DISTRIBUTED MULTIMEDIA ACCESS SYSTEM AND METHOD

• US 20090119265A1
• Filed: 11/05/2008
• Published: 05/07/2009
• Est. Priority Date: 11/05/2007
• Status: Active Grant

• Current Assignee: National Taiwan University (Government of The Republic of China)
• Original Assignee: National Taiwan University (Government of The Republic of China)
• Inventors: Lin, Ching-Ju, Chang, Yi-Ting, Chou, Cheng-Fu, Hsu, Chih-Cheng
• Granted Patent: US 8,688,639 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/1837 Management specially adapted to peer-to-peer storage networks topology management mechanisms of peer-to-peer networks H04L67/1042
• CPC Class Codes: G06F 16/48 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually

Abstract

A distributed media access system and method. The method includes making each of a plurality nodes in a P2P network designate at least one nearest neighbors as a first node candidate and perform at least one random walk to obtain a second node candidate according to the peer list; calculating user preferences of the request node and user preferences of the first and second node candidates according to the first and second node candidates and the preference parameters corresponding to the first and second node candidates so as to obtain a plurality of similarity values; selecting the more similar candidate as buddies of user of the node according to the calculated similarity values; and establishing a buddy list according to the buddies of user and storing the buddy list such that the P2P network can quickly find buddies of the node through the buddy list.

Claims

1. A distributed multimedia access system applied in a P2P network having a plurality of nodes, wherein users located at each of the nodes are required to input preference parameters to the P2P network to obtain a peer list located at some of the nodes of the P2P network, the system comprising:

a buddy candidate selection module allowing any one of the nodes to designate at least one node nearest to a request node as a first node candidate for the request node and perform at least one random walk to obtain a second node candidate according to the peer list;

a similarity calculation module for calculating similarity between user preferences of the request node and user preferences of the first and second node candidates according to the first and second node candidates obtained by the buddy candidate selection module and the preference parameters corresponding to the first and second node candidates so as to obtain a plurality of similarity values;

a buddy selection module for selecting at least a more similar node candidate between the first and second candidates as buddies of users located at the request node according to the calculated similarity values calculated by the similarity calculation module; and

a buddy list storage module for establishing and storing a buddy list according to the buddies of users such that the P2P network can quickly find buddies of the request node through the buddy list.

2. The distributed multimedia access system of claim 1, wherein the preference parameters of each of the nodes are determined through multimedia file contents input by users located at the node.

3. The distributed multimedia access system of claim 1, wherein each of the node candidates stores objects related to preference of the users located at the node candidates, characteristics of the node candidate can be determined through attribute description of the objects, the objects have at least one attribute, Ka represents a set of keywords describing an attribute a if an attribute set is A and a is one attribute of A, the objects can be categorized according to tagged keywords, and preferences for different categories of objects can be used to distinguish different nodes.

4. The distributed multimedia access system of claim 3, wherein when the preference parameters are determined through attributes of the objects input by users located at the request node, the similarity calculation module let object(c ) represent the objects input by a user c_i located at the request node and be defined as a vector of weights {right arrow over (w_i)}=(w_ik_i, w_ik₂, . . . , w_ik_n, wherein the weight w_ik_n represents user c_i'"'"'s preference for the objects and is described by keyword k_n, the similarity calculation module performs the following calculation:

$w_i<br><br>k_n=\frac{<br><br>O_{i,k_n}<br><br>}{<br><br>O_i<br><br>}$

wherein O_i is the set of the objects held by the user c_i, O_i,kn is a subset of O_i containing the objects tagged by the keyword k_n, Cosine similarity measure is used to quantify similarity sim(c_i,c_j) between the user c_i and one of the M candidates c_j, the similarity calculation module performs the following calculation;

$\mathrm{sim}<br><br>\left(c_i,c_j\right)=\cos <br><br>\left(\stackrel{\to <br><br>}{w_i},\stackrel{\to <br><br>}{w_j}\right)=\frac{\stackrel{\to <br><br>}{w_i}·<br><br>\stackrel{\to <br><br>}{w_j}}{{<br><br>\stackrel{\to <br><br>}{w_i}<br><br>}_2\times <br><br>{<br><br>\stackrel{\to <br><br>}{w_j}<br><br>}_2}=\frac{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>w_{i,k}<br><br>w_{j,k}}{\sqrt{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>w_{i,k}^2}<br><br>\sqrt{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>w_{j,k}^2}}$

wherein K is the total number of keywords. If c_i and c_j have similar tastes in certain styles of multimedia content, then sim(c_i,c_j) returns a smaller value.

5. The distributed multimedia access system of claim 1 further comprising a query unit for searching multimedia files interested by user of the request node, the query unit comprising:

a receiving module for receiving multimedia file query condition sent by the request node;

a search module for sending the query condition to the plurality of node candidates so as to search multimedia files that meet the query condition, wherein the plurality of node candidates further sends the query condition to other node candidates so as to search multimedia files that meet the query condition until a predefined time is reached;

a capturing module for capturing the multimedia files found by the search module; and

a transferring module for transferring the multimedia files captured by the capturing module to the node.

6. The distributed multimedia access system of claim 5, wherein if a plurality of multimedia files is captured by the capturing module, a multimedia file list is established by the capturing module and sent by the transferring module to the node.

7. The distributed multimedia access system of claim 5, wherein the query condition is a keyword or a unique identification code

8. The distributed multimedia access system of claim 7, wherein the keyword is an attribute or an attribute set describing the multimedia files

9. A distributed multimedia access method applied in a P2P network having a plurality of nodes, wherein user of each of the nodes is required to input preference parameters to the P2P network to obtain a peer list located at some of the nodes of the P2P network, the method comprising the following steps of:

(1) making any one node designate at least one nearest neighbor as a first node candidate and perform at least one random walk to obtain a second node candidate according to the peer list;

(2) calculating user preferences of the request node and user preferences of the first and second node candidates according to the first and second node candidates and the preference parameters corresponding to the first and second node candidates so as to obtain a plurality of similarity values;

(3) selecting at least a more similar candidate between the first and/or the second node candidates as buddies of user of the request node according to the calculated similarity values; and

(4) establishing a buddy list according to selected buddies of user and storing the buddy list such that the P2P network can quickly find buddies of the node through the buddy list.

10. The distributed multimedia access method of claim 9, wherein the preference parameter of each node is determined through multimedia file content input by user of the node.

11. The distributed multimedia access method of claim 9, wherein each candidate stores objects related to preference thereof, characteristic of the candidate can be determined through attribute description of the objects, the objects have at least one attribute, if an attribute set is A and a is one attribute of A, Ka represents a set of keywords describing the attribute a, the objects can be categorized according to tagged keywords, preferences for different categories of objects can be used to distinguish different nodes.

12. The distributed multimedia access method of claim 11, wherein when the preference parameter is determined through attributes of the objects input by user of the node, object(c_i) represents the objects input by user c_i of the node, which is defined as a vector of weights {right arrow over (w_i)}=(w_ik_i, w_ik₂, . . . , w_ik_n, . . . ), wherein weight w_ik_n represents user c_i'"'"'s preference for the objects described by keyword k_n, w_ik_n is calculated through the following equation:

$w_i<br><br>k_n=\frac{<br><br>O_{i,k_n}<br><br>}{<br><br>O_i<br><br>}$

wherein O_i is the set of the objects held by the user c_i, O_i,kn is a subset of O_i containing the objects tagged by the keyword k_n. Cosine similarity measure is used to quantify similarity sim(c_i,c_j) between the user c_i and one of candidates c_j, sim(c_i,c_j) is calculated through the following equation;

$\mathrm{sim}<br><br>\left(c_i,c_j\right)=\cos <br><br>\left(\stackrel{\to <br><br>}{w_i},\stackrel{\to <br><br>}{w_j}\right)=\frac{\stackrel{\to <br><br>}{w_i}·<br><br>\stackrel{\to <br><br>}{w_j}}{{<br><br>\stackrel{\to <br><br>}{w_i}<br><br>}_2\times <br><br>{<br><br>\stackrel{\to <br><br>}{w_j}<br><br>}_2}=\frac{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>w_{i,k}<br><br>w_{j,k}}{\sqrt{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>w_{i,k}^2}<br><br>\sqrt{\underset{k=1}{\overset{K}{\sum <br><br>}}<br><br>w_{j,k}^2}}$

wherein K is the total number of keywords. If c_i and c_j have similar tastes in certain styles of multimedia content, then sim(c_i,c_j) returns a smaller value.

13. The distributed multimedia access method of claim 9 further comprising:

receiving multimedia file query condition sent by the node;

sending the query condition to the plurality of candidates so as to search multimedia files that meet the query condition, wherein the plurality of candidates further sends the query condition to other node candidates so as to search multimedia files that meet the query condition until the predefined time (or called time to live, TTL) is reached; and

capturing the multimedia files that meet the query condition and transferring the multimedia files to the node.

14. The distributed multimedia access method of claim 13, wherein if a plurality of multimedia files is captured, a multimedia file list is established and sent to the node.

15. The distributed multimedia access method of claim 13, wherein the query condition is a keyword or a unique identification code

16. The distributed multimedia access method of claim 15, wherein the keyword is an attribute or an attribute set describing the multimedia files.

17. The distributed multimedia access method of claim 13, wherein the TTL is determined by an estimation process.

18. The distributed multimedia access method of claim 17, wherein the estimation process comprises the following steps of:

(1) estimating popularity of objects to be searched;

(2) multiplying the number of all objects with the popularity to obtain the number of the objects;

(3) dividing the number of the all objects by the number of the objects to obtain an average visit number;

(4) estimating similarity cluster ratio of neighboring nodes; and

(5) integrating the number of nodes of the clusters, the average visit number and the similarity cluster ratio to estimate the TTL.

19. The distributed multimedia access method of claim 9, wherein nodes in different clusters have a certain probability to be designated as the first node candidate.