Database Reference
In-Depth Information
embedded system, but the amount of this information for user satisfaction is limited.
Thus, the previously constructed information on the Web can give the efficient and
satisfactory ones to the users. So, the next subsection describes the embedded data
mining method using the Web.
3.4 The Association of the Web
An embedded hardware system has not enough memory devices to manage the data
because it has a resource limitation problem and low performance capability. Actu-
ally, our system has a 512Mbyte working memory (NOR flash memory) and a
256Mbyte Compact Flash (CF) memory. The working memory includes operating
system and some files to boot. It cannot store some information permanently. The
CF memory includes 200Mbyte map data for car navigation, and 30Mbyte TTS DB.
This is due to the cost of car navigation product. However, the user wants to utilize
various information and services from a lot of different information sources. To re-
solve this problem, this system stores and manages only multiple level-of-abstraction.
This mined data for multiple-level association rules are performed on the server-side.
The data mining server plays a role in performing the Web mining. Mining typical
user profiles and URL associations from the vast amount of access logs is an impor-
tant feature. It deals with tailoring the interaction with Web information space based
on information about the users.
The multiple level-of-abstraction is composed of multiple-level association rules
and a summary table. The methods for mining associations at a generalized abstrac-
tion level by extension of the Apriori algorithm is applied as in [14]. The summary
table forms the topic based indexing scheme. It stores basic information about groups
of tuples of the underlying relations. This summary table is incrementally updateable
and is able to support a variety of data mining and statistical analysis tasks. The
summary table forming the indexed file is downloaded from the data mining server
when the system is first switched on at the start of the day and the information is
changed in the data mining server. The generalization process using attribute-
oriented induction approach [14] for summary tables is performed on the server-side.
It extracts a large set of relevant data in a database from a low concept level to a rela-
tively high one. Thus, the system does not spend extra calculation time for data min-
ing on an embedded system.
The sample structure of the multiple level-of-abstraction is as shown in Figure 4. It
has a hierarchy form to index the data. We use two kinds of mined data; news and
traffic information. (a) of Figure 4 depicts the news information. (b) of Figure 4 de-
picts the traffic information. The summary table basically includes the primary key,
data, title, associated URL, and comments. Embedded applications that represent the
news and traffic information just display the multiple level-of-abstraction informa-
tion. If the user wants to see the specific information, that information is downloaded
and displayed on the screen by selecting the specific button, or speaking the title. The
speech interactive agent requests the URL for information to be sent to the data min-
ing server, then the server sends the requested information in a form of HTML type
text using HTTP protocol. The received text information is parsed and passed to the
TTS, then the TTS reads this texts.
Search WWH ::
Custom Search