This paper describes a novel application of text categorization for mathematical word problems , namely Multiplicative Compare and Equal Group problems. The empirical results and analysis show that common text processing techniques such as stopword removal and stemming should be selectively used. It is highly beneficial not to remove stopwords and not to do stemming. Part of speech tagging should also be used to distinguish words in discriminative parts of speech from the non-discriminative parts of speech which not only fail to help but even mislead the categorization decision for mathematical word problems. An SVM classifier with these selectively used text processing techniques outperforms an SVM classifier with a default setting of text processing techniques (i.e. stopword removal and stemming). Furthermore, a probabilistic meta classifier is proposed to combine the weighted results of two SVM classifiers with different word problem representations generated by different text preprocessing techniques. The empirical results show that the probabilistic meta classifier further improves the categorization accuracy.

One of the most important goals in Intelligent Tutoring is to create applications that can evaluate the quality of a text in a human-like manner. The aim of this study is to compare three methods of using Latent Semantic Analysis (LSA) to evaluate the quality of summaries written by students in Spanish. The sample is made up by 226 summaries written by Chilean students based on both expository and narrative texts. Each summary was first assessed by human judges in order to compare the results with the scoring provided by three different LSA methods: a) comparing the summaries with the original text divided in paragraphs, b) comparing the summaries with the text as one unit, and c) comparing the summaries with other summaries written by four human experts. Results show that comparison between each student’s summary and the text as a one unit constitutes the method which most closely resembles human evaluation.

Our past work has investigated the use of the Cognitive Model Software Development Kit (SDK) for creating the cognitive models that underlie model-tracing Cognitive Tutors. Though successful at increasing the number of people who could author such a cognitive model, for certain kinds of situations the Cognitive Model SDK proved cumbersome. The present work discusses a new authoring system, xPST, that allows an example-based tutor to be built on top of existing software.

In this paper we present a syntactical way to develop the adaptation capability in logical-based intelligent agents. We use prime implicants to represent the beliefs of an agent and present how syntactical belief update operators can be obtained by correlating models and prime implicants. Using prime implicants allows the introdution a new notion of belief update. We characterize this new operator both in terms of postulates and in terms of explicit operators.

Numerous logic frameworks have been proposed for modeling uncertainty and reasoning with such data. While different in syntax, the approaches of these frameworks have been classified into "annotation based" (AB) and "implication based" (IB). In this paper, we present a unified framework which allows evaluating programs in either approach. It extends existing query processing techniques to handle certainty constraints and uses heuristics to further improve the performance. Our experiments indicate that the proposed techniques yield useful tools for uncertainty reasoning.

This paper introduces a qualitative ranking function that uses signed integers to describe the surprise associated with the occurrence of events. The measure introduced, kappa++, is based on the kappa calculus but differs from it in that its semantics enable an explicit representation of complements. As a result, the kappa++ is more capable of enforcing probability theory-like constraints to carry on reasoning.

We compare two approaches to Bayesian network inference, called variable elimination (VE) and arc reversal (AR). It is established that VE never requires more space than AR, and never requires more computation (multiplications and additions) than AR.

Time series are found widely in engineering and science.  We study multiagent forecasting in time series, drawing from literature on time series, graphical models, and multiagent systems.  Knowledge representation of our agents is based on dynamic multiply sectioned Bayesian networks (DMSBNs), a class of cooperative multiagent graphical models.  We propose a method through which agents can perform one-step forecast with exact probabilistic inference.  Superior performance of our agents over agents based on dynamic Bayesian networks (DBNs) are demonstrated through experiment.

We address the problem of information fusion in uncertain environments. Imagine there are multiple experts building probabilistic models of the same situation and we wish to aggregate the information they provide. There are several problems we may run into by naively merging the information from each. For example, the experts may disagree on the probability of a certain event or they may disagree on the direction of causility between two events (e.g., one thinks A causes B while another thinks B causes A). They may even disagree on the entire structure of dependencies among a set of variables in a probabilistic network. In our proposed solution to this problem, we represent the probabilistic models as Bayesian Knowledge Bases (BKBs) and propose an algorithm called Bayesian knowledge fusion that allows the fusion of multiple BKBs into a single BKB that retains the information from all input sources. This allows for easy aggregation and de-aggregation of information from multiple expert sources and facilitates multi-expert decision making by providing a framework in which all opinions can be preserved and reasoned over.

Belief revision always results in trusting new evidence, so it may admit an unreliable one and discard a more confident one. We therefore use belief change instead of belief revision to remedy this weakness. By introducing  epistemic states, we take into account of the strength of evidence that influences the change of belief. In this paper, we present a set of postulates to characterize belief change by epistemic states and establish representation  theorems to characterize those postulates. We show that from an epistemic state, a corresponding ordinal conditional function by Spohn can be derived and the result of combining two epistemic states is thus reduced to the result from combining two corresponding ordinal conditional functions proposed by Laverny and Lang. Furthermore, when reduced to the belief revision situation, we prove that our results induce all the Darwiche and Pearl's postulates.