This paper presents Mining Knowledge TV, a module for data mining that is part of the Knowledge TV (KTV) Project. KTV proposes the specification of a semantic layer that is embedded in a Digital TV (DTV) environment, improving the way that content is accessed by other applications.
Interactive Digital TV [
In this scenario, this paper aims at presenting the specification of
Mining Knowledge TV- MKTV, which focuses on the integration
of data mining [
The main aim of MKTV is the implementation of a KDD
environment, which focuses on data mining and semantic
information on the Knowledge TV platform [
The mining process will be carried out on the data from many
sources, mainly the sources that come from the Service
Information (SI) metadata table, which uses the MPEG2 standard
in the Ginga DTV environment. This standard is used to represent
information about TV programs, services and multimedia
interaction. Examples of such information are channels, program
schedule, program classification, etc. User behaviour is also an
important kind of information source because it indicates, for
example, the channels usually watched with start time and total
watching period. The useful content obtained by means of data
mining will be semantically enriched through the use of
ontologies and then provided as a service to NCL or Java
languages application developers. This is possible because Ginga
supports the development of applications using both languages on
its architecture. More information about the Ginga architecture
can be seen in [
The data mining process acts on all these sources and generates new information that is semantically enriched by means of a domain ontology. This semantic process enables a better analysis and turns more explicit the meaning of the data mining resultant discovered knowledge. This semantic is provided as a service and creates opportunities, which can be used for NCL or Java developers to implement more powerful and sophisticated applications.
Brazilian DTV is being characterized as an environment of technological convergence, new and extremely susceptible to changes. It is not yet completely standardized and it is constantly being updated. In this way, these aspects impose restrictions that we must consider during the architectural modelling. These evaluated aspects can be highlighted as restrictions:
channels, i.e.; the change of channel will delete all application information related to that channel.
All these limitations in the architecture of the STB lead us to use a
hybrid approach detached from the middleware. That means that
the components of the KTV (and consequently the MKTV) with
highest consumption of resources (such as processing power and
memory) will be exploring the Ginga middleware return channel
[
The Mining Knowledge TV (MKTV) is the component of the
Knowledge TV architecture that accounts for the discovery and
treatment of useful knowledge from the DTV data, users
behaviour and other sources such as the Web. These data are
initially stored in a local relational database and gradually we will
start the process of extraction, transformation and load (ETL) of
information. After the ETL process, the data will follow for the
next module that is the Data Warehouse (DW) [
The historical data will be organised in the DW. Next, the Data Mining module applies data mining algorithms, searching and discovering useful patterns and information not known in the existent DW. The knowledge extracted through the MKTV will be encapsulated in semantic files with more expressive power (OWL files). Ontologies specification in OWL will be the standard for communication between the modules of the KTV. Figure 1 illustrates the KTV conceptual architecture and the MKTV module.
One application scenario is the problem of recommendation and personalization of content. To deal with such problems, specific modules, specified on the conceptual architecture, will be instantiated and executed. First the system stores the data that comes from the STB to a database. Then, the information related to user watched programs will be extracted to the Data Mart Personalization in the DW. After this process, it will be used clustering algorithms to find groups with similar preferences. Such knowledge discovered will feed and enrich the ontology specified in the semantic modelling layer and will return the pattern discovered in the form of recommendation to the user. For example, the next available programs similar to the ones the user uses to watch. Depending on the data mining goal, other tasks and algorithms can be applied to discover the desired knowledge.
This paper describes our initial works on the Mining Knowledge TV(MKTV), which is part of the KTV project. The major aim of this project is to provide semantic knowledge to be used for other DTV applications. At the moment, the MKTV is in a development stage, so that we have carried out a survey of the state of the art in data mining for DTV. In addition, we have also identified the main data mining methods and algorithms that are currently used in the DTV, together with a list of tools that are compatible to this new computational and interactive platform.
We can testify the innovation feature of this proposal if we consider the few DTV works that focus on the joint use of knowledge representation and data mining techniques to generate a better quality set of data.
The next MKTV activities intend to simulate DTV data traffic and integrate content from the data mining process and semantic modelling sub-layer. As future work, during its validation stage, MKTV will collaborate with the JCollab Project [16], whose aim is to develop a platform to create journalistic content via a social network. Another potential future work is the investigation about the integration of MKTV solution to other Digital TV systems.