Increasing the enrollment of women, minority, and other underrepresented populations in undergraduate information systems and computing programs is an important social issue. We explore ways of attracting and retaining community college transfer students—an important source of underrepresented students —by examining their sensemaking efforts as they transition to four-year universities. We conducted a qualitative study to test sensemaking theory and develop recommendations for retaining community college transfer students in undergraduate information systems, computer science, and computer engineering programs.

Mississippi State University (MSU), like many other institutions across the country, has seen a significant decline in the number of computing majors since the early 2000's when the Dot Com crash caused many students to shy away from majors involving computing. In addition, the diversity of the students who have remained in the field has decreased in two ways. First, female students make up a smaller percentage of majors in these fields. Second, the percentage of underrepresented minorities has decreased significantly. Therefore, MSU's computing departments (Management and Information Systems, Computer Science and Engineering, and Electrical and Computer Engineering) combined forces in 2010 to propose an NSF funded Broadening Participation in Computing program that would work with the state's large community college system to recruit more students into these computing related majors at MSU. This program received funding in the spring of 2010 and has run two successful Summer Computing Experience programs for this targeted group. This paper presents an overview of the program components and discusses the assessment results so far for the program.

The wide spread use of ontologies in many scientific areas creates a wealth of ontology-annotated data and necessitates the development of ontology-based data mining algorithms. We have developed generalization and mining algorithms for discovering cross-ontology relationships via ontology-based data mining from data represented using multiple ontologies.We present new interestingness measures to evaluate the discovered cross-ontology relationships. The methods presented in this dissertation employ generalization as an ontology traversal technique for the discovery of interesting and informative relationships at multiple levels of abstraction between concepts from different ontologies. The generalization algorithms combine ontological annotations with the structure and semantics of the ontologies themselves to discover interesting cross-ontology relationships.The first algorithm uses the depth of ontological concepts as a guide for generalization. The ontology annotations are translated to higher levels of abstraction one level at a time accompanied by incremental association rule mining. The second algorithm conducts a generalization of ontology terms to all their ancestors via transitive ontology relations and then mines cross-ontology multi-level association rules from the generalized transactions.Our interestingness measures use implicit knowledge conveyed by the relation semantics of the ontologies to capture the usefulness of cross-ontology relationships. We describe the use of information theoretic metrics to capture the interestingness of cross-ontology relationships and the specificity of ontology terms with respect to an annotation dataset.Our generalization and data mining algorithms are applied to the Gene Ontology and the post-natal Mouse Anatomy Ontology. The results presented in this work demonstrate that our generalization algorithms and interestingness measures discover more interesting and better quality relationships.

The notion of cluster ensembles has been accepted in recent years as an effective alternative to individual clustering. Ensemble based clustering methods have typically focussed on combining different clustering results from a single data set and have been limited to partitional clustering only. In this paper, we present an effective ensemble algorithm for combining the results of hierarchical clustering of multiple datasets. We use a graph theoretic approach to combine multiple cluster hierarchies into a single set of partitional clusters. A graph is generated from the cluster hierarchies based on the association strengths of objects in the hierarchies. A graph partitioning algorithm is then applied to generate partitional clusters. Our algorithm can handle multiple contextually related heterogeneous datasets that use different feature sets, but consist of non-disjoint sets of objects. We used our algorithm to solve a document clustering problem and experimental results demonstrate the effectiveness of our algorithm.

In the fall of 2009, three women faculty members in the Bagley College of Engineering and the College of Business at Mississippi State University initiated a leadership development program for female faculty members in those colleges. This paper provides a summary of the first two years of that program. In this paper we describe the motivation for establishing this program, the environment in which the program was conducted, and the discussion topics for the sessions that were held. We also provide feedback from the participants who responded to a survey about the program. We conclude with some general observations about the program and plans for its continuation.

An important problem faced by many Computer Science and Software Engineering programs is declining enrollment. In an effort to reverse that trend at Mississippi State University, we have instituted pair programming for the laboratory exercises in the introductory programming course. This paper describes a study performed to analyze whether using pair programming would increase retention. An important goal of this study was not only to measure increased retention, but to provide insight into why retention increased or decreased. The results of the study showed that retention significantly increased for those students already majoring in Computer Science, Software Engineering, or Computer Engineering. In addition, survey results indicated that the students viewed many aspects of pair programming to be very beneficial to their learning experience.

Clustering has traditionally been used for partitioning the objects of a single dataset. Some applications may require the clustering of multiple related heterogeneous datasets where it may not be easy to compute a useful and effective integrated feature space. In this paper, we present an algorithm called CEMENT (Cluster Ensemble using Mutual ENTropy) to address the problem of clustering two related datasets where the datasets represent the same or overlapping sets of objects but use different feature sets. The algorithm takes the partitional clusters generated from two datasets as input and uses a constraint-based approach to generate a single set of clusters. CEMENT is an EM (expectation maximization) approach where the objective function is the mutual entropy between the two sets of clusters. The algorithm was applied to the problem of clustering a document collection consisting of journal abstracts from ten different Library of Congress categories. These documents were pre-processed using several NLP (natural language processing) steps to extract syntactic and semantic feature sets. We present empirical results and statistical tests showing that CEMENT yields higher quality clusters with this dataset than several baseline clustering approaches.

Traditional clustering is typically based on a single feature set. In some domains, several feature sets may be available to represent the same objects, but it may not be easy to compute an integrated feature set. We have developed the EPaCH (Ensemble method for generating Partitional clusters from multiple Cluster Hierarchies) algorithm to address the problem of combining the results of hierarchical clustering from multiple related datasets where the datasets represent the same set of objects but use different feature sets. EPaCH uses a graph theoretic approach to combine the hierarchies into a single set of partitional clusters. A graph is generated from the hierarchies based on the association strengths of objects in the hierarchies. A graph partitioning algorithm is then applied to generate flat clusters. EPaCH was tested empirically with a document collection consisting of journal abstracts from ten different Library of Congress categories. Both syntactic and semantic feature sets were extracted and the resulting datasets were clus- tered individually using average-link agglomerative hierarchical clustering. EPaCH was then used to generate a single set of flat clusters from the dendrograms. In the document clustering domain, EPaCH is shown to yield higher quality clusters than phylogeny-based ensemble methods and than clustering based on a single feature set for three of four measures of cluster quality.

Document clustering generates clusters from the whole document collection automatically and is used in many fields, including data mining and information retrieval. In the traditional vector space model, the unique words occurring in the document set are used as the features. But because of the synonym problem and the polysemous problem, such a bag of original words cannot represent the content of a document precisely. In this paper, we investigate using the sense disambiguation method to identify the sense of words to construct the feature vector for document representation. Our experimental results demonstrate that in most conditions, using sense can improve the performance of our document clustering system. But the comprehensive statistical analysis performed indicates that the differences between using original single words and using senses of words are not statistically significant. In this paper, we also provide an evaluation of several basic clustering algorithms for algorithm selection.

Document clustering is a widely used strategy for information retrieval and text data mining. Partitioning and hierarchical clustering methods are most widely used algorithms. Other investigators proposed to use bisecting K-means method for document clustering and their experimental results have indicated that the bisecting K-means method is the preferred method for document clustering [16]. However, in our research we have found that, whereas the bisecting K-means method has advantages when working with large datasets, a traditional hierarchical clustering algorithm still achieves the best performance for small datasets.

Document clustering is a widely used strategy for information retrieval and text data mining. In traditional document clustering systems, documents are represented as a bag of independent words. In this project, we propose to enrich the representation of a document by incorporating semantic information and syntactic information. Semantic analysis and syntactic analysis are performed on the raw text to identify this information. A detailed survey of current research in natural language processing, syntactic analysis, and semantic analysis is provided. Our experimental results demonstrate that incorporating semantic information and syntactic information can improve the performance of our document clustering system for most of our data sets. A statistically significant improvement can be achieved when we combine both syntactic and semantic information. Our experimental results using compound words show that using only compound words does not improve the clustering performance for our data sets. When the compound words are combined with original single words, the combined feature set gets slightly better performance for most data sets. But this improvement is not statistically significant. In order to select the best clustering algorithm for our document clustering system, a comparison of several widely used clustering algorithms is performed. Although the bisecting K-means method has advantages when working with large datasets, a traditional hierarchical clustering algorithm still achieves the best performance for our small datasets.

WebDoc is an automated classification system that assigns Web documents to appropriate Library of Congress subject headings based upon the text in the documents. We have used different classification methods in different versions of WebDoc. One classification method is a statistical approach that counts the number of occurrences of a given noun phrase in documents assigned to a particular subject heading as the basis for determining the weights to be assigned to the candidate indexes. The second classification method uses a naïve Bayes approach. In this case, we experimented with the use of smoothing to dampen the effect of having a large number of 0s in our feature vectors. The third classification method is a k-nearest neighbors approach. With this approach, we tested two different ways of determining the similarity of feature vectors. In this paper, we report the performance of each of the versions of WebDoc in terms of recall, precision, and F-measures.

Document clustering is a kind of text data mining and organization technique that automatically groups related documents into clusters. Traditionally single words occurring in the documents are identified to determine the similarities among documents. In this work, we investigate using compound words as features for document clustering. Our experimental results demonstrate that using compound words alone cannot improve the performance of clustering system. Promising results are achieved when the compound words are combined with the original single words to be the features. An evaluation of several basic clustering algorithms is also performed in our work for algorithm selection. Although the bisecting K-means method has been proposed as a good document clustering algorithm by other investigators, our experimental results demonstrated that for small datasets, a traditional hierarchical clustering algorithm still achieves the best performance.

Document clustering is a widely used strategy for information retrieval and text data mining. This paper describes the preliminary work for ongoing research of document clustering problems. A prototype of a document clustering system has been implemented and some basic aspects of document clustering problems have been studied. Our experimental results demonstrate that the average-link inter-cluster distance measure and TFIDF weighting function are good methods for the document clustering problem. Other investigators have indicated that the bisecting K-means method is the preferred method for document clustering. However, in our research we have found that, whereas the bisecting K-means method has advantages when working with large datasets, a traditional hierarchical clustering algorithm still achieves the best performance for small datasets.

This dissertation provides an ad hoc integration methodology to manage and integrate heterogeneous online distributed databases on demand. The problem arises from an impending demand from scientific users to conveniently manage existing Web data along with the complexity involved in the construction of a functional data federation system using existing data integration technologies. We close this gap with a databases management framework accompanying novel Web data specification languages, wrapper generation technologies, and distributed query processing techniques. A major achievement of this dissertation is the establishment of a sound relational data model for Web data. Under this model, the Web becomes a synthetic extension of the traditional database systems. Consequently, a novice user of our system can cheaply integrate a large number of distributed Web sources with in-house databases for daily scientific data analysis purpose. The relational Web modeling leads to a practical ad hoc integration system—the Meteoroid system (a MEthodology for ad hoc inTEgration of Online distributed heteROgeneous Internet Data)—in the context of biological data interoperability. We identify that a main difficulty for ad hoc integration lies in the lack of a fully automated wrapper generation and maintenance technique for general semi-structured data such as HTML, XML and plain text documents. We address this issue through a thorough study of characteristics of online Web data and devise various automated wrapper techniques to facilitate robust data wrapping tasks. With this technique, form-based Web data and table-based Web data can be treated like traditional relational databases. A seamless interoperation environment for Web data and in-house databases is possible. Another difficulty impeding ad hoc integration is in the query processing for heterogeneous distributed sources, where conflict of data is common and on demand mediation of distributed sources is desirable. The dynamicity and unpredictability of Web data further complicate the query processing task. We studied limitations posed by the Web environment for integration query processing and developed innovative techniques to expedite the early appearance of available results. Finally we demonstrate a prototype system for ad hoc integration of heterogeneous biological data. In the system, visual Web-based interfaces guide the integration of heterogeneous data for novice users. A declarative environment is supported for ad hoc querying and management of distributed data sources.

This paper presents an automatic document classification system, WebDoc, which classifies Web documents according to the Library of Congress classification scheme. WebDoc constructs a knowledge base from the training data and then classifies the documents based on information in the knowledge base. One of the classification algorithms used in WebDoc is based on Bayes' theorem from probability theory. This paper focuses upon three aspects of this approach: different event models for the naive Bayes method, different probability smoothing methods, and different feature selection methods. In this paper, we report the performance of each method in terms of recall, precision, and F-measures. Experimental results show that the WebDoc system can classify Web documents effectively and efficiently.

Knowledge discovery from large acoustic images is a computationally intensive task. The data-mining step in the knowledge discovery process that involves unsupervised learning (clustering) consumes the bulk of the computation. We have developed a technique that allows us to partition the data, distribute it to different processors for training, and train a single system to join the results of the independent categorizers. We report preliminary results using this approach for knowledge discovery with large acoustic images having more than 10, 000 training instances.

Scientists at the Mississippi State University Diagnostic Instrumentation and Analysis Laboratory and the Idaho National Engineering and Environmental Laboratory (INEEL) have developed an expert system for a noninvasive characterization of containerized radiological waste. The characterization of the containers is necessary for determining their proper disposition. Three prototypes were developed, with each using a different method of handling uncertainty-a fuzzy system, a Bayesian network system, and a neural network system. The performance of each expert system was assessed to determine how well it modeled the decisions made by the INEEL domain expert. The prototype systems were also analyzed to measure the agreement in their decisions, the domain expert's decisions, and the decisions made by two additional experts. The neural network prototype was further analyzed to determine how consistent it was in its assessments. This paper describes the analysis of the performance of the three expert system prototypes.