Comparison of Semantic-Enhanced Search Services in UPnP-AV Multimedia Systems

Ubiquitous home multimedia systems can be realized by the use of UPnP-AV standard so that home users can play and share multimedia contents across a wide range of devices from any locations inside the local home network. However, due to the lack of semantic support in the native search service, user queries are merely limited to mediaspecific properties of audio/video resources in locality. This paper aims to investigate the performance of semanticbased searching mechanisms on the support of user queries with more usability advantages. Based on our simulated data set of enhanced UPnP-AV data with external annotation from the Linked Data cloud, the semantically enhanced search service semantic with indexing search algorithm is far more appropriate than the keyword-based counterpart. Thus, it should be recommended by default, especially in the latest version of UPnP-AV, where the power of new features is utterly driven by the semanticbased content description. 


I. INTRODUCTION
Universal Plug and Play-Audio/Video (UPnP-AV) is an implementation framework for building ubiquitous home entertainment system [1], where users can search computer, and manage to run or display these files on any AV devices locating in the same UPnP-based network environment.Over the last decade, the basic architecture of UPnP-AV system has been researched extensively to cope better with new demands of innovative applications in many domains, such as media cloud service [2]- [3], and ubiquitous mobile learning [4].However, a problem of poor usability occurring in the primitive searching service remains existed so far.Therefore, due to the limitation of using keyword-based matching scheme by default, searching with a rather complex string, e.g."Find all pop songs of the singer named Taylor Swift" will not be applicable in the scheme.
In this paper, the problem of limited search capability in the UPnP-AV system will be tackled.The goal is to enable advanced semantic searching functionality by the notable means of Semantic Web technologies and Linked Data principle [5].Although our semantically-enhanced UPnP-AV system shares a common architecture with Manuscript received December 20, 2015; revised March 10, 2016.those works (such as [6]- [9]), our distinctive concerns are on the efficacy of different semantic searching algorithms, and related metadata management, deploying at the content sharing server.The key contribution is an advanced search engine that is capable of facilitating the user queries with more usability advantages and wider search coverage, i.e. including media description files from multiple devices or even the Cloud drives.Therefore, the user's burden can be relieved greatly.
The rest of this paper is organized as follows.In the next section, the discussion on why the UPnP searching mechanism must be worked on the semantic basis instead of the syntactic matching style is carried out, following with the semantically-enhanced UPnP-AV architecture under consideration including some extra functionalities.Then, the performance evaluations of retrieval system affected by different semantic indexing algorithms in the initial prototype is reported.Finally, the conclusion of this paper is in the last section.

A. Limited Media Search Capability in UPNP-AV
According to the UPnP-AV reference [1], all media contents (i.e..mp3or .mp4files) must be associated at least with a metadata (XML) file in order to specify their descriptions using the MPEG-7 metadata and structuring in the DIDL-Lite format.For instance, a DIDL-Lite XML document of media contents given in Fig. 1 is a logical package containing two digital items (DIs).In each DI, a sub-item (also called as Component) will be used for binding to actual media resource (i.e. a media stream or a picture file).To search for media information in hierarchical elements of the DIDL-Lite XML documents described above, users are demanded to consult the Content Directory Service (CDS) located in the Media Server as shown in Fig. 2 as the CDS works on a basis of syntactic comparisons, various attributes of MPEG-7 metadata (as shown in Table I) will be searched to find a good match for the given strings.In this regard, user's media queries are restricted merely to the media-specific properties of AV resources.Then, some users, who may prefer to specify their queries with broader properties or in an ambiguous manner, will never be friendly with the limited capability of native UPnP-AV search.To resolve the limited UPnP-AV search capability, many solutions have been proposed in the literature.For example, a semantic model relying on the RDF storage was suggested in [6] for enabling the data linkage to collect more data from the web.In [7], a new repository is advocated for enabling users to search the contents from a single point, rather than browsing them on many servers by themselves.However, this work just facilitated users for having a convenient search, but ignoring the improvement of searching quality.This is in contrast to the studies in [10]- [12] where semantic-based methods are actively performed and better searched results can be then yielded.Therefore, it is certain that the semantic functionality must be involved for efficiently handling the limited media search capability in UPNP-AV.

B. Limited Storage for Advanced Semantic Search
The current form of media storage in UPnP-AV is particularly designed for basic XML document storage, since only the tag-based searching is required.This is inadequate for supporting the advanced semantic search, such as semantic indexing search, where the XML-to-RDF conversion is required [13]- [15] so that 3 different forms of RDF storages can be made possible, according to the storage organization, namely triple table, property table and vertical partitioning.Hence, it is our intention to be study further on the suitability of these RDF forms for serving our advanced searches.

III. UPNP-AV ARCHITECTUR UNDER EXPERIMENT
In this section, the proposed extension of UPnP-AV architecture, which contains three types of data storages will be described so that it can be used adequately to support our experiments.

A. Gateway Functionality
The architecture that can alleviate the aforementioned limitations of search capability in UPnP-AV environment can be depicted in Fig. 3 Indeed, it is 3-tier standard UPnP-AV architecture [6]- [7], but the Media Server is now acting as a gateway and is proposed to provide advanced features: a) handling of the RDF data from the external sources (Web of Data) via the Linked Data [16] connectivity and b) enabling a more-capable search service of media contents via semantic-based operations.Three of four components serving for different management functions within a gateway will be described below.

B. Metadata Management
It includes two key processes: a) integrating metadata from two different sources of Media server (1.1a) and the external sources via Linked data (1.1b), and b) converting them to both RDF and JSON formats (1.3) by using the vocabulary mapping method.Table II gives examples of RDF conversions of metadata (on the left side) by using the annotations from the music ontology (mo) [17], RDF [18], RDF schema (RDFs) [19].

C. Storage and Search Management
To provide an effective search of the new type of contents integrated form external and internal source, the database manager must be carefully indexed the content so that the fast access and retrieval can be resulted.In this regard, two possible forms of data can be stored: One is in XML format as same as the other native data.The other one is in non-native formats (i.e.RDF or JSON format), which is a preferred choice of data format used in the semantic process related to the Linked Data.Hence, it is worth to investigate on the performance of different search mechanisms in order to identify which format should be used in the database storage.

IV. PERFORMANCE EVALUATIONS
In this section, the experiments and results will be explained in the two following sub-sections:

A. Evaluation Methodology
The proposed architecture is evaluated by effectiveness and performance the searching in three situations to find the quality performance.Hence, three different searching mechanisms will be involved: Metadata search, Semantic search by triple table, and Semantic search by property table.This is corresponded to different types of data storage as shown in Table III, which demands the different searching mechanism for yielding effectiveness.

B. Experiment Setup and Dataset
The dataset used in our experiments contains different size of media contents, i.e. 20 k, 120 k, 250k and 500 k imported from the SWAT Projects -the Lehigh University Benchmark (LUBM) [20].In the case of metadata indexing, we deploy the Apache Lucene [21] to perform the metadata search, while Allegrograph [22] is used for the case of semantic search by using the triple table.Finally, the Apache Siren [23] will be used in the case of semantic search by using the property table.In essence, when comparing between two searching mechanisms in our case, they both contain advantages and disadvantages in somewhat extent as shown in Table V.While the triple-table based semantic search is easy to manipulate on the RDF data, it is slow in searching response.In contrast, the property-table based semantic search has a fast response, but is complex in the manipulation of JSON data due to the relevant conversion of RDF to JSON at the beginning.

VI. CONCLUSIONS
In this paper, the weakness of typical media search merely on media-specific properties defined in MPEG-7 metadata are especially addressed.In favors of semantic technology, it can be possible that user queries can be facilitated with more usability advantages.Based on our experiments and results, the indexing semantic searching algorithm can be best confirmed on the features of fast and smart responses altogether.While the test system showed the feasibility of proposed approach, it unveiled the need for further research in semantics with ontology in order to enable automatic reasoning supports beyond the Linked Data queries.

Figure 2 .
Figure 2. Simplified view of media search in UPnP-AV

Figure 3 .
Figure 3. Extending UPnP AV infrastructure to exploit the external media contents

TABLE I .
METADATA FOR MEDIA DESCRIPTION

TABLE II .
EXAMPLE RULES OF VOCABULARY MAPPING

TABLE III .
TYPES OF STORAGE AND MATCHING APPROACHESBased on the data shown in TableIV, we can see the possible queries that can be applied for several purposes in different searching mechanism.We can conclude the advantage of semantic searches that can allow users to do sophisticated query and hence gaining more usability than the counterpart of metadata search.

TABLE IV .
RESULT OF SEARCHING FROM THREE MODELS