JP2005196415A - Information recommendation program, server, and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique that enables real-time processes even if the number of object contents increases, while keeping the accuracy of recommending information from decreasing. <P>SOLUTION: This information recommendation program causes a computer to execute (a) a procedure for reading a user profile including one or two or more user keywords and user keyword values; (b) a procedure for reading an information profile including the one or two or more information keywords of one or two or more pieces of information included in a group of object information and their information keyword values; (c) a procedure for extracting top n pieces of user keywords with the greater keyword values from the user profile; (d) a procedure for searching for and extracting an information profile including as information keywords the same keywords as the top n pieces of user keywords; (e) a procedure for multiplying the keyword values of the top n pieces of user keywords by the information keyword values of the same information keywords as the top n pieces of user keywords and for accumulating the results; and (f) a procedure for extracting information with top m pieces of information profiles with greater cumulative values as recommended information. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a technique for recommending information in accordance with a user's preference in real time, and more particularly to an information recommendation program, an information recommendation server, and an information recommendation method using the technique.

One of the information recommendation methods currently proposed is to form a keyword vector for the content from the text information of the content, and also to form a keyword vector for the user profile and take the inner product of both keyword vectors, for example. In some cases, content is extracted after calculating (see, for example, Patent Document 1). However, such calculation of similarity and content extraction are becoming difficult for real-time processing as the target content increases.
JP 2002-215665 A

  An object of the present invention is to provide an information recommendation program, an information recommendation server device, and an information recommendation method that can perform real-time processing even if the target content increases and do not impair the accuracy of information recommendation.

  A first feature of the present invention is an information recommendation program, in which a computer includes a user profile including (a) one or more user keywords and a user keyword value corresponding to each user keyword. (B) a procedure for reading an information profile including one or more information keywords of one or more information included in the target information group and an information keyword value corresponding to each information keyword; ) A procedure for extracting the top n user keywords having a large user keyword value from the user profile; and (d) information for searching and extracting information profiles including the same keyword as the top n user keywords as information keywords. Profile search procedure, (e) top n user keyword keyword values, top n user keywords, A step of multiplying and integrating the information keyword values of the same information keyword and (f) a step of extracting information having the top m information profiles having a large integrated value obtained by the integration as recommended information. (N and m are both natural numbers).

For example, a user profile rp whose (a) (user keyword, user keyword value) is (rkw1, rkwv1), (rkw2, rkwv2), (rkw3, rkwv3) is read.

(B) an information profile jp1 whose (information keyword, information keyword value) is (jkw11, jkwv11), (jkw12, jkwv12), (jkw13, jkwv13);
(Jkw21, jkvv21), (jkw22, jkwv22), (jkw23, jkwv23) and the information profile jp2,
The information profile jp3 of (jkw31, jkwv31), (jkw32, jkwv32), (jkw33, jkwv33) is read.

(C) When rkwv1> rkwv2> rkwv3, rkw1 and rkw2 are extracted from the user profile as the top two user keywords with the largest user keyword values.

(D) When rkw1, jkw11, and jkw21 are the same, and rkw2, jkw12, and jkw22 are the same, the information profiles jp1 and jp2 are extracted as information profiles that include the same keyword as the user keywords rkw1 and 2 as information keywords. .

(E) (rkwv1 * jkvv11) + (rkwv2 * jkwv12) = V1,
(Rkwv1 * jkwv21) + (rkwv2 * jkwv22) = V2 is obtained.

(F) Information profile (jkw11, jkwv11), (jkw12, jkwv12), (jkw13, jkwv13) having the largest integrated value when V1> V2
Is extracted as recommendation information.

  The second feature of the present invention is that the information profile includes, as an information keyword, a keyword that is included in the top n user keywords and that is included in the top p appearance frequencies in the target information group in the information profile search procedure. Is searched and extracted (p is a natural number).

The third feature of the present invention is that (a) a procedure for reading an information profile including one or more information keywords of information referred to by a user and an information keyword value corresponding to the information keyword, (b) a user For each keyword value,
(1-update coefficient) * user keyword value + overall normalization scale value * update coefficient * information keyword value / procedure for updating with a value obtained by the number of information keywords is to further execute (the update coefficient is 1 Less than).

  According to a fourth aspect of the present invention, a procedure for deleting an information keyword whose information keyword value is less than a predetermined value and forming an information profile using the keyword value of the information keyword remaining without being deleted is further executed. There is.

  According to the first feature of the present invention, it is possible to quickly recommend information useful for a user by determining recommendation information using a user keyword having a large user keyword value.

  According to the second feature of the present invention, by using a keyword having a large user keyword value and having a high appearance frequency in the target product group as an information keyword, determining recommendation information, It becomes possible to recommend information useful for the user more quickly.

  According to the third aspect of the present invention, when the user actually refers to the information, the user profile is updated under the influence of the referenced information, so that the user profile always reflects the user's preference. Will be. For this reason, even if a user's preference changes, it becomes possible to always recommend the information according to a user's preference.

  According to the fourth feature of the present invention, it is useful for the user by deleting the information keyword with low importance and further determining the recommended information using the value of the information keyword remaining without being deleted. It becomes possible to recommend information more quickly.

  FIG. 1 shows the overall configuration of an information recommendation system to which the present invention is applied. This system allows client computers (PCs) 1 and 2 used by users to view product information using a web browser, and web browsers to view information on optimal products by sending HTML files. A Web server 8 that makes a state of interest, and an information recommendation server 10 that recommends optimum information to each user based on user profile data and the like. The client personal computers 1 and 2 and the Web server 8 can transmit and receive data via the communication network 5 such as the Internet.

  The information recommendation server 10 includes an input device 11 used for inputting product profile data and the like, a display device 12 used for confirming the contents of data input at the time of data input, and data A transmission / reception unit 13 that can transmit / receive data, an input / output control unit 14 that controls input from the input device 11, output to the display device 12, input from the transmission / reception unit 13, output to the transmission / reception unit 13, and the like. The apparatus 20 is provided with the main control part 15 which controls each component.

  The input device 11 includes a keyboard and a pointing device. Pointing devices include mice, trackballs, trackpads, sticks, and the like. The display device 12 includes a CRT display, a liquid crystal display, an EL display, an organic EL display, a plasma display, and the like. The transmission / reception unit 13 includes a network interface card (NIC) and the like.

  The storage device 20 is a hard disk (HD), a flexible disk (FD), an optical disk, a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), etc., a product profile database (DB), keyword statistics An information DB, a user profile DB, and a recommended rule DB are stored. Although not shown, a basic input / output system (BIOS), an operating system (OS), and the like are also stored.

  FIG. 2 shows an example of the product profile data structure. As shown in FIG. 2, the product profile data includes a product profile basic information table and a product profile basic keyword table.

  The product profile basic information table illustrated in FIG. 2A includes a product ID, a product name, a profile category type, and product classification types 1, 2, and 3. The profile category type indicates a profile category when a user profile is formed. The product classification types 1, 2, and 3 are classification information necessary for making a more appropriate recommendation.

  The product profile basic keyword table shown in FIG. 2B has a product ID and a product profile keyword (abbreviated as “product keyword” as appropriate). One or more product keywords exist for a single product ID.

  FIG. 3A shows an example of the user profile data structure. As shown in FIG. 3A, the user profile data includes a user ID, a profile category type, a user profile keyword (appropriately referred to as “user keyword”) and a user profile keyword value (appropriately “ Abbreviated as “user keyword value”). The user keyword value indicates the weight of the user keyword in the user profile. There are one or more profile category types for a single user ID. There are one or more user profile keywords for a single profile category type. There is a single user profile keyword value for a single user profile keyword.

  FIG. 3B shows an example of user profile data. As shown in FIG. 3B, for the user ID: 100, there are “product for women” and “product for men” as profile category types. For the combination of “user ID: 100, profile category type: product for women”, user profile keywords include “skirt”, “gather”, and “hip”. For the combination of “user ID: 100, profile category type: product for men”, the user profile keyword includes “rear” and “exterior”. The “keyword, keyword value” is “skirt, 800”, “gather, 630”, “hip, 20”, “rear, 940”, “exterior, 550”.

  FIG. 4 shows an example of the keyword statistical information data structure. As shown in FIG. 4, the keyword statistical information data includes a keyword and a keyword appearance frequency. The keyword is an element constituting the product profile. The appearance frequency is the number of products including the keyword in the profile.

  The recommended rule is a rule that is referred to during recommendation information determination processing. For example, when searching for the product profile of the target product group with the upper keyword having a large keyword value in the user profile, only the upper “three” (from No. 1 to No. 3) are used as search keywords. It is not used after the second.

  An overall image of processing in this embodiment will be described.

  First, as shown in FIG. 5A, as preprocessing, product data is input (step S10), product data is recorded (step S20), a product profile is formed (step S30), and keyword statistical information is stored. Generate (step S40). The formation of the product profile and the generation of the keyword statistical information will be described in detail later.

  Next, as shown in FIG. 5B, user action information is collected (step S50), and a user profile is formed (step S60). The product profile formed in step S30 is used for forming the user profile. The formation of the user profile will be described in detail later.

  Further, as shown in FIG. 5C, recommended information is determined (step S70), and recommended information is displayed (step S80). For the determination of recommended information, the product profile formed in step S30, the keyword statistical information generated in step S40, and the user profile formed in step S60 are used. The determination of recommended information will be described in detail later. The information recommendation server determines “recommendation information”. The Web server provides the client PC with an HTML file associated with the “recommendation information” and capable of displaying the “recommendation information”. The HTML file is interpreted by a Web browser running on the client PC, and “recommendation information” can be browsed.

(About product profile formation processing)
Based on FIG. 6, the product profile forming process will be described.

  Step S101: Product related content is input. The product related content includes a product ID, a product name, a profile category type, and a keyword.

  Step S102: Extract a keyword vector. As a keyword vector extraction method, a TF / IDF method is mainly used. A weight value generated for each keyword by the IT / IDF method is used as a keyword value.

  The keyword vector is composed of “keyword” and “keyword value”. Each product ID has one or more keywords, and each keyword has a keyword value.

FIG. 7A shows an example of the extracted keyword vector. As shown in FIG. 7A, for example, a keyword vector of a product having a product ID of “12531” is
(Keyword, keyword value) = (frill, 5), (warm, 4), (feminine, 4), (white, 2)
It becomes.

Step S103: Discard the keyword vector with low regularity. For example, FIG. 7B shows an example in which keywords whose keyword value is “2” or less are removed from the keyword vector of FIG. As shown in FIG. 7B, the keyword “white” with the keyword value “2” is removed for the product with the product ID “12531”. The keyword vector of the product whose product ID is “12531” is
(Keyword, keyword value) = (frills, 5), (warm, 4), (feminine, 4)
It becomes.

The keyword to be removed is not limited to a keyword having a keyword value of “2” or less. Keywords with a weight of “3” or less may be removed. When a keyword with a keyword value of “3” or less is removed, the keyword vector of the product whose product ID is “12532”
(Keyword, keyword value) = (UV, 5), (shape memory, 5)
It becomes only.

Instead of removing keywords whose keyword values are less than or equal to a predetermined value, it is also possible to remove keywords whose keyword value is a predetermined number or more from the top for each product. For example, when removing the keyword whose keyword value is the fourth or higher from the top, the keyword vector of the product whose product ID is “12533”
(Keyword, keyword value) = (warm, 4), (feminine, 5), (wool, 3)
become. In addition, when removing the keyword whose keyword value is the fifth or higher from the top, the keyword vector of the product whose product ID is “12534”
(Keyword, keyword value) = (water repellency, 3), (long, 3), (rain, 2), (cold, 2)
become.

  Step S105: The profile category type and the product classification type of each product are determined. FIG. 8 shows an example of the profile category type and the product category type. As shown in FIG. 8, a large category such as “product for women” and “product for men” is entered as the profile category type, and “middle / senior”, “hand-washing”, “dry”, etc. are classified as product category types. Enter the appropriate type.

  Step S106: The product profile information is stored in the product profile DB. The product profile formed as a result of the processing as described above is stored in the product profile DB, and is used in keyword statistical information generation processing, user profile update processing, and recommendation information determination processing described later.

(About keyword statistics generation processing)
Based on FIG. 9, keyword statistical information generation will be described.

  Step S201: Acquire product profile data.

  Step S202: Extract keyword statistical information from the product profile data. The keyword statistical information includes, for example, the keyword appearance frequency.

  FIG. 10 shows the keyword appearance frequency related to the keyword set in FIG. As shown in FIG. 10, the keyword “Frill” is used only for the product ID: 12531, so the appearance frequency is “1”, and the keyword “warm” is related to the product IDs: 12531, 12533, 12535. Since it is only used, the appearance frequency is “3”.

  Step S203: Save the keyword statistical information in the keyword statistical information DB.

(User profile formation process)
Based on FIG. 11, a user profile formation process will be described. In this example, a process for forming a user profile when a user purchases a product on the Web service or refers to product information is described.

Step S301: The product profile A of the product referenced by the user is acquired from the product profile DB. The product profile A is, for example, (keyword, keyword value) = (LSI 2.0, America 3.0).

Step S302: The user profile of the user is acquired from the user profile DB. The user profile P is, for example, (keyword, keyword value) = (computer 400, communication 350, wireless 150, LSI 100).

Step S303: Each keyword value of the user profile P is divided by the overall normalized scale value T to obtain a user profile P ′. When T = 1000, the user profile P ′ is
(Keyword, keyword value) = (computer 0.4, communication 0.35, wireless 0.15, LSI 0.1).

The overall normalization scale value T is a total value of all keyword values of the user profile P. User profile P is
When (keyword, keyword value) = (computer 400, communication 350, wireless 150, LSI 100),
(400 + 350 + 150 + 100) = 1000
Therefore, the overall normalization scale value T is “1000”.

Step S304: The value of each keyword of the user profile P ′ is multiplied by (1−update coefficient α) to obtain the user profile P ″. If the update coefficient α = 0.2, (1−update coefficient α) = 0.8, the user profile P ″ is
(Keyword, keyword value) = (computer 0.32, communication 0.28, wireless 0.12, LSI 0.08).

  The update coefficient α is not limited to “0.2”.

When importance is placed on the user's long-term preference, it is preferable that the update coefficient is small, so that it is set to “0.1”, for example. This increases the influence of the selected product profile on the user profile. On the other hand, when the user's short-term preference tendency is emphasized, the larger update coefficient is preferable, and for example, “0.3” is set. As a result, the influence of the selected product profile on the user profile is reduced.

Step S305: A value obtained by dividing the update coefficient α by the number N of keywords of the product profile A is multiplied by each keyword value of the product profile A to obtain a product profile A ′. Since the update coefficient α = 0.2 and the number of keywords N of the product profile A = 2,
(Update coefficient α / number of keywords in product profile N) = 0.1,
Product profile A '
(Keyword, keyword value) = (LSI 0.1, America 0.1).

  Step S306: The keyword value of the user profile P "is added to the keyword value of the product profile A 'to obtain a user profile P".

User profile P ''
(Keyword, keyword value) = (computer 0.32, communication 0.28, wireless 0.12, LSI 0.08),
Product profile A '
Since (keyword, keyword value) = (LSI 0.1, America 0.1),
User profile P '''
(Keyword, keyword value) = (computer 0.32, communication 0.28, radio 0.12, LSI 0.18, USA 0.1).

  Step S307: Each keyword value of the user profile P "" is multiplied by the overall normalization scale value T to obtain a user profile P "".

User profile P '''
(Keyword, keyword value) = (computer 0.32, communication 0.28, radio 0.12, LSI 0.18, USA 0.1),
Since the overall normalization scale value T = 1000,
User profile P ''''
(Keyword, Keyword value) = (Computer 320, Communication 280, Wireless 120, LSI 180, USA 100).

  Step S308: The user profile P "" "is stored in the user profile DB as an updated user profile.

Through these series of processes, the user profile reflects the product profile of the product referred to by the user (keyword, keyword value) = (computer 400, communication 350, wireless 150, LSI 100) from (keyword, keyword value). ) = (Computer 320, Communication 280, Wireless 120, LSI 180, USA 100).

  Therefore, it is possible to update the user profile sequentially while appropriately following changes in the user's preference.

  Although the above description has been described as “update” of the user profile, the initial value of the user profile can also be formed in a similar manner.

For example, in the initial state where there is no user profile, product profile A
(Keyword, keyword value) = (LSI 1.0, US 1.0)
Suppose you browsed the product.

In this case, since the user profile P in step S302 is “0”,
The user profile P ′ in step S303 is also “0”,
The user profile P ″ in step S304 is also “0”.

The product profile A in step S301 is (keyword, keyword value) = (LSI 1.0, America 1.0).
Because
The product profile A ′ in step S305 is
(Keyword, keyword value) = (LSI 0.1, USA 0.1)
And
The user profile P ′ ″ in step S306 is (keyword, keyword value) = (LSI 0.1, USA 0.1).
So that the keyword value ratio remains at 0.1: 0.1 and the keyword value sum is 1000, multiply by 5000,
The user profile P ″ ″ in step S307 is (keyword, keyword value) = (LSI 500, USA 500).
It becomes.

(Recommended information determination process)
Based on FIG. 12, the recommendation information determination process will be described.

Step S401: A user profile of a user is acquired. User profile is
It is assumed that (keyword, keyword value) = (computer 300, communication 280, LSI 200, wireless 120, USA 100).

  Step S402: The keyword statistical information of the target product group is acquired. The target product group refers to a set of products having the same category type as the category type stored in the user profile.

  For example, it is assumed that a certain user has two user profiles, the profile category type of one user profile is category 001, and the profile category type of the other user profile is category 002. When the user accesses a certain Web page, the products that can be browsed on the Web page are those whose profile category type of the product profile is category 001 and category 003.

  In such a case, in step S402, only products having a category type that matches one of the user's profile category types are included in the target product group. That is, in the above example, only products whose profile category type of the product profile is category 001 are included in the target product group.

The keyword statistical information in Example 1 is
(Keyword, appearance frequency) = (USA 30, Japan 20, computer 20, LSI 15, network 10, light 10).

  Step S403: Search the keyword statistical information with the upper keywords of the user profile. For example, if the number of upper keywords is “3”, the highest keyword “computer”, the second keyword “communication”, and the third keyword “LSI” are search keywords.

  The number of upper keywords is not limited to “3”. This number depends on how much recommendation information can be narrowed down. For example, if it is performed in 3 and hundreds of corresponding products match, the number should be increased, and conversely if no one matches, the number should be decreased.

Step S404: The keyword statistical information is searched with the upper keyword in the user profile. For example, if the high-order keyword in the user profile is “computer, communication, LSI”, the keyword in the keyword statistical information that matches these and the appearance frequency thereof are:
(Keyword, appearance frequency) = (Computer 20), (LSI 15).

  Step S405: Search for and extract a product that includes the keyword acquired in Step S404 in the product profile.

  Step S406: The product profile of the product extracted in step S405 is acquired. For example, the following product profiles of products A to C are acquired.

Product profile of product A (keyword, keyword value) = (computer 1.0, light 2.0, Japan 1.0)
Product profile for product B (keyword, keyword value) = (LSI 2.0, Wireless 1.0, America 3.0)
Product profile of product C (keyword, keyword value) = (computer 2.0, LSI 1.0)
Step S407: The degree of coincidence between the product profile acquired in step S406 and the user profile is calculated. Specifically, the keyword values of keywords that exist in common in both profiles are multiplied, and the sum of the obtained values is obtained.

For example, first, the degree of coincidence between the user profile and the product profile of the product A is calculated. The only common keyword in both profiles is the computer,
The computer keyword value in the user profile is "300"
Since the keyword value of the computer in the product profile of product A is “1.0”, the degree of coincidence is 300 * 1.0 = 300.

Next, the degree of coincidence between the user profile and the product profile of the product B is calculated. There are three common keywords in both profiles: "LSI", "wireless", and "USA"
The LSI keyword value in the user profile is “200”, the wireless keyword value is “120”, the American keyword value is “100”,
Since the LSI keyword value in the product profile of the product B is “1.0”, the wireless keyword value is “1.0”, and the American keyword value is “1.0”,
The degree of coincidence is 200 * 2.0 + 120 * 1.0 + 100 * 3.0 = 820.

Further, the degree of coincidence between the user profile and the product profile of the product C is calculated. There are two common keywords in both profiles: "Computer" and "LSI"
The computer keyword value in the user profile is “300”, the LSI keyword value is “200”,
Since the keyword value of the computer in the product profile of the product B is “1.0” and the keyword value of the LSI is “1.0”,
The degree of coincidence is 300 * 2.0 + 200 * 1.0 = 800.

  Therefore, the coincidence degree of the product A is 300, the coincidence degree of the product B is 820, and the coincidence degree of the product C is 800.

  Step S408: A product having a high degree of coincidence is set as a recommended product. For example, when the top two products are recommended products, the products B and C are picked up as recommended products. Note that the recommended products are not limited to “top two”.

  As described above, recommendation information is determined based on the user profile and keyword statistical information. Using only the top keywords in the user profile, calculate the degree of coincidence between the user profile and the product profile, and recommend only products with a high degree of coincidence to quickly recommend information according to user preferences It becomes possible.

  In the above embodiment, the optimum “product” along the user profile is recommended. However, the present invention is not limited to the case of recommending “product”. It can also be used when recommending information such as “news” instead of “product”. In that case, “news / keyword” and “news / keyword value” are set for each news.

It is a figure which shows the whole structure of the information recommendation system to which this invention is applied. (A) shows an example of a product profile basic information table, and (b) shows an example of a product profile basic keyword table. (A) shows an example of user profile data structure, (b) shows an example of user profile data. An example of a keyword statistical information data structure is shown. (A) is a flowchart showing the flow of pre-processing, (b) is a flowchart showing the flow of user profile formation processing, and (c) is a flowchart showing the flow of recommendation information determination processing. It is a flowchart which shows the flow of a product profile formation process. (A) shows an example of the extracted keyword vector, and (b) shows an example in which a keyword having a weight of “2” or less is removed from the keyword vector of FIG. An example of a profile category type and a product category type is shown. It is a flowchart which shows the flow of keyword statistical information generation. The keyword appearance frequency regarding the keyword set of FIG.7 (b) is shown. It is a flowchart which shows the flow of a user profile formation process. It is a flowchart which shows the flow of a recommendation information determination process.

Explanation of symbols

1 ... client PC, 2 ... client PC, 5 ... communication network,
8 ... Web server, 11 ... Input device, 12 ... Display device,
13: Transmission / reception unit 14 ... Input / output control unit 15: Main control unit (CPU),
20 ... storage device,

Claims (12)

On the computer,
A procedure for reading a user profile including one or more user keywords and a user keyword value corresponding to the user keyword;
A procedure for reading an information profile including one or more information keywords of one or more pieces of information included in the recommendation target information group and an information keyword value corresponding to the information keyword;
A procedure for extracting the top n user keywords having a large user keyword value from the user profile;
An information profile search procedure for searching and extracting the information profile including the same keyword as the top n user keywords as the information keyword;
Multiplying and multiplying the keyword value of the top n user keywords by the information keyword value of the same information keyword as the top n user keywords;
A procedure for extracting, as recommendation information, information having upper m information profiles having a large integrated value obtained by the integration;
Information recommendation program (n and m are both natural numbers).   In the information profile search procedure, the information profile including, as the information keyword, a keyword included in the top n user keywords and having an appearance frequency in the top p in the target information group is searched, The information recommendation program according to claim 1, wherein p is a natural number. On the computer,
A procedure for reading an information profile including one or more information keywords of information referred to by a user and an information keyword value corresponding to the information keyword;
Each of the user keyword values is
3. The procedure of updating with (1-update coefficient) * user keyword value + total normalized scale value * update coefficient * value obtained by the information keyword value / number of information keywords. Information recommendation program (update coefficient is less than 1). On the computer,
The procedure for deleting the information keyword whose information keyword value is less than a predetermined value and forming the information profile using the keyword value of the information keyword remaining without being deleted is further executed. 3. The information recommendation program according to 3. Means for reading a user profile including one or more user keywords and a user keyword value corresponding to the user keyword;
Means for reading an information profile including one or more information keywords included in the recommendation target information group and an information keyword value corresponding to the information keyword;
Means for extracting the top n user keywords having a large user keyword value from the user profile;
Information profile search means for searching for and extracting the information profile including the same keyword as the top n user keywords as the information keyword;
Multiplying and multiplying the keyword value of the top n user keywords by the information keyword value of the same information keyword as the top n user keywords;
Means for extracting, as recommended information, information having the top m information profiles having a large integrated value obtained by the integration;
An information recommendation server (where n and m are natural numbers).   The information profile search means searches the information profile including, as the information keyword, a keyword included in the top n user keywords and having an appearance frequency in the top p in the target information group, The information recommendation server according to claim 5, wherein p is a natural number. Means for reading an information profile including one or more information keywords of information referred to by a user and an information keyword value corresponding to the information keyword;
Each of the user keyword values is
7. The method further comprising: (1−update coefficient) * user keyword value + total normalized scale value * update coefficient * updated with a value obtained from the information keyword value / number of information keywords. Information recommendation server (update coefficient is less than 1).   6. The apparatus according to claim 5, further comprising means for deleting the information keyword whose information keyword value is less than a predetermined value and forming the information profile using the keyword value of the information keyword remaining without being deleted. 7. The information recommendation server according to 7. Computer
A procedure for reading a user profile including one or more user keywords and a user keyword value corresponding to the user keyword;
A procedure for reading an information profile including one or more information keywords of one or more information included in the target information group and an information keyword value corresponding to the information keyword;
A procedure for extracting the top n user keywords having a large user keyword value from the user profile;
An information profile search procedure for searching and extracting the information profile including the same keyword as the top n user keywords as the information keyword;
Multiplying and multiplying the keyword value of the top n user keywords by the information keyword value of the same information keyword as the top n user keywords;
A procedure for extracting, as recommendation information, information having upper m information profiles having a large integrated value obtained by the integration;
Information recommendation method (n and m are both natural numbers).   In the information profile search procedure, the information profile including, as the information keyword, a keyword included in the top n user keywords and having an appearance frequency in the top p in the target information group is searched, The information recommendation method according to claim 9, wherein p is a natural number. Computer
A procedure for reading an information profile including one or more information keywords of information referred to by a user and an information keyword value corresponding to the information keyword;
Each of the user keyword values is
11. The procedure for updating with (1-update coefficient) * user keyword value + total normalized scale value * update coefficient * value obtained by the information keyword value / number of information keywords is further executed. Information recommendation method (update coefficient is less than 1). Computer
10. The procedure further comprising: deleting the information keyword whose information keyword value is less than a predetermined value, and forming the information profile using the keyword value of the information keyword remaining without being deleted. 11. The information recommendation method according to 11.

Images