Abstract We have proposed that scientific research reports should be constructed entirely of structured knowledge rather than text. In an earlier paper, we emphasized Research Designs as a framework for structured research reports and described how a structured implementation might be applied to Pasteur’s classic swan-neck flask experiment. In this paper, we examine some of the issues encountered in developing that implementation using dynamic models. For instance, we consider issues associated with modeling state transitions.

Learning to program, especially in the object-oriented paradigm, is a difficult undertaking for many students. As a result, computing educators have tried a variety of instructional methods to assist beginning programmers. These include developing approaches geared specifically toward novices and experimenting with different introductory programming languages. However, determining the effectiveness of these interventions poses a problem. The research presented here developed an instrument to assess student learning of fundamental and object-oriented programming concepts, then used that instrument to investigate the impact of different teaching approaches and languages on university students’ ability to learn those concepts. Extensive data analysis showed that the instrument performed well overall. Reliability of the assessment tool was statistically satisfactory and content validity was supported by intrinsic characteristics, question response analysis, and expert review. Preliminary support for construct validity was provided through exploratory factor analysis. Three components that at least partly represented the construct “understanding of fundamental programming concepts” were identified: methods and functions, mathematical and logical expressions, and control structures. Analysis revealed significant differences in student performance based on instructional language and approach. The analyses showed differences on the overall score and questions involving assignment, mathematical and logical expressions, and code completion. Instructional language and approach did not appear to affect student performance on questions addressing object-oriented concepts.

Supervised learning methods rely on large sets of labeled training examples. However, large training sets are rare and making them is expensive. In this research, Latent Semantic Indexing Subspace Signature Model (LSISSM) is applied to labeling for active learning of unstructured text. Based on Singular Value Decomposition (SVD), LSISSM represents terms and documents as semantic signatures by the distribution of their local statistical contribution across the top-ranking LSI latent dimensions after dimension reduction. When utilized to an unlabeled text corpus, LSISSM finds the most important samples and terms according to their global statistical contribution ranking in the corresponding LSI subspaces without prior knowledge of labels or dependency to model-loss functions of the classifiers. These sample subsets also effectively maintain the sampling distribution of the whole corpus. Furthermore, tests demonstrate that the sample subsets with the optimized term subsets substantially improve the learning accuracy across three standard classifiers.

We describe the latent semantic indexing subspace signature model (LSISSM) for semantic content representation of unstructured text. Grounded on singular value decomposition, the model represents terms and documents by the distribution signatures of their statistical contribution across the top‐ranking latent concept dimensions. LSISSM matches term signatures with document signatures according to their mapping coherence between latent semantic indexing (LSI) term subspace and LSI document subspace. LSISSM does feature reduction and finds a low‐rank approximation of scalable and sparse term‐document matrices. Experiments demonstrate that this approach significantly improves the performance of major clustering algorithms such as standard K‐means and self‐organizing maps compared with the vector space model and the traditional LSI model. The unique contribution ranking mechanism in LSISSM also improves the initialization of standard K‐means compared with random seeding procedure, which sometimes causes low efficiency and effectiveness of clustering. A two‐stage initialization strategy based on LSISSM significantly reduces the running time of standard K‐means procedures.

No abstract available.

Abstract Some collections cover many topics, while others are narrowly focused on a limited number of topics. We introduce the concept of the “scope” of a collection of documents and we compare two ways of measuring it. These measures are based on the distances between documents. The first uses the overlap of words between pairs of documents. The second measure uses a novel method that calculates the semantic relatedness to pairs of words from the documents. Those values are combined to obtain an overall distance between the documents. The main validation for the measures compared Web pages categorized by Yahoo. Sets of pages sampled from broad categories were determined to have a higher scope than sets derived from subcategories. The measure was significant and confirmed the expected difference in scope. Finally, we discuss other measures related to scope.

We examine metadata and data-structure issues for the Historical Newspaper Digital Library. This project proposes to digitize and then do OCR and linguisting processing on several years worth of historical newspapers. Newspapers are very complex information objects so developing a rich description of their content is challenging. In addition to frameworks for the logical structure and physical layout, we propose metadata relevant to the image processing and to the historians who will use this collection. Finally, we consider how the metadata infrastructure might be managed as it evolves with improved text processing capabilities and how an infrastructure might be developed to support a community of users.

Boltzmann-based models with asymmetric connections are investigated. Although they are initially unstable, these networks spontaneously self-stabilize as a result of learning. Moreover, pairs of weights symmetrize during learning; however, the symmetry is not enough to account for the observed stability. To characterize the system it is useful to consider how its entropy is affected by learning and the entropy of the information stream. The stability of an asymmetric network is confirmed with an electronic model.

A reinforcement training procedure was developed for sequential back-propagation networks and applied in several studies demonstrating interaction between agents in multiple-agent networks. In the first study, a network was trained to predict the next position of an agent which was moving in a complex pattern around the corners of a square. The network quickly learned to predict the position without error. In particular, the network may be said to have developed an agent or user model of the moving agent. In two additional studies, a joint contingency was applied to two agents and limited cooperation was developed between them. Overall, the results provide support for the application of neural networks in distributed AI (artificial intelligence).>

Users without prior computer experience employed a menu-and-tree interface to search a database relevant to telecommunications services. Detailed data were recorded on the latency of each response and whether an error was made. The depth, level, vertical position, number of options, and trial were found to significantly affect response latency. The level, content, and side of the screen on which the choice was displayed significantly affected error rate. The results of the experiment and additional issues in cognitive human factors are discussed.

KeystrokesThe Keystroke-Level Model asserts that the time for an expert to enter a task using a command language is a function of specific task-acquisition, mental, and motor-response times.The evidence for the model is critically reviewed, and new data are presented.The fit of the new data to the model is modest even when several modifications of the model are considered.It is proposed that a more complex model, based explicitly on cognitive processes, is necessary.