Ontology Based Content Management for Digital Television Services

Ontology Based Content Management for Digital Television Services

Benjamin Lui1, Dickson K.W. Chiu2,3, Haiyang Hu4,5, Hua Hu4,6,Yi Zhuang4,5

1Department of Information Engineering, The ChineseUniversity of Hong Kong

2Dickson Computer Systems, 7 Victory Avenue, Homantin, Kowloon, Hong Kong
3Department of Computer Science and Engineering, The Chinese University of Hong Kong

4College of Computer Science and Information Engineering, Zhejiang Gongshang University, China

5State Key Laboratory for Novel Software Technology, Nanjing University, China

6College of Computer Science, Hangzhou Dianzi University, China

email: , , {hhy, huhua, zhuang}@mail.zjgsu.edu.cn

Abstract

Traditional Content Management Systems (CMS)for Digital Television (TV) services usually attempt to provide friendlyuser interfaces with some content information for attracting new subscriptions. However, the searching and subscription functionsprovidelimited abilities foruser requirement elicitation and matchmaking of contents. With the help of ontology, this paper purposes an agent-based Content Management System for Digital TV services. Not only users can retrieve matched contents effectively, service providers can also obtain multi-level analysis of user behaviors and achieve cross-sale of contents and services. We demonstrate the flexibility of our approach with a sample ontology and system architecture.

Introduction

Content management systems (CMS) are widely used in various industries. Financial content distribution, multimedia contents,and books sharing are some of the examples that rely on CMS to store and distribute different information through electronic means. Although CMS can provide some basic support for information searching, it is difficult to provide users with the relevant information efficiently and effectively. This is mainly due to the difficulty in classifying the contents. Smeulders [1] called such difficulty of classifying the information “semantic gap” - the discrepancy between the way video contents are coded digitally and the way they are experienced by human users.

In the Digital TV (DTV)industry, channel subscription is the core business process for making profit. A portal subscription system is usually provided for users to subscribe channels through set-top boxes. However, the portal systems usually operate in a single direction (see Figure 1):

It allows users to subscribe contents and various services, with limited ability in providing recommendations of other channels or services.
It provides a friendly user interface, but the subscription is user-driven. This means users subscribe the channels or Video-on-demand (VOD) contents after they preview the content and feel they like that.

Figure 1. Process of (a) channel browsing and subscription in DTV services; (b)VOD or other multimedia clips searching and subscription

Recently, agent-based system design has been maturing to make system interactions more flexible and automated [2]. Agents can react to environment changesand customer interactions with DTV servicessuch as channel viewing, subscription, and Electronic Program Guide (EPG) browsing. Agent-based content management systems embedded in DTV portal subscription system can help proactively sense those interactions, match the right content or channels to user preferences, and provide recommendations to users for up-sale. We use ontology for achieving effective matchmaking and recommendation, since ontology provides the key knowledge about the inter-relationships among the issues and alternatives of the user subscription preferences. Ontology-based analysis of the preferences can be automated in the content management and subscription processes. Table 1 analyzes how ontology can improvecontent management system.

Limitation of traditional content management system / Contributions of Ontology
Match-
making / Match-making is often ineffective because of rigid definition of multimedia contents (usually, categories are used) predefined by service providers / Shared and agreed ontology provides common, flexible, and extensible definitions of multimedia contents for match-making and subsequent business processes
It is difficult to specify unclear types of multimedia content which are out of predefined categories / Complicated requirements can be decomposed into simple genres for streamlining the elicitation of options
Recom-
mendation / Recommendations are often only possible within the same category / Ontology helps elicit alternatives for recommendation
Pre-defined formulae for every type of multimedia contents are needed for evaluation / Ontology help recommendation by evaluating offers in terms of flexible overall scaling
Business analysis / Data mining is simple which mainly depends on viewership of single channel / program only / Ontology help analyze the viewership of related channels and programs to achieve a better marketing strategy

Table 1. Contributions of ontology to CMS of Digital TV services

As such, we introduce an end-to-end agent-based content management system (ACMS) in this paper, which helps improve customer satisfaction in enjoying more interesting contents while exploring potential business growths of services and content providers. We employ an agent-based design to provide the following key functionalities:

-proactively search channels or VOD contents for users based on different categories;

-recommend channels potentially interested by users for subscription based on channel classifications, user subscription, and profile preferences for up-sale;

-recommend and notify users if there are new contents that users may potentially interested in;

-provide statistical reports of users subscription in both vertical and horizontal domains so thatthe service provider can review the business needs and then look for new potential contents.

The remaining part of the paper is arranged as follows. Section 2 analyzes different CMS and content classification methods purposed in other studies. Section 3 presents a conceptual model of our ACMS based on ontology. Section 4 describes a motivating example ontology and discusses how ontology is useful in the end-to-end flow of DTV services. Section 5 outlines our system architecture and some implementation details before we discuss our summary in section 6.

Related Work

Advancement of multimedia encoding methods and distribution of network storage attracts many researches to look for a better way in annotating multimedia contents and eventually distributing to users effectively. Some of them focus on classifying and annotating the video content, while some of them analyze the components required for the system in order to satisfy the needs of service providers and users.

2.1.Content management and content annotation

There exist many literatures on the content management, especially for matchmaking of the content. TV-Anytime forum is one of the active working groups in suggesting metadata specifications for multimedia content matchmaking [4]. However, keywords for multimedia contents are subjective. There is a semantic gap between the content providers (who categorize the multimedia sources) and content users (who want to find the best fit content to by entering certain keywords). It is a hottopic in minimizing the gap to allow an effective and efficient way to matchmaking the multimedia content.

Kubota[5] proposed using XML-based knowledge cards to store the multimedia content and description. Story cards and Question-Answer (QA) cards are used as a knowledge channel for content providers to distribute their “stories,” while end-users can view the description of the stories and perform queries through QA cards. The Hong Kong CyberLibrary (HKCL) [6] developed a pair of Internet Client (Content Management Agent) and Server (Content Management Server) for searchingmultimedia contents by using Universal Multimedia Resource Link (UMRL) or infrastructure related information stored in the CMS. However, the correlation between XML / UMRL keywords and search criteria is limited.

Ontology that encodes knowledge possibly spanning different domains as well as describes their inter-relationships, is applied in the iJADE CMS [7] and aSDMS CMS [8]. iJADE CMS focuses on how Web mining techniques can be effectively applied with Chinese contents, with the integration of various artificial intelligence (AI) techniques such as intelligent agents, ontology, and fuzzy logic based data mining schemes. For aSDMS, a set of ontologies representing information used by students and faculty members are described with the Resource Description Framework (RDF) and RDF Schema (RDFS). This supports users to navigate and edit the information through a Web interface. A prototype using RDF Gateway has been developed for demonstration.

Antonella et al. [9] [10] and Garnaud et al. [11] suggest the use of ontology to improve the matchmaking process. They develop systems that can perform automatic shot detection and supports users during the annotation phase in a collaborative framework by providing suggestions on the basis of actual user needs as well as modifiable user behavior and interests. Their systems automatically assign annotation for video content with the help of ontology.

However, the above-mentioned researches concentrated in describing the contents and searching methodology. The RDF and RDF schema can be referenced for the categorization. However, they ignore user requirements and preferences, which are important forservice provider in reviewing the multimedia contents to offer in order to eventually achieve win-win situations. In addition, our previous research on the application of ontology for user requirements and preferences elicitations in e-marketplaces [3] provides a strong foundation for this research.

2.2.End-to-end systems

End-to-end systemsare designed to favor both service providers and users, emphasizing on interactions.

Goto and Miyazaki [12] developed anagent-based TV system that controls the TV and peripherals on behalf of usersas well as provides information to the users. The system includes multiple agents that can answer users’questions about TV programs. Relevant information can be searched for users to understand more about the TV programs. However, only documents distributed in limited knowledge resources (such as relational databases) are searched based on just single-word extraction. So, matchmaking is not very effective.

Oh and Lim [13] implemented a hybrid EPG agent system that can provide detailed information for a specific program by combining information from broadcast and scattered Webpages. The searching mechanism is based on program title, broadcasting station, broadcasting time, and duration. One of the advantages of the system is its integration with the Internet. In addition, keywords definition is easy. Only the Program and System Information Protocol (PSIP) table is required to be extracted from digital Transport Stream (TS). Modularized architecture allows system enhancement and maintenance with ease. However, keywords based on PSIP table in TS are subjective,and therefore this limits the searching and recommendation ability. The system focuses on user functionality but provides very little information and functionalities to service provider to react with users in either content enrichment and communication.

Multi-Agent Open Architecture for a TV Recommender System [14] is another interesting idea that applies OWL Ontology for video content annotation and recommendation. The recommender system, called AVATAR1,is a modular multi-agent architecture that combines different knowledge inference strategies (such as Bayesian techniques, profilesmatching, and semantic reasoning). TV contents ontology, based on TV-Anytime metadata specification, is employed to infer new data from the known information. However, this system does not consider service providersvery much. There is no sub-system in evaluating viewership in a macroscopic view, which is useful for service providers to review the business strategy, and consequently also help in improving customer satisfaction. There are no alert and news sub-systems to proactive notificationof new or relevant contents to customers.

Conceptual model and methodology

In this section, we present our conceptual model for content searching and subscription. We also discuss the ability to support key business processes of DTV services using ontology. The processes involve users’ searching requirement elicitation, matchmaking, and recommendation, together with business analysis and formation of new marketing strategy.

Figure 2. Conceptual model of an ontology-based content searching and subscription in UML class diagram

Figure 2 illustrates a conceptual model of our CMS for DTV services. In our CMS, users look for attractive contents provided. They provide the criteria for matchmaking the contents and decide if the recommendations suit them.

Each searchprocess (see Figure 3) consists of tasks that compose with requirement issues. Based on the requirement issues, an agent tries to map each issue as a genre based on an agreed ontology. There are two types of genre. Basic genre can be mapped by the issue if there is exactly one genre mapping in ontology. However, if an issue can break down into several genres according to the ontology (e.g., concert => music performance), auxiliary genres are mapped. Alternative genre with corresponding contents can be recommended (e.g., fighting vs. Kung-fu) to users for better serving user needs and for increasing business opportunities. If users accept the alternative, such contents can besuggested to users for further selection.

The agreed ontology helps the CMS maps users’ requirements to genre and also the inter-relationships among genres. Recommendations of content can be achieved. From theperspective of service providers, ontology inter-relates genres so that segmentation of contents and users can easily be formed for statistical analysis. Figure 3 illustrates our process model. The requirement elicitation phase achieves matchmaking based on the issues identified by the user. Various contents are retrieved based on the ontology. In business exploring phase, recommendations of the confirmed content can be listed for the users to attract their further subscriptions.

Figure 3. Process model of the purposed CMS in UML Activity diagram

Application of Ontology

This section presents a motivating ontology and discusses how ontology helps the overall content management process. Figure 4 shows an example ontology (in UML class diagram) for a selection of requirements of channel or multimedia content searching. Subscription consists of a channel and is also characterized by user profiles (i.e.,customer segment and subscription history) and marketing offers. Content consists of a number of attributes that are used as the selection criteria, such asyear, cast, director, and genre. Genre can further be decomposed by different attributes, like class, category, and language.Figure 5 lists partially the example ontology used our ACMS based on the ontology shown in Figure 4.

Figure 4. A simplified ontology of content annotation in UML class diagram

<owl:Ontology rdf:about="#Content">

<rdfs:comment>SimplifiedContent Ontology</rdfs:comment>

<owl:Class rdf:ID="Content" />

<owl:Class rdf:ID="Genre" />

<owl:Class rdf:ID="Category">

<rdfs:subClassOf rdf:resource="#Genre" />

...

</owl:Class>

<owl:ObjectProperty rdf:ID="hasGenre">

<rdfs:domain rdf:resource="#Content" />

<rdfs:range rdf:resource="#Genre" />

</owl:ObjectProperty>

...

<owl:DatatypeProperty rdf:ID="class"> <!-- Enumeration --!>

<rdfs:domain rdf:resource="#Genre"/>

<rdfs:range> <owl:DataRange> <owl:oneOf> <rdf:List> <rdf:rest> <rdf:List> <rdf:rest>

<rdf:List> <rdf:rest<rdf:List>

<rdf:rest rdf:resource="

<rdf:firstrdf:datatype="

</rdf:List</rdf:rest>

<rdf:first rdf:datatype="

</rdf:List</rdf:rest>

<rdf:first rdf:datatype="

</rdf:List</rdf:rest>

<rdf:first rdf:datatype="

</rdf:List>

</owl:oneOf</owl:DataRange</rdfs:range>

</owl:DatatypeProperty>

…

</owl:Ontology>

Figure5. Partial Ontology Listing of the Multimedia Content annotation

4.1.Understanding requirements from ontology

The main difficulty of searching suitable contentsfor users is the existence of the semantic gap. The inconsistency in the represented values from users’ and service providers’ perspective usually limit the potential content matchmaking. Ontology addresses this constraint and presents machine understandable semantics of the requirements about the contents. The requirement elicitation processes are suggested as follow:

Match key requirement issues like class, category, language as genre

For each issue, check if a direct mapping to a genre based on an agreed ontology is possible. If not, the agent tries to see if sets of relating genres can be mapped by well-known graph search algorithms [15].

If the user accepts the new genre, new contents are recommended. Further the agent refines the issue with new search criteria.

Another set of related genres can be extracted and eventually more and more content matched users’ interests can be provided.

4.2.Analyzing the market with the help of ontology

With the help of ontology and appropriate ontology language, the inter-relationship of contents can be effectively segmented for the service providers to have multi-level market analysis. Statistics of a single contents and set of inter-related contents can be generated. The agent-based CMS can therefore increase business opportunities by mapping of potential cross-sale, grouping users for forming segment-based marketing strategies.

System Architecture and implementation

Figure 7. System implementation architecture

Figure 7 shows the implementation architecture of our ACMS. The architecture is designed not only to support matchmaking and recommendation for users by applying a mutual understanding ontology, but also explore potential business opportunities for service and content providers. The design consists of the following four subsystems.

The Ontology generation and maintenance subsystem aims to annotate multimedia content automatically by using ontology as illustrated in section 4. An ontology editor helps the service provider adjust the instance granted to contents for matching the business strategy. The Matchmaking and recommendation subsystem allows users to specify and edit their required issues and alternatives based on ontology. It supports agents to proactively provide updates to users based on user profile and subscription behaviors. Agents monitor user browsing and subscription practices and advises new program / channel content offers. Users can retrieve contents potentially interested by help of the search engine. The search engine in this subsystem selects the most appropriate ontology based on a given set of criteria and issues.