KR101565339B1 - Recommendation system using collective intelligence and method thereof - Google Patents

Abstract

A recommendation system and method using collective intelligence are disclosed. The recommendation system using collective intelligence includes a rating information collection unit for collecting rating information posted on the Internet related to a product in association with a user and a product that posted rating information; A similar tendency user cluster generating unit for measuring a degree of similarity between users using the rating information, and then generating a cluster in which users are clustered according to the degree of similarity between users; And a service providing unit for providing a recommendation service for the product to the service target based on the rating information of the user included in the cluster to which the service target party belongs.

Description

[0001] RECOMMENDATION SYSTEM USING COLLECTIVE INTELLIGENCE AND METHOD THEREOF [0002]

Embodiments of the present invention relate to a recommendation system and method for providing products recommended by people with similar tendencies using collective intelligence.

When consuming cultural properties such as movies and novels, relative evaluation based on taste may play a more important role than absolute evaluation.

If you use movie reviews or book reviews left on portal sites or internet bookstores, you can easily see the public's evaluation of the work, but it's hard to know how people with similar tastes have evaluated it.

The recommendation system using collective intelligence is applied to many fields, but it is difficult to build evaluation database of many users. Also, in order to use the service properly, there is an entry barriers to input my information (for example, my favorite products) first.

There is provided a recommendation system and method using collective intelligence capable of collecting evaluation documents in which individuals' ratings of products are recorded in a micro blog, a rating site of a portal site, and building a product evaluation database of a group.

There is provided a recommendation system and method using collective intelligence that can present product information recommended by users having similar tendencies using a product evaluation database of a group.

A rating information collection unit for collecting rating information posted on the Internet in association with a product in association with a user and a product that posted rating information; A similar tendency user cluster generating unit for measuring a degree of similarity between users using the rating information, and then generating a cluster in which users are clustered according to the degree of similarity between users; And a service providing unit for providing a recommendation service for the goods to the service target based on the rating information of the users included in the cluster to which the service target person belongs.

According to one aspect, a recommendation system using collective intelligence includes a document collection unit for collecting posted documents posted on the Internet; A document filtering unit for extracting a publication document related to a product among the publication documents; And an affirmative degree calculating unit for calculating a word positive degree with respect to the posted document. At this time, the rating information collection unit can collect word affinity as rating information of the product.

According to another aspect, the recommendation system using collective intelligence may further include a product keyword database that holds at least one keyword related to the product. At this time, the document filtering unit may extract a publication document including a word matching the keyword based on the product keyword database.

According to another aspect, the recommendation system using the collective intelligence includes positive and negative keywords assigned to positive keywords and positive keywords, affirmative / negative keywords corresponding to negative keywords, negative / negative keywords corresponding to negative keywords, And a negative keyword database. At this time, the positive likelihood calculator extracts a word matching the positive keyword or the negative keyword in the publication document based on the positive / negative keyword database, and calculates the word affinity using the positive word weight or the negative word weight corresponding to the extracted word .

According to another aspect, the rating information collection unit includes a user database for holding a unique key assigned to each site ID and a site ID of a user collected from at least one web site; And a rating information database for managing the rating information posted by the user based on the user database according to the ID of the product and the unique key value of the user. At this time, a unique key having a different value is assigned to the site ID having the same ID among the site IDs. However, if it is determined that the user is the same user through the authentication of the user, the unique key can be reassigned to the same value.

According to another aspect, the similarity degree between users is a value obtained by comparing the rating information of the same goods among users, and the similarity incidence user cluster generating unit includes a similarity database that maintains similarity with different users for each user; And a cluster database that manages users whose similarities between users are equal to or higher than a set value in the same cluster.

According to another aspect, the recommendation system using the collective intelligence is a similarity tendency in which users who have posted rating information on the Internet when the service target person is not included in the cluster are searched for users whose similarity degree with the service target is higher than the set value A user search unit; And a database updating unit for updating the degree of similarity of the service target person with the searched user in the similarity degree database and adding the service target person to the cluster database.

According to another aspect of the present invention, the service providing unit calculates a final rating for a product by dividing a total of weights of products obtained by multiplying the rating information of the user by the degree of similarity of the user with the rating information of the user for each product, A final rating calculation unit; And a product recommendation unit for recommending the product based on the final rating.

A recommendation method for recommending a product using collective intelligence, the method comprising: collecting rating information posted on the Internet related to a product, in association with a user and a product that posted rating information; A similarity-based user cluster generating step of generating a cluster in which users are clustered according to the degree of similarity between users after measuring the similarity between users using rating information; And a service providing step of providing a recommendation service for a product to the service target based on the rating information of the user included in the cluster to which the service target person belongs.

A microblog used by a rating site of a portal site or a microblog used by an individual is collected to construct a product evaluation database of a group to provide a recommendation service based on a rating of a user having a similar tendency even if the service target person does not input separate taste information . Therefore, by using the database of collective intelligence, it is possible to obtain product information recommended by users having similar tastes to the service target person.

FIG. 1 is a block diagram illustrating an internal configuration of a recommendation system using collective intelligence for providing a recommendation service by constructing a product evaluation database of a similarity group according to an exemplary embodiment of the present invention. Referring to FIG.
2 is a diagram for explaining an example of collecting rating information.
3 is a block diagram illustrating the structure of a rating information collecting unit that collects rating information posted on the Internet and users who have published rating information and converts the rating information into a database in an embodiment of the present invention.
4 is a block diagram illustrating an additional configuration of a recommendation system using collective intelligence for converting rating information and a user into a database through filtering of a publication document collected in a microblog in one embodiment of the present invention.
FIG. 5 is a block diagram illustrating the configuration of a similar tendency user cluster generation unit in which similarity tendency users are clustered in an embodiment of the present invention. FIG.
6 is a diagram for explaining an example of clustering users of similar tendencies.
FIG. 7 is a block diagram illustrating an additional configuration of a recommendation system using collective intelligence to update a database by searching for users who are similar to the users that are not included in the database, according to an exemplary embodiment of the present invention.
8 is a block diagram illustrating the configuration of a service providing unit that recommends a product using rating information of a user having a similar tendency in an embodiment of the present invention.
FIG. 9 is a flowchart illustrating a recommendation method using a collective intelligence for providing a recommendation service by constructing a product evaluation database of a similarity group according to an exemplary embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating an internal configuration of a recommendation system using collective intelligence to provide a recommendation service by building a product evaluation database of a group showing similar tendencies in an embodiment of the present invention. 1 shows a recommendation system 100 using a collective intelligence for recommending a product to a service target based on an evaluation of the public about the commodity.

In the present specification, the term 'commodity' may refer not only to cultural properties such as movies, dramas and novels, but also to objects or services for sale.

The recommendation system 100 according to an exemplary embodiment may include a rating information collection unit 110, a similar tendency user cluster generation unit 120, and a service provider 130.

The rating information collecting unit 110 collects and stores rating information posted on the Internet in association with a product in association with a user who is a publisher and a corresponding product. The rating information collecting unit 110 collects rating information from a rating site that is a product introduction site (for example, a movie introduction site) that introduces a product or a rating information from a publication document posted on a microblog And the rating information for each user can be constructed as a database by a method of extracting and collecting.

The similar tendency user cluster generating unit 120 plays a role of creating a user cluster clustering users according to the degree of similarity between users measured based on the rating information for each user. Likewise, the similar tendency user cluster generating unit 120 means a means for storing similarity between users and a user cluster. That is, the similarity-based user cluster generating unit 120 can maintain and manage the similarity with different users for each user, and maintain and manage similarity-oriented users in the same cluster.

The service providing unit 130 may provide a recommendation service for the product to the service target based on the rating information of the users in the cluster to which the service target person belongs. In other words, the service providing unit 130 can use the rating information of a user having a similar tendency to the service target through the cluster to service a product recommended by users of similar tendencies.

The configuration and operation of the recommendation system using collective intelligence according to an embodiment will be described in detail with reference to FIGS. 2 to 8. FIG.

Rating information Collecting section

First, we explain how to collect rating information for each user in a rating site.

2, the rating information collecting unit 110 collects rating information such as a user's site ID 201 and a rating value of the user's rating in the rating site 210, such as a product introduction page provided on a portal site, Information 202 may be collected. As shown in FIG. 2, when a web page having a rating in the rating site 210 is a plurality of pages, the rating information collecting unit 110 collects the user's site ID 201 and the rating information 202 while changing pages .

The rating information collection unit 110 constructs the collected rating information for each user as a database. As shown in FIG. 3, the rating information collection unit 300 may include a user database 310 and a rating information database 320.

As shown in Table 1, the user database 310 maintains and manages a site ID of a user collected from at least one web site and a unique key allocated to each site ID. The user database 310 basically assigns unique keys having different values to each other even if the site IDs are the same for each site, and if the user subsequently authenticates the user, the user database 310 regards the same user as the same user, Can be reallocated.

id Site ID (user) Site name Unique key One xlos naver One 2 xlos Yes24 One 3 chaehyun interpark One 4 nomad buxmovie 2

The rating information database 320 maintains and manages the rating information posted by the user in association with the product ID and the unique key value of the user based on the user database 310 as shown in Table 2. In other words, the rating information database 320 records ratings of each product that each user has built in the user database 310. In this case, the rating information is a value that is normalized and stored within -1 to 1. In particular, the closer the rating is to -1, the closer to 1 the negative opinion on the product means the value reflecting the positive opinion on the product.

id Product ID Unique key Grade One One One One 2 One 2 One 2 One 2 One 4 2 2 -0.4

The rating information database 320 may be constructed by collecting rating information of each user for each product or a certain number (n) of products may be arranged in descending order of ratings of users during a predetermined period (period changeable) The rating information database 320 can be constructed.

According to the above configuration, the rating information collection unit 300 may collect rating information from rating sites and store and maintain rating information for each user associated with each product in a database.

Next, we explain how to collect rating information for each user in a microblog.

4, the recommendation system 400 using collective intelligence according to an embodiment includes a document collection unit 410, a product keyword database 420, a document filtering unit 430, an affirmative / The negative keyword database 440, and the affirmity calculator 450 as additional configurations.

The document collecting unit 410 collects the site ID of the user and the posted document posted by the user at the microblogging site as shown in Table 3. Here, microblog is a type of social network service that can communicate with a plurality of people using a short message of about one or two sentences. It is a kind of social network service that includes overseas sites such as twitter, fanfou, me2day, (yozm) are the representative micro blogs.

id Site name Site ID
(user) Publishing document (text) Posted One me2day xlos Inception was really awesome !! 2010.08.02 12:22:40 2 twitter nomad Moss .. the worst movie 2010.07.22 22:15:23 3 me2day xlos Today is too sleepy ㅠ. 2010.08.01 10:12:40 4 twitter chae Theatrical Liar 2 Recommended ~~ 2010.05.05 12:15:23 5 me2day xlos I saw Mid-Damage today, and it was fun. haha 2010.08.02 10:00:15

The product keyword database 420 plays a role of maintaining at least one keyword related to a product. At this time, the product keyword database 420 can maintain and manage representative keywords and related keywords for each product in correspondence with each other as shown in Table 4. In addition, the product keyword database 420 may classify each product into categories and convert the product-specific keywords into a database.

id category keyword Related Keywords One movie Inception inception 2 movie Moss 3 play Liar 2 lier2 4 drama damages Damage, Damage

The document filtering unit 430 filters the posted document related to the product based on the product keyword database 420. In other words, the document filtering unit 430 extracts a publication document including a word matched with a keyword in the product keyword database 420 among the publication documents collected in Table 3. In Table 5, a shaded publication corresponds to a publication document related to a product, which is extracted through keyword matching.

id Site name Site ID
(user) Publishing document (text) Posted One me2day xlos Inception was really awesome !! 2010.08.02 12:22:40 2 twitter nomad Moss .. the worst movie 2010.07.22 22:15:23 3 me2day xlos Today is too sleepy ㅠ. 2010.08.01 10:12:40 4 twitter chae Theatrical Liar 2 Recommended ~~ 2010.05.05 12:15:23 5 me2day xlos I saw Mid-Damage today, and it was fun. haha 2010.08.02 10:00:15

The positive / negative keyword database 440 classifies affirmative and negative keywords to construct a database as shown in Table 6. The negative keywords and negative keywords are classified into positive keywords corresponding to positive keywords and positive keywords corresponding to positive keywords, negative keywords corresponding to negative keywords, And plays a role of maintaining adjective weight given to each keyword. The positive / negative keyword database 440 can be prepared by learning in advance through machine learning.

id Category Keywords (positive / negative) Weight One Positive (POSITIVE) recommendation One 2 Positive (POSITIVE) good One 3 Positive (POSITIVE) It is interesting 0.7 4 Positive (POSITIVE) Best One 5 NEGATIVE Shine -One 6 NEGATIVE worst -One 7 NEGATIVE Sleepy -0.5

The positive likelihood calculator 450 calculates the word affinity of the publication document of Table 5 extracted through filtering based on the positive / negative keyword database 440. The positive likelihood calculator 450 may calculate a word affinity using an affirmative weight or negative word weight corresponding to the extracted word after extracting a word matching the positive keyword or the negative keyword in the posted document. For example, the positive likelihood calculator 450 can calculate the word affinity of the posted document by using Equation (1).

The positive likelihood calculator 450 classifies the posted document into words of morpheme units and determines positive or negative word weights assigned to words matching the positive or negative keywords among the classified words as a keyword weight . The positive likelihood calculator 450 may calculate the sentence distance weight according to Equation (2).

In other words, the affirmity calculator 450 classifies a publication document as a sentence-by-phrase and counts the total number of phrases classified as sentence-by-sentence phrases, i.e., a phrase including a word matching a positive keyword or a negative keyword, And the sentence distance weight is calculated by dividing the value obtained by subtracting the distance between the phrases including the sentence distance.

For example, the publishing document was 'Inception'. If this is more fun than I thought ~ recommended ~ ', the word positivity is calculated in the following manner.

(1) Divide the posted document into morphemes.

<I saw the insception. This is more fun than I thought. I recommend it.> -> <[INSEPT], [VIEW], [THIS], [THOUGHT]

(2) Calculate the keyword weight.

[Fun] = 0.7

[Recommendation] = 1.0

(3) Split the posted document into sentences.

<I saw the insception. This is more fun than I thought. I recommend it ~> -> <[I saw the insception], [It is fun more than I thought], [I recommend]>

(4) Calculate sentence distance weight.

[Saw inception] = (3-0) / 3 = 1

[More fun than I thought] = (3-1) / 3 = 0.66

[Recommended] = (3-2) / 3 = 0.33

(5) The word affinity is finally judged using the keyword weight and the sentence distance weight.

Word positive = [fun] (0.7 * 0.66) + [recommended] (1.0 * 0.33) = 0.79

If the word positive degree calculated through the above process is positive, the rating of the product posted on the posted document is positive, and if it is negative, it can be determined that the rating of the product is negative.

In addition, the positive likelihood calculator 450 normalizes the sum of the word affinities per user for each product and the sum to within one to one. Here, the word affinity of the posted document can be used as the rating information of the user regarding the article related to the posted document.

Through the above-described configuration, the rating information collection unit 300 can construct a user database 310 as shown in Table 1, in which a unique key is assigned to each user collected from the microblog. In addition, the rating information collection unit 300 stores the word affinity of the publication document calculated by the affirmity calculation unit 450 as rating information of the corresponding product, The star rating information can be constructed in the rating information database 320 as shown in Table 2. [

Similarity User Cluster Generating unit

5 shows a similar tendency user cluster generating unit 500 in which similarity tendency users are clustered. As shown, the pseudo tendency user cluster generating unit 500 may include a similarity degree database 510 and a cluster database 520.

The similarity degree database 510 plays a role of maintaining and managing similarity with different users for each user. The similarity of users' propensity can be measured by various methods using Pearson correlation score, Euclidian distance, Jacquard coefficient, Manhattan distance, and so on. For example, a Pearson correlation score is a measure of how well two datasets are represented in a straight line. The Pearson correlation score can be used to measure the similarity between users. For example, after locating other users who have evaluated the same product as a particular user-rated product, the two users can calculate the Pearson correlation score of the two users by comparing the ratings that they have left on the product. Pearson correlation scores range from -1 to 1, of which 1 indicates that two people gave the same score to all products. That is, the similarity database 510 can construct the similarities between users calculated based on the product-specific rating information of users and other users as shown in Table 7.

User ID_1 User ID_2 Similarity One 2 0.8 One 3 0.9 One 4 0.1

The cluster database 520 plays a role of managing users having similarities between users to the same cluster or more in the same cluster. Since a large number of users exist in the rating information database 320 that maintains rating information for each product associated with each product and it takes a long time to calculate the similarity between the users each time, To update the similarity database 510, and to update the cluster database 520 by clustering users having a similar tendency. For example, when a certain number (n) of commodities are selected according to the order in which users are ranked high, the degree of similarity between users can be calculated for users who have rated n commodities. Referring to FIG. 6, users whose similarity value exceeds the set value by applying a set value to the similarity degree are connected to a graph 601 and a user 602 bound to the same graph 601 (602) Can be stored.

User ID Cluster ID One One 2 One 3 One

According to the above configuration, the pseudo-tendency user cluster generating unit 500 can construct a per-user cluster in the cluster database 520 so that users of similar tendencies can be managed through the cluster generated based on the degree of similarity between users .

service Offering

The recommendation system using collective intelligence according to an exemplary embodiment provides a function of adding a service target to a cluster when the service target is a user not included in the cluster by checking the cluster to which the service target is belonging before providing the recommended service can do.

As shown in FIG. 7, the recommendation system 700 using collective intelligence according to an embodiment may further include a similar-tendency user search unit 710 and a database update unit 720 as an additional configuration have.

The similarity user search unit 710 checks whether the service target person is a user included in the cluster in the cluster database 520 through personal information when the service target person inputs personal information (e.g., site ID). If the service target is not included in the cluster, the similarity-based user search unit 710 can search for users included in the cluster and find a user whose similarity with the service target is equal to or higher than a set value. In addition, if there is no user having a tendency similar to that of the service target among the users included in the cluster, the similarity orientation user search unit 710 may target the user having the rating information for each product, It is possible to find a user who posts the rating information for each product in real time and whose similarity with the service target is equal to or higher than a set value.

The database updating unit 720 plays a role of updating the similarity degree database 510 and the cluster database 520 with respect to the service target who is a new user. In other words, if a cluster in the cluster database 520 finds a user having a tendency similar to that of the service target, the service target is added to the cluster to which the user belongs, and a user having a tendency similar to the service target If it is found, the cluster database 520 can be updated by additionally generating a cluster including the user and the service target. Also, if the database updater 720 finds a user having a tendency similar to the service target, the similarity of the service target to the user may be additionally updated in the similarity database 510. [

In addition, the recommendation system using the collective intelligence according to an exemplary embodiment of the present invention provides a service recommendation service to a service target person using the similarity between users included in the cluster to which the service target person belongs, the user included in the cluster, .

As shown in FIG. 8, the service providing unit 800 may include a final rating calculation unit 810 and a product recommendation unit 820 in the recommendation system using collective intelligence according to an embodiment of the present invention .

The final rating calculation unit 810 calculates the similarity score between the users included in the cluster and the product-specific user score using the rating information database 320, the similarity level database 510, and the cluster database 520, And the like. At this time, the user list included in the cluster to which the service target person belongs, the similarity score to each user, and the rating information for each product can be arranged as shown in FIG.

user Similarity Moss
grade Moss
weight Inception
grade Inception
weight Liar
grade Liar
weight Lee Chae-hyun 0.99 0.90 0.89 0.60 0.59 0.70 0.69 Hong Gil Dong 0.38 0.40 0.15 0.40 0.15 1.00 0.38 Hong Gil Soon 0.78 0.70 0.55 0.90 0.70 Kim Cheol-Soo 0.67 0.80 0.54 0.90 0.60 weight
Sum 2.13 1.35 1.78 Similarity
Sum 2.82 2.04 2.15 Normalization
(Final rating) 0.75 0.66 0.83

In detail, the final rating calculation unit 810 calculates a weighted sum of the products obtained by multiplying the rating of each product by the degree of similarity, calculates the sum of similarities of the users who left the rating on the same product, To obtain a normalized value. At this time, the normalized value becomes the final rating for the product.

The product recommendation unit 820 recommends the product to the service target based on the final rating for each product. For example, when a service target person logs in to a site providing a recommended service, the product recommendation unit 820 divides the category into categories (for example, movies, novels, dramas, etc.) . As another example, when the service target person logs in to a site providing a recommendation service and selects a specific category, the commodity recommending unit 820 can recommend products corresponding to the category selected by the service target shop in descending order of the final rating. As another example, when the service target person searches for a product, the product recommendation unit 820 obtains the general rating information (i.e., the average of the ratings left by all the users) And the final score of each product calculated based on the rating information of the products. As another example, when the service target person searches for a product, the product recommendation unit 820 can recommend products recommended by users having a similar tendency to the service target person in relation to the product.

According to the above configuration, the service provider 800 calculates the final score for each product using the rating information of the user having a similar tendency to the service target person and the degree of similarity between the users, can do.

FIG. 9 is a flowchart illustrating a recommendation method using a collective intelligence for providing a recommendation service by constructing a product evaluation database of a similarity group according to an exemplary embodiment of the present invention. The recommendation method using collective intelligence according to the present embodiment can be performed by the recommendation system described with reference to FIG. 1 and FIG.

In step 910, the recommendation system establishes a rating information database by associating rating information posted on the Internet in association with the product with the user who is the publisher and the corresponding product.

As an example of building a rating information database, in step 911, the recommendation system can use a method of collecting rating information on a rating site, which is a product introduction site (for example, a movie introduction site). In other words, the user's site ID and the rating information remaining on the rating site can be collected from the rating site such as the product introduction page provided on the portal site.

As another example for constructing the rating information database, in step 912, the recommendation system may use a method of extracting rating information from the posted document posted on the microblog and collecting the rating information. In other words, the referral system collects the user's site ID and the posted document posted by the user at the microblogging site. In step 913, the recommendation system extracts a publication document including keywords related to the product among the published articles collected in step 912 based on the product keyword database holding at least one keyword related to the product. In step 914, the recommendation system is based on a positive / negative keyword database that maintains affirmative weight values assigned to positive keywords and positive keywords, negative keywords corresponding to negative keywords, and negative keywords assigned to negative keywords, , And calculates the word affinity of the posting document extracted through filtering in step 913. The recommendation system can extract a word matched to a positive keyword or a negative keyword in a publication document, and then calculate a word affinity using an affirmative weight or negative word weight corresponding to the extracted word. At this time, the word affinity of the posted document can be stored as rating information of the user for each product in building the rating information database.

In step 920, the recommendation system establishes a similarity database that maintains similarities with different users for each user, and a cluster database that maintains user clusters based on the similarity between users. For example, the recommendation system may find other users who have evaluated the same product as a product evaluated by a particular user, and then calculate the similarity between the two users by comparing the ratings of the two users on the product. In addition, the recommendation system can construct similarity degree with other users for each user as a database, and group users for each cluster into a database by grouping users whose similarities are equal to or higher than a set value. That is, the recommendation system can maintain and manage the similarity with other users for each user, and maintain and manage the users having similar tendency in the product rating by the same cluster.

In step 930, the recommendation system uses a rating information database established in step 910, and similarity database and cluster database established in step 920, The user can recommend a product recommended by the user. The recommendation system can calculate the final score for each product by extracting the similarity between the users and the rating information of the users according to the product through the cluster to which the service target person belongs, and then dividing the sum of the weights multiplied by the similarity degree by the rating information, . Therefore, the recommendation system can recommend the product to the service person in descending order of the final rating.

As described above, according to the embodiments of the present invention, it is possible to collect rating information for each user of a product in a micro blog used by a rating site of a portal site or individuals, and to build a product evaluation database of a group. Accordingly, even if the service target person does not input the preference information, the recommendation service based on the rating of the user having similar tendency to the service target person can be provided.

While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

110: rating information collection unit
120: pseudo-tendency user cluster generation unit
130: Service Offering

Claims (22)

A rating information collection unit for collecting rating information posted on the Internet related to a product in association with a user who posted the rating information and the product;
A similar tendency user cluster generating unit for measuring a degree of similarity between the users by using the rating information, and then generating a cluster in which the users are clustered according to the degree of similarity between the users; And
A service providing unit for providing a recommendation service for the product to the service target person based on the rating information of the user included in the cluster to which the service target person belongs,
Lt; / RTI &gt;
The rating information collecting unit,
A user database for holding a site ID of the user collected from at least one web site and a unique key allocated for each site ID; And
A rating information database for managing the rating information posted by the user on the basis of the user database according to the ID of the goods and the unique key value of the user,
Lt; / RTI &gt;
The user database comprises:
A unique key having a different value is assigned to a site ID having the same ID among the site IDs and if the same user is determined through authentication of the user, the unique key is reassigned to the same value
A recommendation system using collective intelligence. The method according to claim 1,
The rating information collecting unit,
Collecting the word affinity of the posted document posted on the Internet as the rating information of the goods
A recommendation system using collective intelligence. The method according to claim 1,
The recommendation system comprises:
A document collecting unit for collecting posted documents posted on the Internet;
A document filtering unit for extracting a publication document related to the product among the publication documents; And
A degree of affirmity calculating unit for calculating a word affinity for the posted document,
Further comprising:
The rating information collecting unit,
Collecting the word affinity as rating information of the goods
A recommendation system using collective intelligence. The method of claim 3,
The document filtering unit,
Extracting a publication document including a word matched with the keyword of the publication document based on a product keyword database holding at least one keyword related to the product
A recommendation system using collective intelligence. 3. The method of claim 2,
The recommendation system comprises:
A positive degree calculating unit for calculating the word positive degree,
Further comprising:
The affirmth-
And calculating the word affinity using the positive word weight of the positive keyword or the negative word weight of the negative keyword corresponding to the extracted word after extracting a word matching the positive keyword or the negative keyword in the publication document
A recommendation system using collective intelligence. 6. The method of claim 5,
The affirmth-
Calculating a keyword weight by classifying the posted document into words of morpheme units, calculating a sentence distance weight by classifying the posted document into sentence-based phrases, and then adding a value reflecting the sentence distance weight to the keyword weight, Judge the road,
The keyword weight may be expressed as:
The positive word weight or the negative word weight corresponding to each word of the morpheme unit,
The sentence distance weighting may be expressed as:
The number of the phrases constituting the publication document and the distance between the phrases including the positive keyword or the negative keyword to the phrase including the keyword related to the commodity
A recommendation system using collective intelligence. delete The method according to claim 1,
The degree of similarity between the users,
And comparing the rating information of the same product among the users,
Wherein the pseudo tendency user cluster generating unit comprises:
A degree of similarity database for maintaining a degree of similarity between different users for each user; And
A cluster database for managing users whose similarities between users are equal to or higher than a threshold value in the same cluster
A Recommendation System Using Collective Intelligence. 9. The method of claim 8,
The recommendation system comprises:
A similarity user search unit for searching a user who has posted the rating information on the Internet, if the service target person is not included in the cluster, a user whose similarity with the service target person is equal to or greater than the set value; And
A database updating unit for updating the degree of similarity of the service target person with the searched user to the similarity degree database and adding the service target person to the cluster database,
The recommendation system using collective intelligence. The method according to claim 1,
The service providing unit,
Recommending the product to the service target person using the similarity degree between the service target person and the user included in the cluster to which the service target person belongs and the rating information posted by the user included in the corresponding cluster
A recommendation system using collective intelligence. 11. The method of claim 10,
The service providing unit,
A final rating calculation unit for calculating a final rating for the product by dividing the total of weights obtained by multiplying the rating information of the user by the degree of similarity with the user by the sum of the inter-user similarities for the cluster to which the service target person belongs, ; And
And recommending the product based on the final rating
A Recommendation System Using Collective Intelligence. In a recommendation method for recommending a product using collective intelligence,
In the recommendation method,
A rating information collection step of collecting rating information posted on the Internet related to a product in association with a user who posted the rating information and the product;
A similarity-based user cluster generating step of generating a cluster in which the users are clustered according to the degree of similarity between the users after measuring the similarity between the users using the rating information; And
And a service providing step of providing a recommendation service for the goods to the service target person based on the rating information of the user included in the cluster to which the service target person belongs,
Wherein the rating information collection step comprises:
Constructing a user database that maintains a site ID of the user and a unique key assigned to each site ID collected from at least one web site; And
Establishing a rating information database for managing the rating information posted by the user based on the user database in accordance with the ID of the product and the unique key value of the user
Lt; / RTI &gt;
The user database comprises:
A unique key having a different value is assigned to a site ID having the same ID among the site IDs and if the same user is determined through authentication of the user, the unique key is reassigned to the same value
A recommendation method using collective intelligence. 13. The method of claim 12,
Wherein the rating information collection step comprises:
Collecting the word affinity of the posted document posted on the Internet as the rating information of the goods
A recommendation method using collective intelligence. 13. The method of claim 12,
In the recommendation method,
A document collection step of collecting a publication document posted on the Internet;
A document filtering step of extracting a publication document related to the product among the publication documents; And
An affirmative degree calculating step of calculating a word positive degree with respect to the posted document
Further comprising:
Wherein the rating information collection step comprises:
Collecting the word affinity as rating information of the goods
A recommendation method using collective intelligence. 15. The method of claim 14,
Wherein the document filtering step comprises:
Extracting a publication document including a word matched with the keyword of the publication document based on a product keyword database holding at least one keyword related to the product
A recommendation method using collective intelligence. 14. The method of claim 13,
In the recommendation method,
A positive degree calculating step of calculating the word positive degree
Further comprising:
Wherein the affirmth degree calculating step includes:
And calculating the word affinity using the positive word weight of the positive keyword or the negative word weight of the negative keyword corresponding to the extracted word after extracting a word matching the positive keyword or the negative keyword in the publication document
A recommendation method using collective intelligence. 17. The method of claim 16,
Wherein the affirmth degree calculating step includes:
Calculating a keyword weight by classifying the posted document into words of morpheme units, calculating a sentence distance weight by classifying the posted document into sentence-based phrases, and then adding a value reflecting the sentence distance weight to the keyword weight, Judge the road,
The keyword weight may be expressed as:
The positive word weight or the negative word weight corresponding to each word of the morpheme unit,
The sentence distance weighting may be expressed as:
The number of the phrases constituting the publication document and the distance between the phrases including the positive keyword or the negative keyword to the phrase including the keyword related to the commodity
A recommendation method using collective intelligence. delete 13. The method of claim 12,
The degree of similarity between the users,
And comparing the rating information of the same product among the users,
Wherein the pseudo-tendency user cluster generation step comprises:
Establishing a similarity database that maintains a similarity degree with respect to different users for each user; And
Establishing a cluster database that manages users whose similarities are equal to or greater than a set value to the same cluster
A method of recommendation using collective intelligence. 20. The method of claim 19,
In the recommendation method,
Searching for a user who has posted the rating information on the Internet if the service target person is not included in the cluster, and searches for a user whose similarity with the service target person is equal to or greater than the set value; And
A database updating step of updating the degree of similarity of the service target person with the searched user to the similarity degree database and adding the service target person to the cluster database
A recommendation method using collective intelligence. 13. The method of claim 12,
The service providing step may include:
Recommending the product to the service target person using the similarity between the service target person and the user included in the cluster to which the service target person belongs and the rating information posted by the user included in the corresponding cluster
A recommendation method using collective intelligence. 22. The method of claim 21,
The service providing step may include:
A final rating calculation step of calculating a final rating for the product by dividing the total of weights obtained by multiplying the rating information of the user by the degree of similarity with the user by the sum of the inter-user similarities for the cluster to which the service target person belongs, ; And
Recommending the product to recommend the product based on the final rating
A method of recommendation using collective intelligence.

Images