Sie sind auf Seite 1von 5

Video Recommendation System Based on Personalized

Ontology and Social Network


Rung-Ching Chen Hua-Yuan Shih
Chaoyang University of Technology Chaoyang University of Technology
Taichung, Taiwan Taichung, Taiwan
crching@cyut.edut.tw iversonshih@gmail.com

Yu-Siang Lin Hendry


Chaoyang University of Technology Chaoyang University of Technology
Taichung, Taiwan Taichung, Taiwan
air8426@gmail.com hendry.honk@gmail.com

ABSTRACT Besides, users also affect behavior and thinking each other.
Users often use browser, media or mobile device to watch video. Therefore, by way of user's visiting website and her/his watching
In these platforms, most of recommendation through hit icon to record of videos can be references for personal recommendation
calculate grade. In this study, we design on ontology which including the neglect knowledge for users.
combined with social networks or video recommendation. Users Currently, the movies recommendation is by way of users scores
and their friends data will be added to the ontology. According to or clicks on website and according to videos category. When the
user and friends preferences, the system calculates the similarity platform is changed to mobiles environment, the recommendation
of interests. The system uses a rule generating algorithm to create system do the same works. The personal recommendation can
dynamic inference rules. Video are recommended to users by a follow users experiences on YouTube to do recommendation. If
JENA Inference Engine. Over time, the content of personal users never watch videos, it cant process personal
ontology will be updated. The ontology content at mobile device recommendation. Actually, users hobbies formation is not only
will be kept in the newest update states of owl to let recommender from nature but also greater parts are from interaction and users
increase the accurate results. friends of familiarity [1].
CCS Concepts In recent years, recommendation system is gradually escaping the
Information systems~Personalization traditional single recommendation. Many studies begin to
investigate the recommendation system which combines semantic
Keywords web reasoning mechanism. The related technologies such as
Personalized; mobile device; ontology; social network; video ontology are used to achieve intelligent recommendation. Through
recommendation. combine mobile devices, so digital content can be transmitted
more easily for the users. In the ontology inference developing
1. INTRODUCTION process, some recent studies have focused on the establishment of
Mobile devices and mobile networks are popular currently. personal ontology. By the way of the user's properties and users
Mobile services are regarded to the social network applications use the past records, the belonging to the user's ontology can be
such as Line, Whatsapp, Facebook. The social networks' services produced. The ontology can be adjusted at any time by using the
can bring a lot of benefits for business. The mobile services are record contents to provide a better service for users.
care to response speed. In order to speed up the search time, we
In this paper, we propose a novel method which combined social
will combine personalized recommendation method with mobile
network, real-time communication software and ontology to use
services to provide a more effective application to save time for
the data source of personalized recommendation. The data sources
searching information.
are come from users using record and personalized
recommendation can reference user's friend data on social
network and real-time communication software to update
Permission to make digital or hard copies of all or part of this work for personalized ontology. Finally, the functions of inference engine
personal or classroom use is granted without fee provided that copies are infer user's inferences to the personal results. The system uses the
not made or distributed for profit or commercial advantage and that user's friends' social network to solve cold start problems.so, the
copies bear this notice and the full citation on the first page. Copyrights method can be used to infer the users never notice but may
for components of this work owned by others than the author(s) must be actually be interested in film seven users are log in at the first time.
honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior 2. RELATED WORK
specific permission and/or a fee. Request permissions from
Permissions@acm.org. 2.1 Ontology
ASE BD&SI 2015, October 07 09, 2015, Kaohsiung, Taiwan Studer et. al. described "an ontology is a formal, explicit
Copyright is held by the owner/author(s). Publication rights licensed to specification and supports the concept of sharing" [2]. Neches et
ACM. al. wrote "ontology defines the vocabulary and terminology, as
ACM 978-1-4503-3735-9/15/10$15.00 well as in the field, as well as combinations of these rules" [2].
DOI: http://dx.doi.org/10.1145/2818869.2818921 Through the ontology had all the qualities described above,
Ontology is not only for the individual information but also shares 3. RESEARCH METHODS AND SYSTEM
information processing specification. The ontology advantages are
available to express, describing the entity (or object) and the DEVELOP
associations between each other [3]. The system develop user interface on a mobile APP. When users
use the system, they need to log in a social network such as
Ontology can be edited manually, for example usingProtg [4], Facebook, Google +, Twitter, to let system get users friends list
OntoEdit [5], but the process is tedious. So some researchers on the social network. Then, system can get the user's friends
integrate knowledge acquisition and machine learning technology usage statistics and the movie information from Web servers.
to allow the ontology to be generated automatically[6] which Next, pending the receipt of the relevant information converts the
included the technologies of natural language (NLP) [7], data to OWL ontology file and send it to Jena inference engine.
association rule mining [8], hierarchical clustering [9].They use Finally, the results will be presented to the user. The system
the relationship between the rows of the database and the type of workflow is shown in Figure 1.
conceptual analysis [10-14].
The system obtains user friends information and OWL ontology
2.2 Inference Engine files from the web server. According with the similarity of the
Rule ML is an XML-based inference language which defined contents of OWL, the system recommends movies to the user.
rules and knowledge but cannot be reused [15].Rule ML rules
must continue to accumulate and the rule cannot be used
repeatedly. It adds a new rule for every new thing. The knowledge
must be defined on the prevailing rules and the applications
cannot be used on the similar knowledge and rules repeatedly.
JESS is made by Firedman-Hill inference engine. JESS is adopted
Reta algorithm to strengthen efficiency of the inference [16].
However the inferential knowledge needs to match JESS format
for inferences [3], which means that the knowledge and the
mutual are binding engine. Jena [17] is based on inference engine
and developed by Java language. Its inference has good
permeability and the software is free. Jena resources are rich and
its interface is free. Jena provides AndroidJenas API to let mobile
system development becomes possible and do inference apparatus
of ontology. In this paper, we will use Jena as an inference engine.

2.3 Personal Recommendation System


Currently, many papers focus on the personalized
recommendation system. Fong, et al. [3] and Sicheng et. al. [4]
proposed methods to calculate the emotional intensity within a
certain time period. In this way, they need to do to be Figure 1. The workflow of the system.
photographic equipment measuring tools. Through emotions we
The building processing is based on Figure 1 to develop a mobile
can recommended the results for classification as well as the user's
recommendation system. The steps are described as follows.
experience of the movie, and you can get a good result. However,
the user must set up the camera while watching the movie and that Step1: Build a Web site and APP. When user login a community
need to watch for a long time. At the same time, a lot of image network, the system can obtain the user's information and the
analysis and calculation is not applicable for mobile devices. association of user's friend information quickly.
Besides, the music and videos of YouTube are at the same manner Step2: Store the "video information", "user and user's friend
to do recommendation. The main point of difference between watching video record", "the friendship user" in the web server's
YouTube and our method is the users can only follow their own database.
preferences to find relevant videos but our method can combine Step3: Define and plan OWL to construct ontology. Due to the
social networks information to recommend the music and movies. OWL ontology is more complex, therefore we choose graphical
In this study, Ontology Language (OWL) is used to establish interface of Protg tool to build ontology, it was easier and fast.
ontology. There are user ontology and movies ontology which can The figure of the establish screen is shown in Figure 2.
be used to record the individual information of ontology and the Step4: Load ontology data to the mobile device. Then the system
film category for the user's emotional intensity parameter. In can do inference. The candidate movies are calculated by a
addition, combined with the social circles, the system uses the decision-making algorithm. It generated dynamic rules from
ontologies between user and friends. The mobile device receive inference rules.
the statistical results of personal information from Web server, the
system dynamic generate inference rules and modify users' The system will modify Kwona[17] method which is based on the
personal OWL ontology files. Finally, the system combines the normalization of the user's preferences and then to calculate the
individual ontology of content and use Jena inference engine to similarity between users. The detail processing of establishment is
process and present the results of recommendation movie to the based on a threshold which described as follows.
user.
the system can use the information as condition parts to generate
rules for the system inference.
In general, ontology and the inference rules are static which are
not changed. The static inference rules will generate a fixed result.
In this study, the system will find ways to generate dynamic rules
by way of data mining. However, the rules of pre-decision parts
and condition parts are elusive. Even, the rules are found may be
not suitable for users. This requires a complex mechanism to get a
long-term feedback from users.
This study used a common rule as a rule template, such as If part
and Then part. When the user has made the list of candidates, the
multiple preferences and the similar values are the condition part
of a rule. Due to the condition part may have a number of unique
combinations of variables, the following guidelines need to design
an algorithm to determine the conditions part of the rules how to
choose Figure 4 shows the Dynamic Rule Generation Algorithm.

Figure 2. The ontology structure. Table 1. User preferences and normalization.


Step 1: The system collect user preferences Accept Rate (AR). Class Comedy Adven Action Roman Drama Horror Science
That are commended example of films is shown in Table 1. First, ture ce Fiction

when the user has received a service, the service type of user User1 8 11 23 0 3 0 18
acceptance rate in the cumulative way was record 1. Then, the User2 5 8 4 5 3 15 0
user's preference acceptance rate normalized. Next, set up a
threshold value . When the acceptance rate is greater than the User3 8 4 14 4 6 28 4
threshold , the value of NP result is 1 otherwise it is set up to 0. 1 User4 4 4 22 5 16 3 1
is a clearer user preference but 0 is vice versa. In this paper, we User5 8 16 3 0 4 21 9
set the threshold is 0.5.
User6 11 9 0 5 28 1 4
Step 2: When we normalize all users AR. And Then we use
equation (2) to compare all user' similarity rate. M is the number User7 9 1 5 21 7 9 25
of class and m is the index of class. Through the equation (2), we User8 0 3 5 1 1 1 0
use XOR to calculate similarity rate of users. Then user's User9 27 4 8 3 7 3 11
preferences are close, user's SR are smaller.
User10 4 2 4 0 1 0 10
User11 0 8 3 5 3 6 7
0, (AR) <
= { (1) User12 1 2 7 5 3 8 20
1, (AR)
User13 19 5 8 0 0 0 22
User14 2 5 19 3 6 7 0
, =
=1 ( , ) (2)
User15 5 2 4 1 0 27 2
User16 1 2 1 17 5 18 0
_ = {,1 , ,2 , , , } (3) User17 0 21 13 4 5 12 11
User18 1 0 12 5 0 12 1
Step 3: Sort the value of SR. After calculating the similarity by User19 6 2 11 17 7 12 0
equation (3), the system has user performs sorting to select similar User20 1 26 0 27 1 11 10
interesting users. In formula (3), p is an integer value to stand for
the number of users. The SR_Ranki is the set of order of
candidates recommended in order by the value of ,1. 4. RESULTS
In this experiment, the system has collected 100 films from Yahoo
Algorithm 1 Candidate_Decision is used to calculate SR values movie. Twenty users are used to evaluate the system. Figure 2
between the user and community members based on the similarity shows the structure of the ontology. The system builds a user and
candidate decision and normalization method. When the degree of the user's friends in the Person category. The movies are built in
similarity is 0, then the candidate is included in the first list of Public_Movie category. Public_Moive is the different sub-
recommendation candidate similarity. The similarity value categories of Genre. The left side of Figure 2 shows the movie
calculation is shown in Figure 3. categories. The right of Figure 2 displays the name of Action film
Step 4: Find the multi preferences of user. The preferences entity. The entity is Movie Title. The year, recommended degree
include long-term, short-term preferences and the other and other information is recorded in the entity's property of movie.
preferences. The long-term preference is an user watching movies After calculating by formula 1, the user's preference and
after a period of time but the short-term preferences is the normalization of acceptance rate is gotten. The threshold value
watching movies currently. The others preferences mean user's 0.5. AR can be used for a user to prolonged track behavior and to
friends loving. When the system has multi-preferences of users, calculate long-term preferences.
The similarity calculation is based on user 1 to find the relative
users. Users 4 can get the NP from the table, and the number of
categories for the movie is m; =1 is user 1; j is the index of user's
friend for the similarity comparison. The similarity of user 1
compare to his friends is calculated by Equation 2. The similarity
of user 1 and each friend's preferences are shown in Table 2.
The similarity is calculation by using the formula of the sort
algorithm 3. SR = 0 for the first nominate candidates, SR = 1 for
the second candidate, SR greater than 1 for the third candidate. In
this manner backward sort, the top three candidates just to see the
similarity of the human
In the Experiments use of the contents of the top three candidates
recommended OWL ontology movie to the user as a
recommended reference index. The differences in the degree of
similarity recommendation, explore the potential user preferences.
For all users, no similarity rate is 0, so the first candidates are
users 1 of similarity rate, and then second is 2. Any level
candidates can have many people
Table 2. The calculation of the similarity.
User1 compare with all users
User 2 3 4 5 6 7 8 9 10 11
SR 4 1 2 4 3 3 2 3 2 4
User 12 13 14 15 16 17 18 19 20
SR 1 2 1 3 4 2 2 3 4

Protg OWL ontology file generated into action means follow


Figure 3. The candidate Algorithm. Figure 5. AndroiddJenas rules produce the results of user
recommendation. So long-term use in user, you can continue to
add and modify information in the OWL. For example: User's
friend has new similarity data or recommendation candidates.

Figure 5. Using ontology for android APP.


Due to watch movie will be different because of time or different
user scenarios and friends loved, so this research program a preset
Figure 4. The dynamic ruler generation algorithm.
rule schema. And use dynamic rules algorithm 2 to recommend SemanticWeb. Proc. First Intl Semantic Web Conf. Semantic Web,
results. pp. 221-235.
[6] Maedche, A. and Staab, S. 2001. Ontology Learning for the
On the results dynamically as users recommend results by Semantic Web. IEEE Intelligent Systems, No. 2, pp. 72-79.
similarity (SR) calculated for mobile devices, and the use of
[7] Todirascu, A., De Beuvron, F., Dan Galea, and Rousselot, F. 2000.
dynamic rules generated, showing recommended to the user in the Using Description Logics for Ontology Extraction. Proc. First
mobile device. WorkshopOntology Learning at the 14th European Conf. Artificial
Intelligence.
5. CONCLUSIONS AND FUTURE WORKS [8] Maedche, A. and Staab, S. 2000. Discovering Conceptual Relations
This study combined with community networks, the user, friends from Text. Proc. 14th European Conf. Artificial Intelligence, pp.
and film and other related information in OWL ontology. Then, 321-325.
find the user and the user's friends on the acceptance rate of the [9] Clerkin, P., Cunningham, P. and Hayes, C. 2001. Ontology
film after normalizing. User similarity is calculated to identify the Discovery for the Semantic Web Using Hierarchical Clustering.
recommended movies. Finally, the algorithm generates a dynamic Proc. First Semantic Web Mining Workshop.
inference rules and the Jena inference engine to reason the user [10] Cimiano, P., Staab, S. and Tane, J. 2001. Deriving Concept
promoted videos. So, we can find relative movies and increase the Hierarchies from Text by Smooth Formal Concept Analysis. Proc.
interaction and communication between friends. GI Workshop Lehren-Lernen-Wissen-Adaptivit , pp. 72-79.
[11] Quan, T.T., Hui, S.C., Fong, A.C.M. and Cao, T.H. 2006. Automatic
In the future work, we will collect more data to strengthen the Fuzzy Ontology Generation for Semantic Web. IEEE Trans.
accuracy of the system to modify the inference rules. We also Knowledge and Data Eng, Vol. 18, No. 6, pp. 842-856.
hope to combine with intelligent hand ring, collecting heartbeat [12] De Maio, C., Fenza, G., Loia, V. and Senatore, S. 2009. Towards an
pulse data, to analyze user emotions when watching movies, as a Automatic Fuzzy Ontology Generation. Proc. IEEE Intl Conf.Fuzzy
movie recommended weights. On the other hand, in order to allow Systems, pp. 1044-1049.
the user to reduce the search time of the film, we will consider [13] De Maio, C., Fenza, G., Loia, V. and Senatore, S. 2009. A Multi
joining rough set method to solve the data sparseness problem. Facet Representation of a Fuzzy Ontology Population. Proc. IEEE/
WIC/ACM Intl Joint Conf. Web Intelligence and Intelligent Agent
6. ACKNOWLEDGMENTS Technology , pp. 401-404.
This paper is supported by Ministry of Science, Taiwan, with [14] Xu, X., Wu, Y. and Chen, J. 2010. Fuzzy FCA Based Ontology
project number: MOST-103-2221-E-324-028 and MOST-104- Mapping. Proc. First Intl Conf. Networking and Distributed
Computing, pp. 181-185.
2221-E-324 -019 -MY2
[15] Forgy, C. L. 1991. Rete: A fast algorithm for the many pattern/many
7. REFERENCES object pattern match problem. Artificial Intelligence, Vol. 19, No. 1,
pp. 17-37.
[1] Juang, C.H. 2005. Collaborative filtering recommendation Groups.
National Central University Department of Information Management [16] Liu, C. H., Chang, K. L., Jason, Chen, J. Y. and Hung, S. C. 2010.
Master's Thesis, Taiwan. Ontology-Based Context Representation and Reasoning Using OWL
and SWRL. Processings of Eighth Annual Connumication Networks
[2] Studer, R., Benjamins, V. R., and Fensei, D. 1998. Knowledge
and Services Research Conference, pp. 215-220.
engineering: principles and methods. Data & Knowledge
Engineering, Vol. 25, No. 1-2, pp. 161-197. [17] Kwona, O., Leea, Y. and D. Sarangibc. 2011. A Galois lattice
approach to a Context Aware privacy negotiation service. Expert
[3] http://jena.apache.org/.
Systams with Applications, Vol. 38, No. 10, pp. 12619-12629.
[4] Noy, N. E. and Mcguinness D. I. 2001. Ontology Development 101:
A Guide to Creating Your First Ontology. Technical report, Stanford
Medical Informatics, Palo Alto.
[5] Sure, Y., Erdmann, M., Angele, J., Staab, S., Studer, R. and Wenke,
D. 2002. OntoEdit: Collaborative Ontology Development for the

Das könnte Ihnen auch gefallen