Article Preview
Top1. Introduction
Nowadays, we have access to unprecedented quantities of data which encompass valuable information and knowledge. Generally, these data are composed of heterogeneous data types (HDT) such as texts, images, numbers, metadata, videos, etc. In the past, heterogeneous research groups would focus on specific data types which was the base of establishment of new areas in data mining and knowledge discovery such as text mining which focused on texts, image processing which focused on image processing, and computational biology area which focused on DNA and protein processing. Although, these areas have many similarities, nevertheless, each one of them utilizes different methods especially at the pre-processing phase which leads to better results due to the nature of input data. However, heterogeneous data types contain complementary information that could lead, if properly processed, to richer knowledge and discoveries. Recently, multiple research groups started emphasizing on the heterogeneous data mining (heterogeneous data mining) which consists in mining or processing heterogeneous data types. This interests to heterogeneous data mining seems to be proportional to the heterogeneity of the data; the higher is the heterogeneity of the data, the higher is the interest to heterogeneous data mining. For example, river monitoring (Iliou, Anagnostopoilos, Stephanakis & Anastassopoulos, 2015), sentiments analysis (Leitch & Sherif, 2017), data streams analysis (Chen & He, 2016), data integration (Bourennani, Guennoun & Zhu, 2009), stock market prediction (Qiu, Srinivasan & Street, 2006), are some examples of research areas interested in heterogeneous data mining.
When processing heterogeneous data types, there are two options. The first option consists in processing every heterogeneous data type separately as per classical homogeneous data mining (DM) process, then combine the data mining results which is usually complex and was the initial approach to solve heterogeneous data mining problems (Barbro, Jarmo, Hannu & Ari, 2001; Magnusson et al., 2005). The second option consists in combining the heterogeneous data types at the preprocessing level, then process them in a unified manner as a single entity with a single result which avoids the complex task of combining the data mining results. So far, this second approach has been the most successful option such as in (Iliou, Anagnostopoilos, Stephanakis & Anastassopoulos, 2015; Bourennani, Guennoun & Zhu, 2010; Chen & He, 2016).
In this paper, as detailed in Section 5, we propose to solve a data integration (DI) problem using heterogeneous data mining with focus on the preprocessing phase. Data integration requires the identification of similar data entities from heterogeneous repositories to be merged into a single repository to provide users a unified view of these merged data (Lenzerini, 2002). In the proposed case study, described in Section 3, we aim to combine for heterogeneous data mining processing up to seven heterogeneous data types in single unified data model and by using only a single data mining algorithm. The proposed tool should identify automatically data entities that have similar contents to be merged together as a part of the data integration process. As heterogeneous data mining is in its infancy, several questions need to be addressed to solve the data integration case. In this paper, we attempt to answer these questions as follows.