Toward Intelligent Data Warehouse Mining: An
Ontology-Integrated Approach for Multi-Dimensional
Association Mining
Chin-Ang Wu
1, Wen-Yang Lin
2,*, Chang-Long Jiang
3, Chuan-Chun Wu
41
Dept. of Information Engineering, I-Shou University, Kaohsiung County, Taiwan 840, R.O.C., Email: cwu@csu.edu.tw
2
Dept. of Computer Science & Information Engineering, National University of Kaohsiung, Kaohsiung, Taiwan 811, R.O.C., Email: wylin@nuk.edu.tw
3Dept. of Electrical Engineering, National University of Kaohsiung, Kaohsiung, Taiwan 811,
R.O.C., Email: m0955107@mail.nuk.edu.tw
4
Dept. of Information Management, I-Shou University, Kaohsiung County, Taiwan 840, R.O.C., Email: miswucc@isu.edu.tw
Abstract
A data warehouse is an important decision support system with cleaned and integrated data for knowledge discovery and data mining systems. In reality, the data warehouse mining system has provided many applicable solutions in industries, yet there are still many problems causing users extra problems in discovering knowledge or even failing to obtain the real and useful knowledge they need. To improve the overall data warehouse mining process, we present an intelligent data warehouse mining approach incorporated with schema ontology, schema constraint ontology, domain ontology and user preference ontology. The structures of these ontologies are illustrated and how they benefit the mining process is also demonstrated by examples utilizing rule mining. Finally, we present a prototype multidimensional association mining system, which with intelligent assistance through the support of the ontologies, can help users build useful data mining models, prevent ineffective
pattern generation, discover concept extended rules, and provide an active knowledge re-discovering mechanism.
Keywords. Data mining, data warehousing, intelligent assistance, multidimensional association rule, ontology
1. Introduction
With the fast and massive accumulation of data in the current information era, especially under the pervasion of the Internet, the utilization of knowledge discovery and data mining technology plays a key role in promoting business competition and improving scientific discoveries. To prevent ineffectual efforts of garbage in, garbage out computation, contemporary knowledge discovery platforms usually accommodate the data warehouse (Inmon, 1995) as a consistent and integrated data repository. As with the typical data warehouse mining process shown in Figure 1, the data in a data warehouse is extracted from multiple and heterogeneous data sources, undergoes a complex cleaning and integration process, and finally loads into the data warehouse as a consistent and integrated data repository. In this way, users do not need to worry about the heterogeneity and the consistency of the data.
Figure 1. The process of data warehouse mining
In the past decade, much effort has been devoted to data warehouse mining and many applicable solutions have been provided. J. Han’s research group conducted data mining from data cubes and multi-dimensional databases (Han, 1998). They developed DBMiner (Han et al., 1997), a system combining OLAP (Chaudhuri & Dayal, 1997) and data mining to provide association mining, classification, prediction and clustering. Many commercialized database
..… Selecting and Grouping Knowledge Pattern Cleaning and Transforming Data Warehouse Database Target Data Flatfiles ..… Evaluation and Presentation Data Mining
systems, such as Oracle and Microsoft’s SQL Server, also provide data mining from data warehouses or data cubes. These data mining systems provide an integrity data mining environment allowing users, based on their subjective need, to formulate data mining models (queries) interactively. These systems check the legality of mining model structures, yet not the rationalities [rationales?] of them; the users have to tune the mining models repeatedly until they obtain satisfactory results. This problem is exacerbated when mining multidimensional association rules from data warehouses.
For example, consider the star schema (Kimball et al., 1996) in Figure 2. The following mining model, whose structure will be defined later, represents a user’s query for looking at associations among customer’s daily purchase of products in Japan.
Transaction ID: CustID, TimeID Interested mining attribute: ProdName
Condition before grouping: Country= “Japan”.
CustID Gender Education Country Region City Customer SalesmanID SalesmanName CustID ProdID TimeID SalesmanID Quantity Amount Profit Cost Salefact ProdID ProdName Category Type Brand Size Product TimeID Date Month Year Time
Figure 2. An example of sales star schema
The specification of this example mining model includes deciding on the data granularity (Transaction ID), selecting the pertinent mining attributes, and if necessary, setting the filtering condition before and/or after the grouping process and providing essential information to the system to prepare the target data as depicted in Figure 3. Unfortunately, as will be defined later, the mining model specification is a rather complicated task for novice mining users; many unreasonable mining model settings would be specified. Moreover, some users may even not know how to initiate a mining model. Some of the settings are
syntactically legal but semantically unreasonable. Below are some examples of problematic mining models for further illustration.
Figure 3. Data selection and grouping
Example 1:
Transaction ID: Category
Interested mining attribute: CustID, Gender, Education
This is not a sensible mining model due to the improper inclusion of customer ID as a mining attribute. It will create numerous tedious itemsets such as,
CustID = “C001” Education = “High School”. Example 2:
Transaction ID: Gender
Interested mining attribute: ProdName
This mining model will yield only two transactions, not statistically sufficient to generate convincing rules.
Unfortunately, current data warehouse mining systems do not usually provide any assistance in the formation and semantic checking of the users’mining model settings. We argue a good mining system should provide intelligent assistance in crystallizing a user’s mining intention into a good mining model that is syntactically correct and semantically reasonable. Although data warehouses have solved the data preprocessing problems effectively, there is at least the following issues hindering the realization of such an intelligent assistance by the data warehouse mining system:
(1) Lack of semantic portrayal of data
Contemporary data warehouse systems are mostly based on the relational data model organized in the form of star schema, which cannot present full relationships between data. For example, the star schema in Figure 2 simply shows the structural relationships between the fact table and the dimension tables. It overlooks the concept hierarchical relationships between dimensional attributes, as shown in Figure 4, which provides essential information for performing typical OLAP operations, such as roll-up, drill-down, and slice-and-dice. If a data warehouse mining system can provide the users with the concept hierarchical relationships between attributes, the users can perform better data mining. As well as the concept hierarchical relationships, some data constraints beyond the current data warehouse mining systems exist. For example, Gender is not suitable as the only mining attribute. Such constraints can be helpful in the user’s mining model setting.
Customer
GenderCityEducation
Region Country Product Brand Category Size Time Day Month Year Type Salesman
Figure 4. Conceptual hierarchical relationships
(2) Lacking facilities in understanding users’mining intentions
Data mining process is more or less a subjective process that depends on the individual’s intention. A user’s mining intention is represented by his or her mining model setting to obtain a set of mining results. The satisfactory mining requests conducted previously can be an important asset to provide rich information for sharing with other users. For example, they can be further analyzed to provide recommendations for other users to refer. However, the current data warehouse mining systems provide very few mechanisms in capturing and analyzing the users’previous mining request.
To overcome the above deficiencies of current data warehouse mining systems, we contend integrating background knowledge is necessary. Specifically, the providing of concept hierarchical and constraint relationships between attributes can avoid some settings of unreasonable mining models. This reduces the iterative mining processes and further saves the processing cost. Also, maintaining previously conducted mining processes provides recommendations while the user is setting a mining model. This useful information helps the users clarify their real mining intention. In addition, some domain related knowledge beyond the data warehouse can be helpful to the users for setting a more specific mining scope and also extending the rules derived.
As the concept of ontology has emerged into an effective method for domain knowledge representation and sharing (Uschold & Gruninger, 1996; van Elst & Abecker, 2002), we adopt the ontology technique to construct the above mentioned knowledge into the following four groups: (1) schema ontology: describes the schema structure and relationship between the attributes of the data warehouse; (2) schema constraint ontology: describes the constraints between attributes; (3) domain ontology: collects the related domain and expert knowledge; and (4) user preference ontology: integrates the derived common mining models. With the support of these ontologies, we will show in this paper that a data warehouse mining system can provide intelligent assistance to facilitate more efficient and effective mining. Specifically, we take multidimensional association mining as an example to show how a data warehouse mining system can provide intelligent assistance through the support of the ontologies. We will demonstrate such a system, from the user’s perspective, possesses the following advantages:
(1) Helps the users clarify what they need. In other words, it assists the users in building better mining models by providing semantic checking and model recommendations. This avoids the generation of useless patterns, thus the findings of useful patterns can be achieved successfully.
(2) Finds concept extended rules from the existing primitive data warehouse with the help of the proposed domain ontology.
(3) Allows the mining constraints to be set more precisely by including relationship information defined in the domain ontology.
The rest of this paper is organized as follows. In Section 2, we will introduce the multidimensional association rule mining and define its mining models. In Section 3 the framework of intelligent data warehouse mining system incorporating ontologies is presented followed by the introduction of each ontology. How these ontologies can benefit the intelligent mining system is also illustrated with examples of multidimensional association rule mining. In Section 4, we demonstrate the intelligent assistance through examples. Part of the prototype implementation of intelligent system interfaces is presented in Section 5. In Section 6 we discuss related work and finally we conclude and highlight some future research in Section 7.
2. Multidimensional association rule mining model
An association rule has the form, A B, where A and B are sets of items and A B = . The rule implies transactions in the data warehouse containing A also tend to contain B. A is the body or the antecedent of the rule and B is the head or the consequent of the rule. For this rule to be interesting, A and B have to satisfy the user specified minimum support (ms) and
minimum confidence (mc). The support of the rule, P(A B), measures the percentage of the
total transactions containing both A and B. The confidence of the rule, P(B | A), measures the percentage of transactions containing A and also containing B. If the items in a rule involve only one attribute, it is termed a single-dimensional association rule. Further, where the items in a rule involving two or more attributes, it is termed a multidimensional association rule. An attribute is also called a dimension from the perspective of a multidimensional data model.
Definition 1. (Multi-dimensional association rule) Consider a transaction table composed of k attributes (dimensions). Let ximand yjn be the values of attributes Xi and Yj, respectively. The
form of a multi-dimensional association rule is:
X1= “x1m”,X2= “x2m”,….,Xi= “xim” Y1= “y1n”,Y2= “y2n”,….,Yj= “yjn”
Following the work in (Han & Kamber, 2001; Zhu, 1998), the multi-dimensional association rules can be categorized into three types as follows.
(1) Intra-dimensional association rule
This type of rule shows association within an attribute. The items in an intra-dimensional association rule come from only a single attribute. For example, the following intra-dimensional association rule,
ProdName=“LG DVD Burner” ProdName=“DVD-R 8X Disk”,
shows people purchasing an “LG DVD Burner”are also likely to purchase a “DVD-R 8X Disk”, involving the only attribute ProdName.
(2) Inter-dimensional association rule
This type of rule shows association among multiple attributes. The items in an inter-dimensional association rule involve more than one attribute with each attribute appearing only once in the rule. For example, an inter-dimensional association rule,
Gender = “Female”, ProdName = “JVC UX-C305 Hi-Fi System” City=”Taipei”,
involves three attributes with no repetition of any attribute in the rule. This rule indicates the females buying “JVC UX-C305 Hi-Fi”are likely to live in the city
“Taipei”.
(3) Hybrid association rule
This type of rule is a combination of inter-dimensional and intra-dimensional associations. The items in such a rule also originate from multiple attributes, but unlike the inter-dimensional association rule, it allows repeated attributes. For example, the following hybrid-dimensional association rule,
Education = “College”, ProdName = “Acer PC” ProdName = “HP printer”,
shows people with college education who purchase“Acer PC”tend to also purchase
Data warehouse stores complete and primitive data. The user has to specify what subset of the data in warehouse is of concern and what conditions should be satisfied through the construction of a mining model defined below.
Definition 2. (Mining Model) Suppose a star schema S containing a fact table F and m dimension tables {D1, D2,…,Dm}. Let T be a joined table from S composed of a1, a2,…,ar
attributes, so ai Attr(Dk), 1 i r, 1 k m. Here, Attr(Dk) denotes the attribute set of
dimension table Dk. With tG, tM {a1, a2,….,ar} and tG tM = , a mining model of
multidimensional association rules from T is defined as
MM: <tG, tM, [wc], [hc], ms, mc>,
where tG, tM, wc, hc, ms and mc are the mining model elements, each of which described as
follows is involved in acquiring the target data shown in Figure 3.
tG: the transaction ID (data granularity),
tM: the pertinent mining attributes,
wc: the optional filtering condition before the grouping operation, hc: the optional filtering condition after the grouping operation, ms: the minimum support, and
mc: the minimum confidence.
Table 1 is an example based on the data warehouse in Figure 2. From all the customers’ daily transactions, a user wants to know if there are any associations between customers’ education and the purchased products. In such a case, each transaction in the target data is identified by the composite of the customer ID and the transaction date while the mining attributes are education and product. Therefore the mining model will be tG= {CustID, Date}
and tM= {Education, ProdName}. The corresponding target data is shown in Table 1 with six
transactions. However, if instead customer ID is specified as the only transaction ID, i.e., tG=
{CustID}, the target data will be as shown in Table 2. These two target data represent different views with different granularity. Even if a rule can be generated from both datasets, its meaning will differ.
Table 1. An example of target data grouping by CustID and Date
tG tM
tid
CustID Date Education *ProdName
1 C001 2008-02-01 College B,C,E
2 C003 2008-02-03 High School A,B,E
3 C003 2008-02-10 High School A,D
4 C004 2008-02-05 Middle High C,E
5 C005 2008-02-09 College B,C,D,E
6 C005 2008-02-15 College A,B,E
*A: IBM60GB B: IBM TP C: RAM 512MB D: Ink Cartridge E: Hard Disk
Table 2. An example of target data grouping by CustID
tG tM
tid
CustID Education *ProdName
1 C001 College B,C,E
2 C003 High School A,B,D,E
3 C004 Middle High C,E
4 C005 College A,B,C,D,E
*A: IBM60GB B: IBM TP C: RAM 512MB D: Ink Cartridge E: Hard Disk
3. System Framework of Ontology-Integrated Data Warehouse for Multidimensional Association Mining
In this section we will introduce the proposed data warehouse mining system incorporating various ontologies to fulfill the function of intelligent assistance in mining processes. The feasibility of such a system in amending the aforementioned deficiencies of modern data warehouse mining systems will be clarified through suitable examples.
Figure 5 shows the proposed data warehouse mining system framework. Some specific knowledge helpful in supporting the system’s intelligent services is incorporated in this framework, including the characteristics of the data warehouse schema, constraint relationships between attributes, domain specific knowledge and the user preference in the mining model setting. This knowledge is beyond the presentation capability of current data warehouse systems and is structured into four different forms of ontologies: (1) schema
ontology, (2) schema constraint ontology, (3) domain ontology, and (4) user preference ontology.
Figure 5. An Ontology-Integrated Data Warehouse for Multidimensional Association Rule Mining
A mining process begins with the setting of a mining model by the user. According to the specified mining model, the target data is prepared and the mining engine is then launched. The user tunes the mining model repetitively until the satisfactory results are found. The model elements of a satisfactory mining are saved in the mining log, providing data for further analysis to closely gather related model patterns. The analyzed results then are utilized to construct the user preference ontology.
The contents of the ontologies except the user preference ontology are preserved by some experts, all of which, as will be clarified later, are applied to assist the users in settings the mining models by performing reasonableness checks and offering element recommendations. The domain ontology also provides the mining engine with further information to find rules with extended concepts. In what follows we will illustrate the details of each ontology.
Distillation Mining Engine Data Mart Data Mart Flat File Internet Extract Transform Load Data Warehouse
Schema Constraint Ontology
...
Domain Ontology
...
User Preference Ontology
... Mining Model Setting Mining log Data Cube Expert User Target Data Schema Ontology ... Rule Base Operational Data
3.1 Schema ontology
Multidimensional data model shows the inter-relationships of the fact table and dimension tables through the key structures of relational tables. There are other relationships between the dimensions or attributes not shown in the model, yet can benefit the data mining process, including concept hierarchical relationships and different additive characteristics of fact measures in the data warehouse. We use schema ontology to construct such relationships. Figure 6 is an example schema ontology corresponding to the multidimensional data model in Figure 2.
additive fact semi-addit ive fact non- additive fact
dominate dimension optional dimension Product Type Size Brand Category Amount Quantity Profit Cost Customer Gender City Region Country Time Date Month Salesman Education Name SalesmanID ProdID CustID TimeID key attribute dimension root schema hierarchy attribute node Year
Figure 6. An example of schema ontology
There are three additive types of fact measures for OLAP operation. (1) Additive, which can be summed along all the dimensions. Some typical examples are sale quantity and sale amount which can be summed along all the dimensions. (2) Semi-additive, which can be summed only along certain dimensions. For example, the cost can have sensible aggregation only along product dimension. (3) Non-additive, which will not have any sensible summation along any dimension. For example, profit is calculated by subtracting cost from the sale amount and is a non-additive measure. Another clear example, which is not part of the schema, is the balance amount in a bank statement. It cannot generate any sensible summation
for analysis purposes because it is an accumulated amount that is non-additive. The information of the additive types of measures is valuable in that it helps validate user’s specification of aggregation. For example, the target data for the mining model with
tG={CustID} and tM={Cost} will generate summation of Cost along CustID, which is illegal.
With the help of information of additive types in schema ontology, the system can detect such ineffective settings for mining.
The hierarchical relationship is important in that it supports the user in better data selection, therefore, it minimizes the infeasible settings of the mining model. For example, according to the schema ontology, the products: “ASUS EeePC 900”and “ASUS EeePC 1000”have the
following concept hierarchy:
If a user knows there are hierarchical relationships between the type of a product and its category, he or she will not try to dig the associations between them. If they do, trivial patterns will be generated because a product type decides its category. For example,
Type = “EeePC” Category = “PC”
is known to people, therefore the mining is redundant.
3.2 Schema constraint ontology
In (Perng et al., 2001; Perng et al., 2002), the authors explored all possible mining spaces of mining attribute combinations under variational transaction IDs and a defined allowed range through data constraints, which can minimize the search space efficiently.
In multidimensional association rule mining, similar constraints imposed on tGand tMexist.
In addition, we observed some predicates specifically for multidimensional association rule mining. Also, some constraints are domain dependent. For example, tG={Size} may be valid
in the plastic bag business but not in the 3C business. In this paper we present some example constraints for illustration purposes. All these constraint relationships are beyond the representation capability of data warehouses. The schema constraint ontology is used to
EeePC (Type)
ASUS (Brand) PC (Category)
constraint ontology derived from the schema in Figure 2. Some attributes have multiple constraints. These constraints are valuable for the system to verify the user specified tGand tM
of a mining model, and therefore, avoid invalid mining. The following are some examples of constraint relationships related to multidimensional association rule mining.
Salesman SalesmanID Name Product ProdID Size Brand Type Category Customer CustID Gender Education Nationality Region City ItemOnly GroupOnly Decide Follow Decide Salesman SalesmanID Name Product ProdID Size Brand Type Category Customer CustID Gender Education Nationality Region City
Dimension
Attributes
Predicate
Attributes
Dimension
Time Day Month Year Time Day Month Year NoSingleGroup NoIntraMining Exclude GroupOnly GroupOnly
Figure 7. An example of a schema constraint ontology
(1) Decide
This is the relationship analogous to functional dependency. Given two attributes A1 and A2, we say A1 decides A2if for any two transactions t1, t2, if t1.A1 = t2.A1then t1.A2 = t2.A2.
Whenever a functional dependency relationship exists between attributes, redundant mining space or known patterns will be generated and consequently waste the mining efforts. For example, a mining model with tG1 = {CustID, Education} is a redundant form of tG2 =
{CustID} because Education is decided by CustID. The mining space created for tG1 and tG2
will be exactly the same. Another example is a mining model with tM = {ProdName, Brand}
will generate redundant rules because ProdName decides Brand, therefore associations of
ProdName, “IBM TP” Brand, “IBM”,
is an example of a redundant form.
(2) GroupOnly
An attribute with constraint of Is_GroupOnly can be used as a transaction ID only, for example, CustID. If CustID is chosen as one of the mining attributes, the following tedious rule will be generated:
CustID=”C003” Category=”Printer”.
(3) ItemOnly
This constraint shows an attribute can only be used as an interested mining attribute. For example, Size in the 3C Domain is not suitable for deciding the granularity of data.
(4) Follow
Some attributes should be used by following another. For example, a city is located in a region, but different regions could have cities with the same name. Considering the case in the United States, in both states of Tennessee and Ohio, there is a city called Springfield. If the attribute City is used alone as a transaction ID, then the results will be misleading. Therefore, ‘City’use is confined by following ‘Region’.
(5) NoSingleGroup
An attribute with a constraint of NoSingleGroup cannot be used alone as a transaction ID. For example, gender has only two distinct values namely, male and female. Grouping by
gender will result in too few transactions to generate any patterns.
(6) NoIntraMining
An attribute with a constraint of NoIntraMining cannot be used alone as an interested mining item. In another word, it cannot be used to mine intra-multidimensional association rule. Examples such as tM= {Gender} or tM= {Education}, will generate tedious rules like
Education=”College” Education=”High School”.
Such types of attributes are adapted to a constraint of NoIntraMining.
(7) Exclude
This constraint prohibits the simultaneous appearance of attributes. For example, assume the calendar date and fiscal date are both in the time dimension. Suppose a fiscal year starts from Sep 1; it is across two calendar years. Then, an example such as using FiscalYear and
CalendarMonth together as the transaction ID would be misleading. The exclude constraint
can be used to avoid an invalid combination of transaction ID.
In short, the schema constraint ontology is particularly helpful for checking mining model settings. User specification of tGand tMcan be verified with the knowledge presented in this
ontology. Through the intelligent user interfaces, incorrect or inefficient tG or tM can be
reminded and adjusted interactively.
3.3 Domain ontology
Domain ontologies are used to construct the domain expert knowledge related to the mining subject of the data warehouse. The contents are specific to certain dimensions, such as product dimension or customer dimension, to enforce the expression of their semantic relationships. Following the W3C recommendation (Heflin, 2004), a domain ontology can be constructed in the triple format of a “Subject-Relationship-Object”. For example, the following triplet indicates the hard disk is a component of the PC:
where the subject PC has a Composition relationship with the object Hard disk.
Figure 8 is an example domain ontology of 3C products exploring the classification (is-a) and composition (has-a) relationships between products. Other relationships, such as the product features or compatibilities of software to OS or memory to motherboards etc, can be included.
One of the advantages of introducing domain ontology into the mining system is to find concept extended rules from the existing primitive data warehouse. For data with hierarchical relationships, some previous research mentioned extension of association rules (Domingues & Rezende, 2005; Han & Fu, 1995; Lisi & Malerba, 2004; Srikant & Agrawal, 1995) from primitive into a generalized or leveled format. Current data warehouse systems cannot reveal possible extended relationships between data values. For example, the hierarchical relationship in the product dimension of Figure 4 describes only the category and brand of products and their types, yet some possible concept extensions between products such as those presented in Figure 8 are not disclosed. Domain ontology can describe such conceptual relationships relevant to each product.
Printer HPLaserJet 1006 EpsonEPL 6000 Ink Cartridge Photo Conductor Toner Cartridge Laser Inkjet Non-impact Dot-matrix Epson LQ2090 Epson 310 Canon i80 All inone HPCM 1312 Ink Cartridge All inone Epson NX105 Ribbon Cartridges ---Memory HardDisk Notebook DesktopPC PC ---RAM 256MB S 60GB IBM 60GB RAM 512MB Sony VAIO Gateway GE IBM TP Composition Classification
Figure 8. An example of a 3C domain ontology
Besides, different product may have different levels of conceptual hierarchies. For example, in Figure 8 we can see Epson LQ2090, Epson EPL 6000 and Epson NX105 have
one, two and three classification levels respectively. Therefore, it is possible to find extended rules by incorporating the domain ontology with data warehouse mining. For example, in Figure 8, “PC”and “Notebook”are the generalized classifications of “IBM TP”while “RAM 256MB”and “IBM 60G”are its compositions. The following rule,
ProdName=“HP DeskJet” ProdName=“IBM TP”,
reveals most customers buying “HP DeskJet”tend to also buy “IBM TP”. According to the
composition relationships in the 3C domain ontology, we know IBM TP is composed of IBM
60GB. Therefore, the following extended rule can be derived:
ProdName=“HP DeskJet” ProdName=“IBM TP”with ProdName=“IBM 60GB”,
showing customers who buy “HP DeskJet“tend to buy “IBM TP”with “IBM 60GB”.
Another motivation of incorporating domain ontology is it helps the user to clarify the scope of the mining target. For example, if a user is only interested in analyzing PC components, he can specify a filtering condition excluding all products not satisfying this characteristic by utilizing the composition relationship expressed in the domain ontology. Precisely, the filtering condition of mining model element, wc, can be defined to have
ProdName in the 3C_Domain_Ontology as follows:
wc = ProdName in 3C_Domain_Ontology (‘PC’, has_component, var_All )
The predicate 3C_Domain_Ontology takes the triple of the subject-relationship-object as the parameters, where var_All is a variable representing all the products that are components of the PC.
3.4 User preference ontology
In a data warehouse mining system, a user manifests his or her mining intention by the mining model settings as defined in Definition 2. As we previously pointed out, a user’s mining model setting is a highly interactive process between the user and the system. Some users might not know exactly what they want or how they can initiate mining models. Therefore, it is important for a system to provide the users with intelligent assistance in setting the mining models closer to their intentions. Our theme toward realizing this assistance is to utilize the
knowledge of experienced mining users in model settings to provide recommendations for questions like how to select the grouping attributes and the interested mining-items, how to judge minimal support and minimal confidence. This is because experienced users have a good sense of the mining processes, so their knowledge is worth sharing with other users.
Specifically, we log the setting history of mining models satisfying the users, and periodically distill and condense them into the structure of the user preference ontology. A detailed description of the distillation process can be found in (Wu et al., 2009). In short, we employ the association rule mining technique over the mining log to find surrogate patterns representative of frequently used queries in the mining history. For example, consider the following rule:
tG{CustID} tM{Gender, Education}.
It indicates if {CustID} is the transaction ID, the pertinent mining attributes always tend to be {Gender, Education}. Similarly, we can discover association rules revealing close relationships between certain (tG, tM) pairs and where conditions wc and/or having conditions
hc, in the form of
tG, tM wc [, hc]
All surrogate rules mined from the system’s mining log are then constructed into the user preference ontology shown in Figure 9, where the attribute indexing level provides connections from each attribute to the corresponding tG and tM for easy and rapid access.
Under a certain tG, the tM with close relationships are grouped into the same node, meaning
the set of tMtends to be used together for a mining. In the same way, the close related filtering
conditions, wc and hc, under certain tM and tG can also be grouped together, indicating the
users usually apply the wc and hc with such tGand tMsettings.
The favorable minimum support and minimum confidence associated with each surrogate pattern are calculated by averaging the ms and mc in the mining models. For example, the following closely related mining models have different minimum supports and minimum confidences:
Model 1: tG{CustID}, tM{ProdName}, ms45%, mc68%
Model 3: tG{CustID}, tM{ProdName}, ms60%, mc62%.
These mining models will be grouped together in the user preference ontology with the suggested minimum support ms59% and the suggested minimum confidence mc 69%. Other statistical summary functions such as mean or mode can be used instead. The user preference ontology can be further connected to the rule base, if any, maintaining the mining results of the mining models to facilitate efficient execution of incremental mining (Chueng et al., 1996) or iterative mining (Liu & Yin, 2001) of the associated mining model. Figure 10 is an example of the user preference ontology.
Figure 9. The user preference ontology
tG1 tG2 … tGn tM1 tM2 … tMj s3, c3 … … a1 a2 a3 … ar
attribute indexing level
s4, c4 … sk, ck
s1, c1 s2, c2
wc1
hc1 hc2
Figure 10. An example of user preference ontology
4. Intelligent Assistance for Mining Model Formulation with Ontologies
Ideally, the user’s mining intention can be reflected in the mining model setting. However, without complete comprehension of the schema and domain related knowledge, the end users may develop mining models based on their experiences or intuition. The mining model formulated by the users can possibly be semantically invalid, leading to incorrect or redundant search space or mining results and wasting the mining efforts they have made. For a common mining model formulation interface, a system often provides as many syntactic error checking mechanisms as possible. For example, it provides a popup list or list box for attribute selection to avoid users’typo. However less effort has been made with semantic checking due to the lack of semantic relationship information beyond the data warehouse. Below, we will elaborate on how an intelligent assistance can be built into the proposed data warehouse mining system under the support of ontologies.
Specifically, we will show through the assistance of the ontologies we have introduced, the mining model formulation interface can provide the semantic error detection and mining model element recommendation in an attempt to improve the effectiveness and the efficiency of mining processes. Users can express their mining intention more precisely and even clarify
CustID Date, Gender, Education ProdName, Salesman
…
60%,90%CustID Date ProdName Gender
attribute indexing level
Educationr
…
CustID Category, City=”Taipei” 60%,85% 45%,80% ProdName, Education…
Rule Base…
…
or renew their original mining intentions. The intelligent checking mechanisms proposed in our system framework are shown in Figure 11.
Element Recommendation Semantic Check User Interface ﴾ ﴿ Ontologies
Figure 11. Intelligent assistance in mining model setting
Intelligent assistance in semantic checking
Through the support of schema ontology and schema constraint ontology, the system can provide semantic checking against mining model elements. Figure 12 shows four different results will be displayed to inform the user of the appropriateness of his mining modeling setting, and provide a rationale if errors occur to help the user reformulate the model. Conforming to the definition of the mining model, our system will check the main elements, including data granularity tG, mining attributes tM, filtering conditions wc and hc.
Figure 12. Semantic checking mechanism
(a) Semantic checking of tG
This checks the semantic legality of the transaction ID for data grouping. The transaction ID set, tG, represents the data granularity and is the key for the mining transactions. If this
question is asked: “What product associations are there between daily customer’s Resulting cases:
Case 1. Pass checking Case 2. Warning message
Case 3. Reject with error message
Case 4. Automatic correction with message Mining Model Setting Semantic Checking Schema Ontology Schema Constraint Ontology
purchase?”, then tG will be {CustID, Date}; if another question is asked: “From daily
product category purchase, are there any associations between customer’s education and gender?”, then tGwill be {Cateogry, Date}. A user can select tG based on their needs or
interests but may also set it incorrectly because of the lack of semantic understanding. Below are some scenarios of incorrect settings.
Example 1: tG{Size}
The system will reject this setting by semantic checking against the constraint
ItemOnly(Size) in the schema ontology.
Example 2: tG{Gender}
The result of the grouping will be only two transactions, which is too few to generate any rules. This will be rejected according to the constraint NoSingleGroup(Gender).
Example 3: tG= {CustID, Gender}
tG1= {CustID, Gender} is actually a redundant form of tG2= {CustID} according to
constraint Decide(CustID, Gender). The mining space for tG1and tG2is exactly the same.
The system will automatically correct the setting of tG2with warning messages.
(b) Semantic checking of tM
This is the semantic legality check of the user’s interested mining items. This checking specifically verifies if tMviolates any of the constraints in the schema constraint ontology.
Example 4: tM= {Year} or tM= {Year, Education}
Tedious rules such as
Year“1999” Year “2000”or
Year“1999”,Education “High School” Year“2002”,Education “Elementary”
will be generated. According to the constraint GroupOnly(Year), the model settings will be rejected.
According to constraint Decide(ProdID, Size), the mining item ProdID determines the value of Size. This setting will generate known rules since it digs the associations between the product name and its size. The following rule is an example:
ProdName“IBM TP” Size “17 inch * 15 inch * 1 inch”.
Example 6: tM= {Gender}
Tedious rules will be generated such as
Gender“Female” Gender “Male”.
The system will reject the senseless setting according to the constraint
NoIntraMining(Gender).
(c) Semantic checking of wc
In the mining model setting, the wc filtering is operated before grouping data into transactions with transaction ID. The checking of wc includes type consistency checking and domain checking.
Example 8: wc(City ’Japan’)
This example has no problem with type consistency but ‘Japan’is actually not a city,
therefore the system will respond with a domain checking warning to the user.
Example 9: wc(ProdName in 3C_DomainOntology (‘All-in-one’, Classification, var_All))
The 3C domain ontology can be used for filtering conditions. If a user is interested in only the “All-in-one”related products in market basket analysis, all the objects with
‘classification’relationship to All-in-one in 3C domain ontology should be retrieved.
The domain ontology retrieved values are then used for selecting related transactions from the data warehouse.
(e) Semantic checking of (tG, hc)
This function checks the semantic legality of aggregation used in the filtering condition hc. Note, in the star schema model, there are three different types of measures, additive,
semi-additive and non-semi-additive, of which the semi-semi-additive measures are defined along some dimensions. For this reason, the checking of hc should be considered in accordance with the grouping ID to avoid invalid aggregation along the wrong dimensions.
Example 10: tG{ProdName, Date}, hc(sum(SaleAmount) > 1000)
In the schema ontology ‘SaleAmount’is an additive measure, therefore the system will
pass the checking.
Example 11: tG{CustID, Date}, hc (sum(Cost) < 100)
The system will reject the setting because ‘Cost’is a semi-additive fact and should be,
as shown in Figure 6, aggregated along with dimensions including Product.
Intelligent assistance in element recommendation
As well as the semantic checking, the system offers recommendations to lead the user, especially the inexperienced one, toward a more efficient mining model formulation process. The functions, taking partial input from the unfinished mining model element created by the user, spontaneously list the recommendations of possible successive mining model constituent drawn from user preference ontology for users to refer to. Based on the example of the user preference ontology in Figure 10, we present some examples as follows:
(a) Recommendation of tGby giving a partial tG
A partial tGis taken as the key to search in the user preference ontology. It can be empty
once the user does not know how to start a model setting. The system will respond with the available tGlist.
Example 11: Given partial tG= {CustID}
The system will prompt with the list {Date},…, {Category} as succeeding tGcandidates
for the user to refer to.
With input of tG and partial tM,, the system will search the available sets of mining
attributes in the user preference ontology for recommendations.
Example 12: Given tG{CustID, Date} and partial tM{ProdName}
The system will prompt the user with the following list: {Education},…,{Salesman} as referable suggestions.
(c) Recommendation of ms and mc by giving tGand tM
Example 13: tG{CustID, Date}, tM{ProdName, Education}
The corresponding ms and mc of the given tGand tMin the user preference ontology will
be listed. In this case, ms60% and mc 85% will be suggested. 5. A system prototype
A prototype implementation of the system is delivered to demonstrate the feasibility of our study. We use Borland C++ Builder to develop this system and the data warehouse is stored in the SQL Server 2005. A data mining job involves the following steps:
(1) The user finishes a mining model setting through the intelligent user interfaces.
(2) Data is selected from data warehouse, data mart or data cube according to the mining model setting and is transformed to target data in the format shown in Table 1.
(3) Target data is fed into mining engine for frequent itemsets and rule generation.
For step 1, we designed a step by step wizard to guide the users toward correct mining model setting, especially for those who are not familiar with data mining. Figure 13 to Figure 17 show some examples of the user interfaces. The dimension tree is a treeview box showing the dimensions and the attributes of the selected data warehouse. The user specifies the transaction ID or pertinent mining attributes by drag-and-drop. If the settings have ever been improper or incorrect, the system will prompt the user with warnings or error messages as shown in Figure 13 to remind the users. The system also provides recommendations of transaction ID or mining attributes, as shown in Figure 14 and Figure 15. After the steps of transaction ID and mining attribute settings, optional filtering conditions before transaction grouping can be set as shown in Figure 18, if necessary. In Figure 16, the system provides
drag-and-drop of dimension attributes to the attribute edit box. List boxes for the relational operator and the attribute value settings are also provided for users to perform data selection. This decreases the user’s manual typo as much as possible. The system also shows the filtering instances from the user preference ontology, serving to support users who have little sense of setting filtering conditions. The optional filtering condition after transaction grouping is also provided in a similar way. The last step is to set the pertinence threshold, as shown in Figure 17. Users set the minimum support and minimum confidence with the suggestion values provided by the system.
The target data selection and grouping is manipulated by customized functions we have developed. In step 3, for the processing of frequent itemset generation and rule derivation, the domain ontology is utilized to derive extended concepts of rules. We use the AROC algorithm in (Tseng et al., 2007) to generate frequent itemsets and rules. It integrates the classification and composition relationships in the domain ontology to extend the implications of the rules.
Figure 14. An example of recommendations for transaction ID setting
Figure 15. An example of recommendations for mining attribute setting
Figure 17. An example of a pertinence threshold setting
6. Related work
This section gives an overview of the literature related to our work in twofold: the use of ontology in data mining and data warehouse mining.
(1) Use of ontology in data mining
If the concept hierarchy or taxonomy can be viewed as an ontology, then the use of ontologies in data mining can be traced back to 1991 when Nunez used information of the classification hierarchy and attribute processing cost to improve the efficiency of the classification process(Nunez, 1991). Later, Han & Fu (1995) and Srikant & Agrawal (1995) also proposed combining classification hierarchies to mine multilevel association rules and generalized association rules, respectively. Their works were later extended by Chien et al. (2007), who not only applied classification but also composition hierarchical knowledge to mining fuzzy association rules. These researches, however, concentrated on the design of the algorithms, yet discussion of ontology structure design and its benefit to data mining were not covered. Until recently, research on applying ontology to data mining was exploited by several studies such as, ontology-based induction of rules (Aronis et al., 1996; Taylor et al., 1997), based business understanding (Sharma & Osei-Bryson, 2009), ontology-based post-processing and explanation of association rules (Domingues & Rezende, 2005; Liao et al., 2009; Marinica et al., 2008; Svatek et al., 2005), ontology-supported selection of classification algorithms (Bernstein et al., 2005; Lin et al., 2006), ontology-guided new
attributes generation from databases (Phillips & Buchanan, 2001), and ontology-based integration and preprocessing of data (Euler & Scholz, 2004; Perez-Rey et al., 2006).
Differing from the above work on dealing with the issue of incorporating ontology in the individual phase of the well known KDD process proposed by Fayyad et al. (1996), there has been work conducted from an integral perspective. For example, Kopanas et al. (2002) pointed out the essence of incorporating ontology (the term domain knowledge is used instead) to the KDD process and demonstrated their viewpoints using a telecommunication customer insolvency case study.Cespivova et al. (2004) conducted a systematic study by discussing the roles of medical domain ontology in each aspect of the KDD process. A similar study was also presented in (Gottgtroy et al., 2004; Kuo et al., 2007). A position paper presented by Charlest et al. (2006) discussed the synergy of combining case based reasoning and ontology in the context of data mining assistance framework, though the issue of realization and implementation was left aside. In 2006, Pan & Pan proposed an ontology supporting data mining from databases. They maintained previous mining results in ontology that can further be applied for incremental association rule mining.
(2) Data warehouse mining
Currently, the research on data warehouse mining is mostly concentrated on data mining from data cubes or multi-dimensional databases. J. Han’s research group pioneered this research subject (Han, 1998; Han et al., 1999). The study conducted by Ester and his colleagues (Ester et al., 1998; Ester & Wittmann, 1998) instead considered the problem of incrementally updating mined patterns from data warehouses. In 2000, Psaila and Lanzi studied multi-level association mining from a primitive data warehouse and proposed a mining algorithm. Since then, substantial works have been devoted to discovering multidimensional association rules from data warehouses (Ng et al., 2002; Chung & Mangamuri, 2005; Tjioe & Taniar, 2005; Messaoud et al., 2006; Yang et al., 2008).
The research by Priebe & Pernul (2003) first exploited the issues of incorporating ontology into knowledge discovery from data warehouses. In particular, it proposed an intelligent web portal integrating OLAP and information retrieval through ontology, yet it focused on information retrieval issues but not on data mining. Subsequent work on multiple source integration for data warehouse OLAP construction includes Niemi et al. (2007) and Shah et al. (2009). In (Wu et al., 2007), we presented the problems with contemporary association rule mining in data warehousing systems, explained the essence that incorporates
ontologies to resolve the problems, anddemonstrated a preliminary framework.
7. Conclusions
The purpose of data mining is for users to find real and useful knowledge they actually want. In this paper we have shown a data warehouse mining system framework with intelligent assistance incorporating schema ontology, schema constraint ontology, domain ontology and user preference ontology. We have demonstrated the intelligent assistance provided by the mining system in guiding users through the mining processes. This improves the mining effectiveness and efficiency in four aspects as follows. First, the processes of the mining model settings are assisted by intelligent functions, minimizing the possibilities of illegal settings of mining models. Also, appropriate recommendations of the mining model elements are provided while the users are setting the mining model. This avoids execution of ineffective or redundant mining processes and also guides the users through the approaching of the mining models that are closer to their mining intention. Second, with the support of domain ontology, mining rules can be extended and generalized. Third, the information in the domain ontology can be included in the filtering condition to obtain a more specific search space. More precise knowledge can be discovered. Fourth, it provides the system with knowledge browsing capability that a mining model can be examined against the user preference ontology for any duplication or similarities. This saves the system’s resources. In this paper, we have discussed the intelligent assistance in general. A preliminary implementation of this system framework has also been provided to demonstrate the claimed benefits.
The ontologies we have proposed in this paper are implemented in relational table structures. Nevertheless, these ontologies are local to the specific mining system we have proposed. Making them globally sharable is challenging and is an important future work.
Acknowledgements
This work was supported by the National Science Council of R.O.C. under grant NSC 95-2221-E-390-024.
References
Aronis, J.M., Provost, F.J., & Buchanan, B.G. (1996). Exploiting background knowledge in automated discovery. In Proceedings of the 2nd International Conference on Knowledge
Discovery and Data Mining (pp. 355–358).
Bernstein, A., Provost, F., & Hill, S. (2005). Toward intelligent assistance for a data mining process: an ontology-based approach for cost-sensitive classification. IEEE Transactions
on Knowledge and Data Engineering 17(4), 503–518.
Cespivova, H., Rauch, J., Svatek, V., Kejkula, M., & Tomeckova, M. (2004). Roles of medical ontology in association rule mining Crisp-Dm cycle. In Proceedings of
ECML/PKDD Workshop on Knowledge Discovery and Ontologies.
Charest, M., Delisle, S., Cervantes, O., & Shen, Y. (2006). Intelligent data mining assistance via CBR and ontologies. In Proceedings of the 17th International Conference on
Database and Expert Systems Applications (pp. 593–597).
Chaudhuri, S., & Dayal, U. (1997). An overview of data warehouse and OLAP technology.
ACM SIGMOD Record 26(1), 65-74.
Chien, B.C., Zhong, M.H., & Wang, J.J. (2007). Mining fuzzy association rules on has-a and is-a hierarchical structures. International Journal of Advanced Computational Intelligence and Intelligent Informatics 11(4), 423–432.
Cheung, D.W., Han, J., Ng, V.T., & Wong, C.Y. (1996). Maintenance of discovered association rules in large databases: An incremental update technique. In Proceedings of
the 12th International Conference on Data Engineering (pp. 106–114).
Chung, S.M., & Mangamuri, M. (2005). Mining association rules from the star schema on a parallel NCR teradata database system. In Proceedings of International Conference on
Information Technology: Coding and Computing (pp. 206–212).
Domingues, M.A., & Rezende, S.O. (2005). Using taxonomies to facilitate the analysis of the association rules. In Proceedings of the 2nd International Workshop on Knowledge
Discovery and Ontologies.
van Elst, L., & Abecker, A. (2002). Ontologies for information management: balancing formality, stability, and sharing scope. Expert Systems with Applications 23(4), 357–366. Ester, M., Kriegel, H.P., Sander, J., Wimmer, M., & Xu, X. (1998). Incremental clustering
for mining in a data warehousing environment. In Proceedings of 24th International
Ester, M., & Wittmann R. (1998). Incremental generalization for mining in a data warehousing environment. In Proceedings of the 6th International Conference on
Extending Database Technology: Advances in Database Technology (pp. 135–149) Euler, T. & M. Scholz. (2004). Using ontologies in a KDD workbench. In Proceedings of
ECML/PKDD Workshop on Knowledge Discovery and Ontologies.
Fayyad U., Piatetsky-Shapiro G., and Smyth P. (1996). The KDD process for extracting useful knowledge from volumes of data. Communications of the ACM 39(11), 27–34. Gottgtroy, P., Kasabov, N., & MacDonell, S. (2004). An ontology driven approach for
knowledge discovery in biomedicine. In Proceedings of the 8th Pacific Rim International
Conference on Artificial Intelligence.
Han, J. (1998). Toward on-line analytical mining in large databases. ACM SIGMOD Record 27(1), 97–107.
Han, J., Chiang, J.Y., Chee, S., et al. (1997). DBMiner: A system for data mining in relational databases and data warehouses. In Proceedings of the 1997 Conference of the Centre for
Advanced Studies on Collaborative Research (pp. 250–255).
Han, J., & Fu, Y. (1995). Discovery of multiple-level association rules from large databases. In Proceedings of the 21st Very Large Databases Conference (pp. 420–431).
Han J., & Kamber, M. (2001). Data Mining: Concepts and Techniques, Morgan Kaufmann. Han, J., Lakshmanan, L.V.S., & Ng, R.T. (1999). Constraint-based, multi-dimensional data
mining. IEEE Computer 32(8), 46–50.
Heflin, J., Editor (2004). OWL Web Ontology Language Use Cases and Requirements, W3C Recommendation, 10 February 2004, http://www.w3.org/TR/2004/REC-webont-req-20040210/
Inmon, W.H. (1995). Building the Data Warehouse, John Wiley & Sons, Inc., New York, NY. Kimball, R. (1996). The Data Warehouse Toolkit Practical For Building Dimensional Data
Warehouses, John Wiley & Sons, Inc.
Kopanas, I., Nikolaos N., Avouris, M., & Daskalaki, S. (2002). The role of domain knowledge in a large scale data mining project. In Proceedings of the 2nd Hellenic
Conference on AI:Methods and Applications of Artificial Intelligence (pp. 288–299).
Kuo, Y.T., Lonie, A., & Sonenberg, L. (2007). Domain ontology driven data mining: A medical case study. In Proceedings of ACM SIGKDD Workshop on Domain Driven Data
Liao, S.H., Ho, H.H., & Yang, F.C. (2009). Ontology-based data mining approach implemented on exploring product and brand spectrum. Expert Systems with Applications 36(9), 11730–11744.
Lin, M.S., Zhang, H., & Yu, Z.G. (2006). An ontology for supporting data mining process. In
Proceedings of IMACS Multiconference on Computational Engineering in Systems Applications (pp. 2074–2077).
Lisi, F.A., & Malerba, D. (2004). Inducing multi-level association rules from multiple relations. Machine Learning 55(2), 175–210.
Liu, J., & Yin, J. (2001). Towards efficient data re-mining (DRM). In Proceedings of the 5th
Pacific-Asia Conference on Knowledge Discovery and Data Mining, Lecture Notes in Computer Science 2035 (pp. 406–412).
Marinica, C., Guillet, F., & Briand, H. (2008). Post-processing of discovered association rules using ontologies. In Proceedings of IEEE International Conference on Data Mining
Workshops (pp. 126–133).
Messaoud, R.B., Rabaséda, S.L., Boussaid, O., & Missaoui, R. (2006). Enhanced mining of association rules from data cubes. In Proceedings of the 9th ACM International
Workshop on Data Warehousing and OLAP (pp. 11–18).
Ng, E.K.K., Ng, K., Fu, A.W.C., & Wang, K. (2002). Mining association rules from stars. In
Proceedings of the 2002 IEEE International Conference on Data Mining (pp. 322–329) Niemi, T. Toivonen, S., Niinimaki, M., & Nummenmaa, J. (2007). Ontologies with semantic
web/grid in data integration for OLAP. International Journal on Semantic Web and
Information Systems 3(4), 25–49.
Nunez, M. (1991). The use of background knowledge in decision tree induction. Machine
Learning 6(3), 231–250.
Pan, D., & Pan, Y. (2006). Using ontology repository to support data mining. In Proceedings
of the 6th World Congress on Intelligent Control and Automation (pp. 5947–5951). Perez-Rey, D., Anguita, A., & Crespo J. (2006). OntoDataClean: Ontology-based integration
and preprocessing of distributed data. Lecture Notes in Computer Science4345,262–272. Perng, C.S., Wang, H., Ma, S., & Hellerstein, J.L. (2001). Farm: A framework for exploring
mining spaces with multiple attributes. In Proceedings of the 1st IEEE International
Perng, C.S., Wang, H., Ma, S., & Hellerstein, J.L. (2002). User-directed exploration of mining space with multiple attributes. In Proceedings of the 2nd IEEE International
Conference on Data Mining (pp. 394–401).
Phillips, J., & Buchanan, B.G. (2001). Ontology-guided knowledge discovery in databases. In
Proceedings of the 1st International Conference on Knowledge Capture (pp. 1230–130). Priebe, T., & Pernul, G. (2003). Ontology-based integration of OLAP and information
retrieval. In Proceedings of the 14th International Workshop on Database and Expert
Systems Applications (pp. 610–614).
Psaila, G., & Lanzi, P.L. (2000). Hierarchy-based mining of association rules in data warehouses. In Proceedings of ACM Symposium on Applied Computing (pp. 307–312). Sharma, S., & Osei-Bryson, K.M. (2009). Framework for formal implementation of the
business understanding phase of data mining projects. Expert Systems with Applications 36(2), 4114–4124.
Shah, N., Tsai, C.F., Marinov, M., Cooper, J., Vitliemov, P., & Chao, K.M. (2009). Ontological on-line analytical processing for integrating energy sensor data, IETE
Technical Review 26(5), 375–387.
Srikant, R., & Agrawal, R. (1995). Mining generalized association rules. In Proceedings of
the 21st Very Large Data Bases Conference (pp. 407–419).
Svatek, V., Rauch, J., & Flek, M. (2005). Ontology-based explanation of discovered associations in the domain of social reality. In Proceedings of the 2nd International
Workshop on Knowledge Discovery and Ontologies, 2005.
Taylor, M., Stoffel, K., & Hendler, J. (1997). Ontology-based induction of high level classification rules. In Proceedings of SIGMOD Data Mining and Knowledge Discovery
Workshop.
Tjioe, H.C., & Taniar, D. (2005). Mining Association Rules in Data Warehouses.
International Journal of Data Warehousing and Mining 1(3), 28–62.
Tseng, M.C., Lin, W.Y., & Jeng, R. (2007). Mining association rules with ontological information. In Proceedings of the 2nd International Conference on Innovative
Computing, Information and Control (pp. 300–303).
Uschold, M., & Gruninger, M. (1996). Ontologies: principles, methods and applications.
Wu, C.A., Lin, W.Y., Tseng, M.C., & Wu, C.C. (2007). Ontology-incorporated mining of association rules in data warehouse. Journal of Internet Technology 8(4), 477–485. Wu, C.A., Lin, W.Y., Jiang, C.L., & Wu, C.C. (2009). Favorable support threshold
recommendation for multidimensional association mining using user preference ontology. In Proceedings of 2009 IEEE International Conference on Granular Computing (pp. 586–591).
Yang, W., Li, Y., Wu, J., & Xu, Y. (2008). Granule mining oriented data warehousing model for representations of multidimensional association rules. International Journal of
Intelligent Information and Database Systems 2(1), 125–145
Zhu, H. (1998) On-Line Analytical Mining of Association Rules. Master’s Thesis, Simon Fraser University, U.S.A.
