Creating natural language tutorial dialogue systems that realize effective strategies is a central challenge for intelligent tutoring systems research. Traditional approaches generally require large development time, do not generalize well across domains, and do not match the flexibility and natural language sophistication of human tutors. A promising approach that may offer several benefits is data-driven system development, in which a dialogue policy is learned from corpora of human tutorial dialogue. To date these learning approaches typically focus on optimizing the tutor’s choice of act, and do not explicitly model the instances in which the tutor chose not to act. This paper reports on a hidden Markov modeling (HMM) approach within human textual tutorial dialogue that explicitly represents the tutors’ choices not to intervene. The results show that an HMM that models tutor non-interventions predicts tutor moves significantly better than a model that does not explicitly represent the non-interventions. The findings have implications for automatically modeling tutorial strategies and for learning dialogue policies from corpora.

Behavior-based and reactive control methods are popular choices for building fast and lightweight intelligent controllers for resource-constrained systems. Reactive methods are extremely useful in highly resource-constrained applications, but at a cost: they tend to be even more susceptible to certain types of failures than deliberative techniques. Without a planner to adapt to changes, even a small failure can result in incorrect behavior from the entire controller. In this paper, we propose extensions to behavior-based subsumption that can detect four types of failures.

In a competitive game it is important to identify the opponent's strategy as quickly and accurately as possible so that an effective response can be planned. In this vein, this paper summarizes our work in exploring using first order inductive learning to learn rules for representing opponent strategies. Specifically, we use these learned rules to perform plan recognition and classify an opponent strategy as one of multiple learned strategies. Our experiments validate this novel approach in a simple real-time strategy game.

There are two primary types of graph-based data miners: frequent subgraph and compression-based miners. With frequent subgraph miners, the most interesting substructure is the largest one (or ones) that meet the minimum support. Whereas, compression-based graph miners discover those subgraphs that maximize the amount of compression that a particular substructure provides a graph. The algorithms associated with these two approaches are not only different, but they also may result in dramatic performance differences, as well as in the normative patterns being discovered. In order to compare these two types of graph-based approaches to knowledge discovery, in the following sections we will compare two publicly available applications: GASTON and SUBDUE.

The New York Public Library is participating in the Chronicling America initiative to develop an online searchable database of historically significant newspaper articles. Microfilm copies of the papers are scanned and high resolution OCR software is run on them. The text from the OCR provides a wealth of data and opinion for researchers and historians. However, the categorization of articles provided by the OCR engine is rudimentary and a large number of the articles are labeled ``editorial" without further categorization. To provide a more refined grouping of articles, unsupervised machine learning algorithms (such as K-Means) are being investigated. The K-Means algorithm requires tuning of parameters such as the number of clusters and mechanism of seeding to ensure that the search is not prone to being caught in a local minima. We designed a pilot study to observe whether humans are adept at finding sub-categories. The subjective labels provided by humans are used as a guide to compare performance of the automated clustering techniques. In addition, seeds provided by annotators are carefully incorporated into a semi-supervised K-Means algorithm (Seeded K-Means); empirical results indicate that this helps to improve performance and provides an intuitive sub-categorization of the articles labeled ``editorial" by the OCR engine.

Community mining is applied in order to identify groups of users which share, e.g., common interests or expertise. This paper presents an approach for mining descriptive patterns in order to characterize communities in terms of their distinctive features: For an efficient discovery approach, we introduce optimistic estimates for obtaining an upper bound for the community quality. We present an evaluation using data from the real-world social bookmarking system BibSonomy.

The filter model of feature selection has been well studied. In previous studies, classification performance has traditionally been proposed as a way to evaluate filter solutions. In this study, a new method of comparing feature ranking techniques is presented providing a straightforward approach for quantifying individual filters’ robustness to class noise. Six commonly-used filters, plus one which is rarely used, are investigated regarding their ability to retain, in the presence of class noise, strong classification performance. Three classifiers and one classification performance metric are considered. The experimental results of this study show that Gain Ratio, one of the well known and widely used filters, is very sensitive to class noise. ReliefF offers the best results with both the NB and kNN learners while Signal-to-noise, though not as widely used in the literature as the others, outperforms all the filters with the SVM learner.

We futher claim that categorization within this domain is a recursive process of subdivision. We describe a theory of perceptual understanding, implemented Despite its central role in our theory, categorization isn't in Mathematica, that forms rich representations of the only way people demonstrate understanding of exemplar very simple visual concepts. We suggest that this theory can sets. In order to accommodate this, our theory supports a variety represent any category distinction that a human is liable to of other operations, such as outlier detection and concept make within its limited domain.

In this paper, we show how to make a cognitive tutoring agent capable of precise causal reasoning by integrating constraints with data mining algorithms. Putting constraints on recorded interactions between the agent and learners during learning activities allows data mining algorithms to extract the causes of the learners’ problems. Subsequently, the agent uses this information to provide useful and customized explanations to learners.

We have experimented with the notion of — conceptualization, and contextualization from situated cognition and psychic reflection from activity theory for identifying context into a method called the activity states framework (ASF). The purpose of the ASF is to provide a method of analysis for identifying collaborators activity during situated context − specific to Web-mediated collaboration. This paper introduces the ASF.