Upload
geun-sik
View
213
Download
1
Embed Size (px)
Citation preview
Building a Semantic Social Network based on Interpersonal Relationships
Kee-Sung Lee, Myung-duk Hong, Jin-guk Jung Dept. of Information Engineering
Inha University Incheon, Korea
{lks, hmdgo, gj4024}@eslab.inha.ac.kr
Geun-sik Jo School of Computer & Information Engineering
Inha University Incheon, Korea [email protected]
Abstract—With the emergence of the Smartphone, people can use Online Social Network Services ubiquitously, leading to a significant increase of the number of participants in online social networks. Under these circumstances, online users will require an intelligent and intuitive social relationship management system such as the ontology-driven browsing method. In this paper, to build a user-centered semantic social network and to represent entities and relationships with ontology to improve retrieval performance of the semantic social network, we will design our ontology extended from FOAF, RELATIONSHIP and propose a new method to compute closeness among friends using resources on social networks. Furthermore, we evaluate our ontology-driven browsing on via implementing a prototype system.
Keywords-component; Interpersonal relationship rules, closeness, semantic social network, ontology population
I. INTRODUCTION The recent trend in online social networks is the
adoption of semantic web technologies, which we call the Semantic Social Network. Several ontologies (FOAF, SIOC, SKOS) are used to represent social networks. For example, FOAF is used for describing people profiles, their relationships and their activities online. The “knows” property, which is one of well-known properties in the FOAF to represent relationship between friends in a social network, tends to be subdivided to more precisely describe the relationship (familial, friendship, or professional relation-ships). RDF based descriptions of social data provide rich graphics and offer a much more powerful and significant way to represent online social networks than traditional models of social network analysis.
However, many of the latter studies focused on social networks with binary rationalities like “friend or not”. For example, [3] has proposed a Semantic Web based framework for social networks. In the latter study, simple relationships such as friend and colleague were defined as being limited in comparison with much diverse relationships offline. Another problem is that most relationships were manually specified by individual users [4, 5]. Therefore, many users should spend much time to manage their relationships and to search items they intend to find.
In this paper, we propose a new approach to build a user-centered semantic social network and to represent entities and relationships with ontology to improve retrieval performance of the semantic social network. To do this, our ontology will first be designed to be able to reflect offline various relationships between people, one between a person and a resource, and one between resources. A semantic social network then will be built from social networks like Facebook using the ontology, while Kinship and Friendship will be computed by the proposed method. The proposed method accomplishes the following: � Provide rich of properties to represent resources and
relationships for semantic social network services. � Define some rules to infer relationships from
information gathered on social network services. � Formulate an equation to computer closeness among
users with resources on social networks. � Provide a way to retrieve information with ontology.
II. RELATED WORKS In recent studies on social networks, varieties of offline
relationship are commonly being used. In visoLink [1], which is a social network based on "friend ranking" by using the connection associated with a network of currently active blogs of users. It is measured by analyzing their online reading and writing content on each blog. In [2], the authors proposed a method for discovering interesting associations among users in personal photos using a new network structure model, Face Co-Occurrence Networks. In [6], the authors proposed a framework that collects the mobile life-logs, and additionally, high-level logs such as activity and emotion to help users annotate them. Based on these life-logs, they build a novel mobile social network by mining the semantic relations between users in life-logs by using Bayesian network. In this study, personal relation-ships (close friend, friend, acquaintance, none) and business relationships (close colleague, colleague, acquaintance, none) were defined as relationships between people. With this, the current social networks of various relationships are found out and actively used for the course of studies. However, in order to infer a variety of off-line relationships
2012 Third FTRA International Conference on Mobile, Ubiquitous, and Intelligent Computing
978-0-7695-4727-5/12 $26.00 © 2012 IEEE
DOI 10.1109/MUSIC.2012.23
90
2012 Third FTRA International Conference on Mobile, Ubiquitous, and Intelligent Computing
978-0-7695-4727-5/12 $26.00 © 2012 IEEE
DOI 10.1109/MUSIC.2012.23
90
2012 Third FTRA International Conference on Mobile, Ubiquitous, and Intelligent Computing
978-0-7695-4727-5/12 $26.00 © 2012 IEEE
DOI 10.1109/MUSIC.2012.23
90
more information is concerned; profiles of university, company, birthday, post, comment, photo etc., are needed.
Therefore, in this paper, we utilize a variety of information and relationships between people in social networks through a variety of inference rules and intimate expressions to build a variety of relationships.
III. SEMANTIC SOCIAL NETWORK BASED ON INTERPERSONAL RELATIONSHIPS
In this section, an online semantic social network based on diverse relationships rather than binary relational ties (e.g., friend or not) will be described with a close investigation on offline social networks, along with the design of an ontology model [6]. The ontology model will include definitions of relationships between person and person, between resource and resource, and between person and resource.
Figure 1. Concept of semantic social network based on
interpersonal relationship
A. Semantic Social Network Model Fig. 1 illustrates the proposing concept of SSNIR
(Semantic Social Network based on Interpersonal Relationships). Most current social network services do not provide ontology-driven information. It means that the social network model implicitly represents only relationships such as “knows” or “friendOf” as undirected edges between a person and a person as nodes. Generally, an online social network is computationally represented by a node-edge undirected graph, and the online social network service provides users a linear friend list and resource list to manage their own relationship and resources without any semantic information.
Based on our observation on online social networks, there are three major actions for users to play on their social networks: creating a self-descriptive profile, building a personal social network, and writing an article. Generally,
the social network can be represented with two components, several vertices and edges corresponding to friends and relationships, respectively. This simple network can have a weight on an edge to depict the strength of closeness between two persons. Some researchers had proposed various methods to quantify closeness [1, 2, 6]. To do this, they had considered other resources such as photos, tags, and messages on social network services like Facebook and MySpace. The proposed Semantic Social Network is a network (directed graph) with two components described in ontology: vertices and edges. The vertex can be divided into user node and resource node in details. Based on the relationship between users and resources, we classify relationships into the following three groups.
Definition 1 (Relationship Category): There are three kinds of relationships that are determined based on both sides of objects: friendship with both users, resourceship with both contents, and ownership with a user and a resource. Let U be a set of users who are members of a social network service, and let C be a set of re-sources/contents written or uploaded by the users to the service. Let R be a set of relationships, and r be an element of R. Therefore, the network can be represented in a 3-tuple as the below definition below.
R = {r | r � U × U} (Friendship) R = {r | r � C × C} (Resourceship) R = {r | r � U× C} (Ownership)
Definition 2 (SSNIR Model): Let U be a set of users who are members of a social network service, and let C be a set of resources/contents written or uploaded by the users to the service. Finally let R be a set of relationships defined above. The network Nssnir is as follows.
Nssnir = {U, R, C}
Furthermore the friendship will be extended to reflect interpersonal relation in details on offline social networks into the SSN.
B. Ontology for SSNIR A new ontology to efficiently handle the closeness of
friendships with resources will be designed in this section. Based on our view on social networks mentioned in the previous section, some classes and relationships are defined (Figure 2). The ontology is used to provide concept-driven navigation on the semantic information.
Our ontology inherits from FOAF, RELATION-SHIP [7] and OntoAlbum [5], and it extends the meaning of each class and properties. For example, the “ssn:User” class to represent a user who is a member of an online social network service is a subclass of the “foaf:Person”, of which class is defined in FOAF, and it has more relationships and properties those of the person class. Most new relationships and properties are designed to describe the ownership
919191
between the class and a resource that is written or uploaded by the user. To represent the friendship in details, the “ssn:Relation-ship” is extended from the “rel:knows”, of which the property is defined in RELATIONSHIP. Newly defined relationships include “juniorOf”, “seniorOf”, “visitWith”, and “closeTo” properties.
ssn:User
foaf:Person
ssn:Photo
foaf:Group
ssn:hasMember
foaf:Organization
ssn:Education
ssn:hasPerson
ssn:Colleage/University
ssn:HighSchool
ssn:NewsFeed/Comment
rdfs:literal
ssn:hasPhoto
ssn:Company
ssn:periodOf
rdfs:literal
rdfs:literal
ssn:classOf
ssn:classOf
foaf:memberOf
rel:knows
ssn:Relationship
ssn:betweenFriend
ssn:affiliatedTo
ssn:writeTo
property
subClassOf
Class
ssn:writenBy
ssn:hasEmployee
ssn:hasRelationshiprdfs:float
ssn:hasCloseness
ssn:Location
ssn:hasLocation
ssn:visitWith
ssn:appearedIn
ssn:visitWhere
Figure 2. Ontology for semantic social network based on interpersonal
relationships
In order to compute the strength of closeness between a user and his friend, newsfeed post, comment, and photos are considered. We assume that if people appear together in a photo, then they know each other and have a close friendship. Like the assumption, newsfeeds and reciprocated articles to a user are also considered as important resources to determine closeness. The way of determining newsfeeds, tags, and photos will be explained next. Representation of the resources and associated relationships in our ontology is also designed (Figure 2).
C. Closeness of the friendship Methods proposed to predict similarity between bloggers
in Blogspace are not suitable for computing the closeness of friendship based on only newsfeeds on SSN. In order to compute the closeness between users who have a friendship, we take account of activeness, loyaltyness, seeness, and mutualness. Activeness represents the rate of actively writing newsfeeds and tagging the newsfeeds on an online social network. It simply is able to be computed on users, ui and uj. Let cij be the number of comments that user ui writes on the posts of user uj , and pij be the number of posts that user ui writes on those of user uj. Note that we count only one as a number of comments even though ui writes multiple comments on a post. For the users, uj and ui, the activeness from ui to uj is denoted by activeness (ui, uj), and computed by the following equation
k ik
ij
k ik
ijji c
cp
puuactiveness ×=),( (1)
where k is the number of friends. This equation is an asymmetric function because our proposed network is a directed graph.
The loyaltyness depicts the faithfulness of user ui to user uj. It means how frequently the user replies to posts of user uj. It is computed by a rate of comments and posts of each user. Note that it is also asymmetric. The following equation shows how loyaltyness is calculated.
ji
ijji p
cuusloyaltynes =),( (2)
Let seeness(ui, uj) be the seeness between user ui and another user uj. T is the total number of photos that belong to the user and his/her friends. Let us assume that n photos among T have the two users together and the symbol h is the number of persons who are in the photo. It implies that if the two users open appear together on photos, then the users may be close on offline social networks. In contrast to the previous two equations, the following equation is a symmetric function.
2( , )i j n
nseeness u uT h
= � (3)
The mutualness, which measures dissimilarity between ui and another user uj, is complementary to the Jaccard coefficient [8] and is obtained by subtracting the Jaccard coefficient from 1, or, equivalently, by dividing the difference of the sizes of the union and the intersection of two sets by the size of the union:
ji
jijiji
uu
uuuu)u,u(mutualness = (4)
Finally the closeness from user ui to user uj is computed in the following equation. It is an average value of activeness multiplied by loyaltyness, seeness and mutualness, which is a normalized value (0 to 1) as divided by the biggest value. The multiplication of activeness and loyaltyness represents how much close the users are based on their posts and comments on online social networks. Note that the range of this value is from 0 to 1 and asymmetric value. =closeness loyaltynesactiveness ×
(5) esslnmutuaseeness++
929292
IV. ONTOLOGY POPULATION In order to build Semantic Social Network based on
online social network services such as Facebook that can provide information of user profiles, posts, photos, and tags, we need to transform the information to semantic information that will be represented in OWL. While performing this process, some relationships can be directly generated from the information, but the other relationships can be created through inference based on the information. In this section, some rules written in Semantic Web Rule Language (SWRL) for inference will be explained to generate instances of our ontology model from the infor-mation that can be taken on social network services. To extend kinship, we assume that some relationships can be directly acquired from SNS such as Facebook.
A. Rules for Kinship For instance, the following relationships are specified by
users on Facebook; “Daughter”, “Son”, “Mother”, “Father”, “Sister”, “Brother”, “Aunt”, “Uncle”, “Niece”, “Wife”, “Husband”, “Nephew”, “Granddaughter”, “Grand-son”, “Grandmother”, and “Grandfather”. In this section, a few rules will be defined to infer an unknown relationship from acquired information. Defined kinship rules for this study are to generate instances of the “childOf”, “parentOf”, “siblingOf” and “spouseOf”. These properties involve inference, which is the process of deriving logical conclusions from premises known to be true. One of the rules is given below. Rule 1 (siblingOf): The meaning of “siblingOf” relation-ship is that a property representing a person having one or both parents in common with this person. We define the rule for this relationship below.
hasFather(?x1,?x2)^hasFather(?x3,?x4)^ swrlb:stringEqualIgnoreCase(?x2, ?x4) � siblingOf(?x1,?x3)
In this rule, the property “hasFather” can be replaced with “hasMother”. Furthermore, the conclusion “sibling-Of(?x1, ?x3)” has a symmetric property so that the “sibling-Of(?x3, ?x1) is also true because the symmetric one means that if a property is symmetric and the pair (x, y) is an instance of the symmetric property P, then the pair (y, x) is also an instance of the property P.
B. Rules for Friendship In order to generate instances of “acquaintanceOf”,
“closeTo”, “colleagueOf”, “friendOf”, “juniorOf”, “Knows-InPassing”, “knowsOf”, “menteeOf”, “mentorOf”, “neigh-borOf”, “seniorOf”, “visitWith” and “worksWith”, friend-ship rules are defined in this section. A part of these rules is only explained to save space. The “mentorOf” and “seniorOf” properties are selected to explain rules as followings.
Rule 2 (mentorOf): The meaning of “mentorOf” relation-ship is that a property representing a person who serves as a trusted counselor or teacher to this person. Define the rule for this relationship as below.
affiliatedTo(?x1,?y1)^ hasPositionion(?y1,”professor”)^ periodOf(?y1,?z1)^affiliatedTo(?x2,?y2)^ swrlb:stringEqualIgnoreCase(?y1, ?y2)^ swrlb:lessThanOrEqual(?z1, ?z2) � mentorOf(?x1,?x2)
The literal “professor” in the “hasPosition” can be
replaced with “teacher”, and “mentorOf” is defined as the inverseOf property with the “menteeOf”. It means that if property P1 is stated to be the inverse of property P2, and if X is related to Y by P2 property, then Y is related to X by the P1 property. Therefore, “mentorOf(?x1, ?x2)” implies that “menteeOf(?x2, ?x1)” is true. Rule 3 (seniorOf): The meaning of “seniorOf” relationship is that entrance upon office or school was anterior to that of
another person. Define the rule for this relationship as below.
affiliatedTo(?x1,?y1)^affiliatedTo(?x2,?y2)^ swrlb:stringEqualIgnoreCase(?y1, ?y2)^ classOf(?y1,?z1)^classOf(?y2,?z2)^ swrlb:lessThanOrEqual(?z1,?z2)� seniorOf(?x1,?x2)
The property “classOf” can be replaced with “periodOf”,
and “seniorOf” is also defined as the inverseOf property of “juniorOf”.
C. Rules for Closeness Based on the score of closeness between user ui and
another user uj that is mentioned in section 3.3, the following rules are defined to generate each instance of 4 properties, “closeTo”, “knowsOf”, “acquaintanceOf”, and “wouldLikeToKnow”. Note that these properties do not satisfy the symmetric and transitive property. A part of these rules is only explained to save space. The “closeTo” and “acquaintanceOf” properties are selected to explain rules as followings.
Rule 4 (closeTo): The meaning of close relationship is a property representing a person who shares a close mutual friendship with this person. Define the rule for this relationship as below.
hasRelationship(?x1,?y1)^betweenFriend(?y1, ?x2)^ hasClosenessValue(?y1 , ?w1)^ swrlb:greaterThanOrEqual(?w1, 0.8) � closeTo(?x1,?x2)
Rule 5 (acquaintanceOf): The meaning of “acquaintance-Of” relationship is a property representing a person having more than slight or superficial knowledge of this person but
939393
short of friendship. Define the rule for this relationship as below.
hasRelationship(?x1,?z1)^betweenFriend(?z1, ?x2)^ hasClosenessValue(?x2 , ?w1)^ swrlb:greaterThanOrEqual(?w1, 0.3)^ swrlb:lessThan(?w1, 0.5) � acquaintanceOf (?x1,?x2)
V. IMPLEMENTATION AND EXPERIMENTS
A. System Prototype Implementation The system proposed in this paper was implemented as
shown in Fig. 3 using Facebook Graph API for collecting user profile on Facebook and JavaScript for user interface. SSNIR automatically was built with acceptance to use our system after the user has logged onto Facebook. It can provide a way to access information based on the designed ontology that supports an intuitive approach to browse the result of user’s queries. For example, Fig. 3 is a screen shot of a user browsing the result of the following question: “Find a web page of a person who is one among seniors of the same university and has visited JEJU Island with the user”.
Figure 3. A screenshot of ontology-driven browsing interface.
The prototype system first displays the user as an
instance of Person class and properties that belong to the class. The instances of Person class and relationships are depicted as a rectangle, while instances of Resource class are depicted as a circle. The user then clicks the “Friendship” node to navigate the “seniorOf” relationship and the “visitWhere” property. At each node, sub-nodes will be extended by the user’s clicks and the path of the user’s clicks will be highlighted to emphasize the history of a click stream. Finally, the user reaches a link of the friend’s web page.
B. Experiments We have proposed a new method that could compute the
closeness with resources on online social network services in the previous section. To show the justification of the method, the following experiment was conducted for 32 users. The average number friend is 120, colleague
information of 32 users are 1765 people, as is shown in Table 1.
Table 1. Experiments data set.
Age #user #Avg. friend
#user colleague
#user company
#user birthday
20-29 18 183 3018 1266 1509 30-39 14 68 301 384 536 Total 32 120 1765 824 1021
First, each user has written down one of four closeness
(“closeTo”, “knowsOf”, “acquaintanceOf”, “wouldLike- ToKnow”) for their friends. After that, our system has compared each value with one computed by “equation (5)” in Section 3.3
We consider what data affects the accuracy of each relationship concerning closeness. Fig. 4 shows such accuracy for them with monthly data sets. To evaluate our consideration, 5 data sets are prepared: 3m (30 Oct., 2011 ~ 3 months ago), 6m (30 Oct., 2011 ~ 6 months ago), 9m (30 Oct., 2011 ~ 9 months ago), and so on.
Figure 4. Result of accuracy for each relationship concerning the
closeness and the Inverse function of the Standard Deviation (ISD) with monthly data sets.
For one year data, the predicted closeTo reaches about
85%. All users communicated with their friends online more frequently than offline by instant messages as posts and comments. However, online closeness did not reflect the offline one outright because some friends of users were not active users of online social network services. In spite of that, the closeTo computed with the latest data is more approximate than that computed with the whole data.
For the closeness except closeTo, the result of data set for 6 months is improved in terms of accuracy compared to that with a data set for 3 months. Along with the result, the accuracy of the data set for 9 months is also improved. However, even if we use a bigger data set for 9 months, accuracy is not improved significantly. Even though one can achieve the best accuracy with the whole data set, one has to process the biggest data set for the longest time.
To directly obtain distribution, we use the Inverse function of the Standard Deviation (ISD) at each data set. After that, we normalize these values. Fig. 4 shows those values. As you can see in Fig. 4, we can achieve enough
949494
accuracy with a data set of 9 months. Another interesting result is about closeTo. The best accuracy of this property that is different from the other properties can be driven with the recent data set. In particular, the normalized ISD almost converges on 95% as you can see in Fig. 4.
The next experiment shows that each component affects the result of closeTo with the line marked by the diamond with a nine-months data set (9m) in Fig. 5. We can predict the degree of friendship based on variable combinations of activeness, loyaltyness, seenness, and mutualness. Each component affects the result differently. For example, activeness is less important than loyaltyness and seenness while closeTo is predicted. As you can see in Fig. 6, loyaltyness and seenness are less important when wouldLikeToKnow is predicted.
Figure 5. Result of each component affects and the Inverse function of
the Standard Deviation (ISD) with 9 months data set.
Mutualness is more suitable for wouldLikeToKnow than closeTo. In other words, if one considers only loyaltyness or seenness when predicting the degree of closeness, then one may get relatively better closeTo, but less wouldLikeTo-Know. One may also have a result with bigger ISD than other components. However, the proposed method produces the best result with the biggest ISD, even though closeTo with the line marked by the diamond is not the biggest accuracy.
C. Discussion The first limitation is about generating a semantic social
network as via applying rules to information gathered from social network services. The precision and recall achieved are 100% and 83% respectively about instances generated from current online social networks via our defined rules due to various reasons; for example, one of reasons is that current online social networks represent different information in spite of having the same information. For example, in Facebook Inha University and
that is Korean for “Inha University” are described with different identifiers, 112975158716498 and 100000582566699 respectively. To handle this problem, the “SameAs” property can be used to represent them in ontology.
The second limitation is that some users do not provide information that is needed to build a semantic social network. Some users do not allow a part of their information to be available to the public, so some instances of concepts and relationships may not be generated from online social networks.
VI. CONCLUSION AND FUTURE WORK In this paper, we proposed a new ontology-driven
browsing method and a novel way to compute closeness between friends using resources that are written and uploaded to social network services by users. To demonstrate our ontology-driven browsing, our ontology was extended from FOAF and the previous works, and we show intuitive browsing as the result of a query via the implemented prototype system. In addition to browsing, our proposed method concerning closeness can produce 85% precision. It means that our system can predict the approximate closeness between users and their friends using resources on social network services.
This information automatically generated by the system can be used for U2Mind [4] and OntoAlbum [5], which systems are multimedia retrieval system using ontology to achieve ontology-driven browsing and retrieval.
ACKNOWLEDGEMENT This work was supported by the National Research
Foundation of Korea (NRF) grant funded by the Korea government (MEST) (No. 2011-0015484).
REFERENCES
[1] L. Fan, and B. Li, “Blog-based Online Social Relationship Extraction”, In the Proceedings of 8th IEEE International Conference on Congnitive Informatics,Hong Kong, China, pp. 457-463, 2009
[2] H. N. Kim, J. G. Jung, and A. El. Saddik, “Associative face co-occurrence networks for recommending friends in social networks,” In the Proceedings of second ACM SIGMM workshop on Social media, New York, USA, pp. 27– 32, 2010.
[3] B. Carminati, E. Ferrari, R. Heatherly, M. Kantarcioglu, and B. Thuraisingham, “A semantic web based framework for social network access control”, In the Proceedings of 14thSACMAT,Stresa, Italy, pp.177-186, 2009.
[4] K. S. Lee, J. G. Jung, K. J. Oh, and G. S. Jo, “U2Mind: Visual semantic relationships query for retrieving photos in social network,” In the Proceedings of 3rd Asian Conference on Intelligent Information and Database System.Daegu, Korea, 347-356, 2011.
[5] Y. Chai, X.Y. Zhu, and J. Jia, “OntoAlbum: An ontology based digital photo management system”, In the Proceedings of ICIAR, Povoa de Varzim, Portugal, pp. 263–270, 2008.
[6] H. S. Park, and S. B. Cho, “Building Mobile Social Network with Semantic Relation using Bayesian Network-based Life-log Mining”, In proceeding of IEEE Second International Conference on Social computing, Minneapolis, USA, pp.401-406, 2010.
[7] Davis, I. andJr, E.V., 2010 [online]. “RELATIONSHIP: A vocabulary for describing relationships between people”, < http://vocab. org/ relationship/>
[8] A. H. Lipkus, “A proof of the triangle inequality for the Tanimoto distance”, Journal of Mathematical Chemistry”, vol 26, pp.263-265, 1999.
959595