according to the category evaluation based on the calculated similarity. The program according to claim 21 or 22, wherein the program is selected. The step of extracting and storing the keyword data included in the target data in association with the determined category data further stores the selection probability data based on the number of selections of the category,
The program according to claim 22 or 23, wherein an evaluation of a category associated with the keyword data is based on the similarity and the selection probability data.

Images