Apparatus and method for measuring contents similarity based on feedback information of ranked user and computer readable recording medium storing program thereof

• US 8,903,822 B2
• Filed: 04/12/2011
• Issued: 12/02/2014
• Est. Priority Date: 04/13/2010
• Status: Active Grant

• Current Assignee: Konkuk University Industrial Cooperation Corp
• Original Assignee: Konkuk University Industrial Cooperation Corp
• Inventors: Shin, Hyoseop
• Primary Examiner(s): Thai, Hanh
• Application Number:
  US13/084,730
• Publication Number:
  US 20110252044A1
• Time in Patent Office: 1,330 Days
• Field of Search: 707/736, 707/737, 707/748, 707/749
• US Class Current: 707/736
• CPC Class Codes: G06F 16/335 Filtering based on additional data, e.g. user or group profiles filtering in web context G06F16/9535, G06F16/9536

Abstract

An apparatus for measuring contents similarity based on feedback information from ranked users includes: a user ranking unit for receiving from a database information on users as contents creators and feedback information on contents and ranking the users based on the received information; and a user grouping unit for classifying the users into a plurality of groups based on the ranking result. The apparatus further includes: a sequence modeling unit for receiving from the database the contents and the feedback information and modeling the number of feedbacks as sequences based on the ranked user groups; and a post similarity calculation unit for measuring similarity between the modeled sequences.

Claims

1. An apparatus for measuring contents similarity based on feedback information from ranked users, comprising:

at least one processor to provide;

a user ranking unit configured to receive from a database information on users as contents creators and feedback information on posts as contents and rank the users based on the received information;

a user grouping unit configured to classify the users into a plurality of ranked user groups based on the ranking result;

a sequence modeling unit configured to obtain feedbacks from the plurality of ranked user groups and model the contents as sequences of data points based on the feedbacks, each of said data points comprising a preference for the contents of one of said posts as a function of a group ranking of a respective said user group; and

a post similarity calculation unit configured to measure similarity between the modeled sequences to obtain similarity of the posts.

2. The apparatus of claim 1, wherein the user ranking unit calculates the user ranking based on a random walk-based algorithm or a HITS (Hyperlink-Induced Topic Search)-based algorithm.

3. The apparatus of a ranked user of claim 1, wherein the user grouping unit classifies the users into the plurality of ranked user groups by using an exponential classification method.

4. The apparatus of claim 1, wherein the sequence modeling unit models preference of the ranked user groups on each post based on an absolute quantity of the feedbacks as in a following equation:

p=<

fb(p,G₁),fb(p,G₂₎, . . . ,fb(p,G_N)>

,

wherein fb(p, G_k) indicates the number of the feedbacks on a post p from user s in a user group G_k.

5. The apparatus of claim 1, wherein the sequence modeling unit models preference of the user groups on each post based on a relative quantity of the feedbacks as in a following equation:

6. The apparatus of claim 1, wherein the post similarity calculation unit calculates cosine similarity or an Euclidean distance while setting the sequences as vectors and performs sequence normalization using a maximum value and a minimum value as in a following equation:

7. The apparatus of claim 1, wherein the post similarity calculation unit calculates cosine similarity or an Euclidean distance while setting the sequences as vectors and performs sequence normalization using a preference threshold value as in a following equation:

8. The apparatus of claim 1, further comprising:

a similar post retrieval unit for retrieving a post similar to a random post by using similarity between the posts measured by the post similarity calculation unit;

a post clustering unit for clustering the posts based on the number of feedbacks on the posts and the similarity between the posts by using at least one of a majority selection clustering algorithm, a K-means clustering algorithm and an agglomerative hierarchical clustering (AHC) algorithm; and

a post classification unit for classifying the posts based on preference of the ranked user groups.

9. A method for measuring contents similarity based on feedback information of ranked users, comprising:

receiving, by a processor, information on users as contents creators and feedback information on posts as contents from a database of a contents similarity measuring apparatus and ranking the users based on the received information;

classifying, by a processor, the users into a plurality of ranked user groups based on the ranking result;

obtaining feedbacks from the plurality of ranked user groups and modeling, by a processor, the contents as sequences of data points based on the feedbacks, each of said data points comprising a preference for the contents of one of said posts as a function of a group ranking of a respective said user group; and

measuring, by a processor, similarity between the modeled sequences to obtain similarity between the posts.

10. The method of claim 9, wherein in said ranking, the user ranking is calculated based on a random walk-based algorithm or a HITS (Hyperlink-Induced Topic Search)-based algorithm.

11. The method of claim 9, wherein in said classifying, the users are classified into the ranked user groups by using an exponential classification method.

12. The method of claim 9, wherein in said modeling, preference of the user groups on each post is modeled based on an absolute quantity of the feedbacks as in a following equation:

p=<

fb(p,G₁),fb(p,G₂), . . . ,fb(p,G_N)>

,

wherein fb(p, G_k) indicates the number of the feedbacks on a post p from user in a user group G_k.

13. The method of claim 9, wherein in said modeling, preference of the user groups on each post is modeled based on a relative quantity of the feedbacks as in a following equation:

14. The method of claim 9, wherein in said measuring similarity, an Euclidean distance is calculated while setting the sequences as vectors, and sequence normalization using a maximum value and a minimum value is performed as in a following equation:

15. The method of claim 9, wherein in said measuring similarity, an Euclidean distance is calculated while setting the sequences as vectors, and sequence normalization using a preference threshold value is performed as in a following equation:

16. The method of claim 9, further comprising retrieving a post similar to a random post by using the similarity between the posts measured by a post similarity calculation unit.

17. The method of claim 9, further comprising clustering the posts based on the number of feedbacks and the similarity between the posts measured by a post similarity calculation unit by using at least one of a majority selection clustering algorithm, a K-means clustering algorithm, and an agglomerative hierarchical clustering (AHC) algorithm.

18. The method of claim 9, further comprising classifying the posts based on preference of the ranked user groups.

19. A computing device storing a program for performing a method for measuring contents similarity via a wire/wireless communication network, the program including stored instructions executable by the computing device for:

receiving information on users as contents creators and feedback information on posts as contents from a database of a contents similarity measuring apparatus and ranking the users based on the received information;

classifying the users into a plurality of ranked user groups based on the ranking result;

obtaining feedbacks from the plurality of ranked user groups and modeling the contents as sequences of data points based on the feedbacks, each of said data points comprising a preference for the contents of one of said posts as a function of a group ranking of a respective said user group; and

measuring similarity between the modeled sequences to obtain similarity between the posts.