An Empirical Study of Category Skew on Feature Selection for Text Categorization.
Abstract. In this paper, we present an empirical comparison of the effects of category skew on six feature selection methods. The methods were evaluated on 36 datasets generated from the 20 Newsgroups, OHSUMED, and Reuters-21578 text corpora. The datasets were generated to possess particular category skew characteristics (i.e., the number of documents assigned to each category). Our objective was to determine the best performance of the six feature selection methods, as measured by F-measure and Precision, regardless of the number of features needed to produce the best performance. We found the highest F-measure values were obtained by bi-normal separation and information gain and the highest Precision values were obtained by categorical proportional difference and chi-squared. 1
Categorical Proportional Difference A Feature Selection Method for Text Categorization.
Supervised text categorization is a machine learning task where a predefined category label is automatically assigned to a previously unlabelled document based upon characteristics of the words contained in the document. Since the number of unique words in a learning task (i.e., the number of features) can be very large, the efficiency and accuracy of the learning task can be increased by using feature selection methods to extract from a document a subset of the features that are considered most relevant. In this paper, we introduce a new feature selection method called categorical proportional difference (CPD), a measure of the degree to which a word contributes to differentiating a particular category from other categories. The CPD for a word in a particular category in a text corpus is a ratio that considers the number of documents of a category in which the word occurs and the number of documents from other categories in which the word also occurs. We conducted a series of experiments to evaluate CPD when used in conjunction with SVM and Naive Bayes text classifiers on the OHSUMED, 20 Newsgroups, and Reuters-21578 text corpora. Recall, precision, and the F-measure were used as the measures of performance. The results obtained using CPD were compared to those obtained using six common feature selection methods found in the literature: χ 2, information gain, document frequency, mutual information, odds ratio, and simplified χ 2. Empirical results showed that, in general, according to the F-measure, CPD outperforms the other feature selection methods in four out of six text categorization tasks.
Pattern Classification in No%2DLimit Poker A Head%2DStart Evolutionary Approach.

Evaluation of a Dominance%2DBased Rough Set Approach to Interface Design.

Exploratory Quantitative Contrast Set Mining A Discretization Approach.

Mining Associations for Interface Design.
Abstract. Consumer research has indicated that consumers use compensatory and non-compensatory decision strategies when formulating their purchasing decisions. Compensatory decision-making strategies are used when the consumer fully rationalizes their decision outcome whereas non-compensatory decision-making strategies are used when the consumer considers only that information which has most meaning to them at the time of decision. When designing online shopping support tools, incorporating these decision-making strategies with the goal of personalizing the design of the user interface may enhance the overall quality and satisfaction of the consumer’s shopping experiences. This paper presents work towards this goal. The authors describe research that refines a previously developed procedure, using techniques in cluster analysis and rough sets, to obtain consumer information needed in support of designing customizable and personalized user interface enhancements. The authors further refine their procedure by examining and evaluating techniques in traditional association mining, specifically conducting experimentation using the Eclat algorithm for use with the authors ’ previous work. A summary discussing previous work in relation to the new evaluation is provided. Results are analyzed and opportunities for future work are described.
Statistical Methodologies for Mining Potentially Interesting Contrast Sets.

Evaluating WordNet Features in Text Classification Models.

A Characterization of Wordnet Features in Boolean Models For Text Classification.

Evaluating WordNet Features in Text Classification Models.

Predicting Itemset Sales Profiles with Share Measures and Repeat%2DBuying Theory.
Given a random sample of sales transaction records (i.e.,  scanner panels) for a particular period (such as a week, month, quarter,  etc.), we analyze the scanner panels to determine approximations for the  penetration and purchase frequency distribution of frequently purchased  items and itemsets. If the purchase frequency distribution for an item  or itemset in the current period can be modeled by the negative binomial   distribution, then the parameters of the model are used to predict  sales profiles for the next period. We present representative experimental  results based upon synthetic data.
Measuring the interestingness of discovered knowledge A principled approach.

Evaluation of Interestingness Measures for Ranking Discovered Knowledge.
When mining a large database, the number of patterns discovered  can easily exceed the capabilities of a human user to identify interesting  results. To address this problem, various techniques have been  suggested to reduce and/or order the patterns prior to presenting them  to the user. In this paper, our focus is on ranking summaries generated  from a single dataset, where attributes can be generalized in many different  ways and to many levels of granularity according to taxonomic  hierarchies. We theoretically and empirically evaluate thirteen diversity  measures used as heuristic measures of interestingness for ranking summaries  generated from databases. The thirteen diversity measures have  previously been utilized in various disciplines, such as information theory,  statistics, ecology, and economics. We describe five principles that  any measure must satisfy to be considered useful for ranking summaries.  Theoretical results show that only four of the thirteen diversity measures  satisfy all of the principles. We then analyze the distribution of the index  values generated by each of the thirteen diversity measures. Empirical results,  obtained using synthetic data, show that the distribution of index  values generated tend to be highly skewed about the mean, median, and  middle index values. The objective of this work is to gain some insight  into the behaviour that can be expected from each of the measures in  practice.  1 
The Lorenz Dominance Order as a Measure of Interestingness in KDD.
Abstract. Ranking summaries generated from databases is useful within the context of descriptive data mining tasks where a single data set can be generalized in many different ways and to many levels of granularity. Our approach to generating summaries is based upon a data structure, associated with an attribute, called a domain generalization graph (DGG). A DGG for an attribute is a directed graph where each node represents a domain of values created by partitioning the original domain for the attribute, and each edge represents a generalization relation between these domains. Given a set of DGGs associated with a set of attributes, a generalization space can be defined as all possible combinations of domains, where one domain is selected from each DGG for each combination. This generalization space describes, then, all possible summaries consistent with the DGGs that can be generated from the selected attributes. When the number of attributes to be generalized is large or the DGGs associated with the attributes are complex, the generalization space can be very large, resulting in the generation of many summaries. The number of summaries can easily exceed the capabilities of a domain expert to identify interesting results. In this paper, we show that the Lorenz dominance order can be used to rank the summaries prior to presentation to the domain expert. The Lorenz dominance order defines a partial order on the summaries, in most cases, and in some cases, defines a total order. The rank order of the summaries represents an objective evaluation of their relative interestingness and provides the domain expert with a starting point for further subjective evaluation of the summaries. 1
Evaluation of Interestingness Measures for Ranking Discovered Knowledge.
When mining a large database, the number of patterns discovered  can easily exceed the capabilities of a human user to identify interesting  results. To address this problem, various techniques have been  suggested to reduce and/or order the patterns prior to presenting them  to the user. In this paper, our focus is on ranking summaries generated  from a single dataset, where attributes can be generalized in many different  ways and to many levels of granularity according to taxonomic  hierarchies. We theoretically and empirically evaluate thirteen diversity  measures used as heuristic measures of interestingness for ranking summaries  generated from databases. The thirteen diversity measures have  previously been utilized in various disciplines, such as information theory,  statistics, ecology, and economics. We describe five principles that  any measure must satisfy to be considered useful for ranking summaries.  Theoretical results show that only four of the thirteen diversity measures  satisfy all of the principles. We then analyze the distribution of the index  values generated by each of the thirteen diversity measures. Empirical results,  obtained using synthetic data, show that the distribution of index  values generated tend to be highly skewed about the mean, median, and  middle index values. The objective of this work is to gain some insight  into the behaviour that can be expected from each of the measures in  practice.  1 
Principles for mining summaries using objective measures of interestingness.
An important problem in the area of data mining is the development of effective measures of interestingness for ranking discovered knowledge. In this paper, we propose five principles that any measure must satisfy to be considered useful for ranking the interestingness of summaries generated from databases. We investigate the problem within the context of summarizing a single dataset which can be generalized in many different ways and to many levels of granularity. We perform a comparative sensitivity analysis of fifteen well-known diversity measures to identify those which satisfy the proposed principles. The fifteen diversity measures have previously been utilized in various disciplines, such as information theory, statistics, ecology, and economics. Their use as objective measures of interestingness for ranking summaries generated from databases is novel. The objective of this work is to gain some insight into the behaviour that can be expected from each of the diversity measures in practice, and to begin to develop a theory of interestingness against which the utility of new measures can be assessed.  1 
Applying Objective Interestingness Measures in Data Mining Systems.
One of the most important steps in any knowledge discovery  task is the interpretation and evaluation of discovered patterns. To  address this problem, various techniques, such as the chi-square test for  independence, have been suggested to reduce the number of patterns  presented to the user and to focus attention on those that are truly statistically  significant. However, when mining a large database, the number  of patterns discovered can remain large even after adjusting significance  thresholds to eliminate spurious patterns. What is needed, then, is an  effective measure to further assist in the interpretation and evaluation  step that ranks the interestingness of the remaining patterns prior to  presenting them to the user. In this paper, we describe a two-step process  for ranking the interestingness of discovered patterns that utilizes  the chi-square test for independence in the first step and objective measures  of interestingness in the second step. We show how this two-step  process can be applied to ranking characterized/generalized association  rules and data cubes.
Ranking the Interestingness of Summaries from Data Mining Systems.
We study data mining where the task is description  by summarization, the representation language is generalized  relations, the evaluation criteria are based on  heuristic measures of interestingness, and the method  for searching is the Multi-Attribute Generalization algorithm  for domain generalization graphs. We present  and empirically compare four heuristics for ranking the  interestingness of generalized relations (or summaries).  The measures are based on common measures of the diversity  of a population, statistical variance, the Simpson  index, and the Shannon index. All four measures  rank less complex summaries (i.e., those with few tuples  and/or non-ANY attributes) as most interesting.  Highly ranked summaries provide a reasonable starting  point for further analysis of discovered knowledge.  
Heuristic for Ranking the Interestigness of Discovered Knowledge.

Heuristic Measures of Interestingness.
The tuples in a generalized relation (i.e., a summary generated  from a database) are unique, and therefore, can be considered to be  a population with a structure that can be described by some probability  distribution. In this paper, we present and empirically compare sixteen  heuristic measures that evaluate the structure of a summary to assign  a single real-valued index that represents its interestingness relative to  other summaries generated from the same database. The heuristics are  based upon well-known measures of diversity, dispersion, dominance,  and inequality used in several areas of the physical, social, ecological,  management, information, and computer sciences. Their use for ranking  summaries generated from databases is a new application area. All sixteen  heuristics rank less complex summaries (i.e., those with few tuples  and/or few non-ANY attributes) as most interesting. We demonstrate  that for sample data sets, the order in which some of the measures rank  summaries is highly correlated.
Temporal Generalization with Domain Generalization Graphs.
This paper addresses the problem of using domain generalization graphs to generalize temporal data extracted from relational databases. A domain generalization graph associated with an attribute defines a partial order which represents a set of generalization relations for the attribute. We propose formal specifications for domain generalization graphs associated with calendar (date and time) attributes. These graphs are reusable (i.e., can be used to generalize any calendar attributes), adaptable (i.e., can be extended or restricted as appropriate for particular applications), and transportable (i.e., can be used with any database containing a calendar attribute).
Data Mining in Large Databases Using Domain Generalization Graphs.
Attribute-oriented generalization summarizes the information in a relational database by repeatedly replacing specific attribute values with more general concepts according to userdefined concept hierarchies. We introduce domain generalization graphs for controlling the generalization of a set of attributes and show how they are constructed. We then present serial and parallel versions of the Multi-Attribute Generalization algorithm for traversing the generalization state space described by joining the domain generalization graphs for multiple attributes. Based upon a generate-and-test approach, the algorithm generates all possible summaries consistent with the domain generalization graphs. Our experimental results show that significant speedups are possible by partitioning path combinations from the DGGs across multiple processors. We also rank the interestingness of the resulting summaries using measures based upon variance and relative entropy. Our experimental results also show that these measures provide an effective basis for analyzing summary data generated from relational databases. Variance appears more useful because it tends to rank the less complex summaries (i.e., those with few attributes and/or tuples) as more interesting.
A Technique for Generalizing Temporal Durations in Relational Databases.
We address the problem of generalizing temporal data  concerning durations extracted from relational databases.  Our approach is based on a domain generalization  graph that defines a partial order specifying the  generalization relations for a duration attribute. This  domain generalization graph is reusable (i.e., can be  used to generalize any duration attribute), adaptable  (i.e., can be extended or restricted as appropriate for  particular applications), and transportable (i.e., can  be used with any database containing a duration attribute)  .  
Mining Market Basket Data Using Share Measures and Characterized Itemsets.
We propose the share-confidence framework for knowledge  discovery from databases which addresses the problem of mining itemsets  from market basket data. Our goal is two-fold: (1) to present new  itemset measures which are practical and useful alternatives to the commonly  used support measure; (2) to not only discover the buying patterns  of customers, but also to discover customer profiles by partitioning customers  into distinct classes. We present a new algorithm for classifying  itemsets based upon characteristic attributes extracted from census or  lifestyle data. Our algorithm combines the Apriori algorithm for discovering  association rules between items in large databases, and the AOG  algorithm for attribute-oriented generalization in large databases. We  suggest how characterized itemsets can be generalized according to concept  hierarchies associated with the characteristic attributes. Finally, we  present experimental results that demonstrate the utility of the shareconfidence  framework.  1 
Generalization Lattices.
Generalization lattices encode domain knowledge relevant to  generalization. They provide a convenient framework for data visualization  during user-guided exploration and for automated guidance during  independent exploration. To reduce the size of a generalization lattice  for an individual attribute, we define six types of pruning. Then we consider  the generalization space defined by the cross product of lattices  for several attributes. To increase the relevance of the data exploration  results, we define five additional types of pruning. An interactive, webbased  system for visualizing the generalization space allows the user to  interactively guide the data exploration process.
Generalization for Calendar Attributes using Domain Generalization Graphs.
This paper addresses the problem of generalizing temporal data based on calendar (date and time) attributes. The proposed method is based on a domain generalization graph, i.e., a lattice defining a partial order that represents a set of generalization relations for the attribute. We specify the components of a domain generalization graph suited to calendar attributes. We define granularity, subset, lookup, and algorithmic methods for specifying generalizations between calendar domains. To reduce the size of the domain generalization graph used in generalization and the number of results shown to the user, we use six types of pruning: reachability pruning, preliminary manual pruning, data range pruning, previous discard pruning, pregeneralization manual pruning, and post generalization pruning. 1 
Mining Association Rules from Market Basket Data using Share Measures and Characterized Itemsets.
The problem of mining association rules from market basket data has recently been an important research topic in the area of knowledge discovery from databases. It was originally introduced in [2] and studied extensively in [1, 5, 25, 26, 31, 19, 23, 29, 30, 3, 4, 33, 14]. The problem is typically examined in the context of discovering buying patterns from retail sales transactions. Although there are many similar data mining applications which can be modelled in this way, we again study the problem using the retail store example because of its intuitive nature and clarity. Consider a retail sales operation with a large inventory consisting of many different products. The operation is situated in a location where the customer base is socio-economically diverse, with annual household incomes ranging from very low to very high, and demographically ranging from young families to the elderly. The sales manager has used data mining to search for association ru...
Data Visualization in the DB%2DDiscover System.
The DB-Discover system is a research software tool for knowledge discovery from databases. Utilizing an efficient attribute-oriented induction algorithm based upon the climbing tree and dropping condition methods, called multi-attribute generalization, it generates many possible summaries of data from a database. In this paper, we present the design of the data visualization capabilities of DB-Discover. We describe how our data visualization techniques help us manage the many possible summaries and how potentially interesting summaries are selected from them. A prototype system has been implemented which allows many alternative visualizations of the summaries. The results of our work enable a domain expert to quickly and efficiently analyze the contents of a database from many different perspectives.  1 
Parallel Knowledge Discovery Using Domain Generalization Graphs.
Multi-Attribute Generalization is an algorithm for attributeoriented  induction in relational databases using domain generalization  graphs. Each node in a domain generalization graph represents a different  way of summarizing the domain values associated with an attribute.
A Note on Regeneration with Virtual Copies.

Attribute%2Doriented Induction Using Domain Generalization Graphs.
Attribute-oriented induction summarizes the information in a relational database by repeatedly replacing specific attribute values with more general concepts according to user-defined concept hierarchies. We show how domain generalization graphs can be constructed from multiple concept hierarchies associated with an attribute, describe how these graphs can be used to control the generalization of a set of attributes, and present the Multi-Attribute Generalization algorithm for attributeoriented induction using domain generalization graphs. Based upon a generate-and-test approach, the algorithm generates all possible combinations of nodes from the domain generalization graphs associated with the individual attributes, to produce all possible generalized relations for the set of attributes. We rank the interestingness of the resulting generalized relations using measures based upon relative entropy and variance. Our experiments show that these measures provide a basis for analyzing summary data from relational databases. Variance appears more useful because it tends to rank the less complex generalized relations (i.e., those with few attributes and/or few tuples) as more interesting.  1 
Performance Analysis of a Regeneration%2DBased Dynamic Voting Algorithm.

