Keyword Extraction from a Single Document using Word Co-occurrence Statistical Information

AAAI Conferences

We present a new keyword extraction algorithm that applies to a single document without using a corpus. Frequent terms are extracted first, then a set of cooccurrence between each term and the frequent terms, i.e., occurrences in the same sentences, is generated. Co-occurrence distribution shows importance of a term in the document as follows. If probability distribution of co-occurrence between term a and the frequent terms is biased to a particular subset of frequent terms, then term a is likely to be a keyword.


Multi-Document Summaries Based on Semantic Redundancy

AAAI Conferences

This paper presents a technique for producing short summaries from multiple documents. This technique promotes the belief that informative short summaries may be generated when using conceptual representations of redundant semantic information, called topic semantic signatures. The results of DUC-2002 evaluations account for the advantages of using the techniques presented in this paper.


Minimal Text Structuring to Improve the Generation of Feedback in Intelligent Tutoring Systems

AAAI Conferences

The goal of our work is to improve the Natural Language feedback provided by Intelligent Tutoring Systems. In this paper, we discuss how to make the content presented by one such system more fluent and comprehensible, and we show how we accomplish this by using relatively inexpensive domain-independent text structuring techniques. Weshow how specific rhetorical relations can be introduced based on the data itself in a bottom-up fashion rather than being planned top-down by the discourse planner.


Algorithms for Large Scale Markov Blanket Discovery

AAAI Conferences

This paper presents a number of new algorithms for discovering the Markov Blanket of a target variable T from training data. The Markov Blanket can be used for variable selection for classification, for causal discovery, and for Bayesian Network learning. We introduce a low-order polynomial algorithm and several variants that soundly induce the Markov Blanket under certain broad conditions in datasets with thousands of variables and compare them to other state-of-the-art local and global methods with excellent results.


Choh Man Teng

AAAI Conferences

The standard formulation of association rules is suitable for describing patterns found in a given data set. A number of difficulties arise when the standard rules are used to infer about novel instances not included in the original data. In previous work we proposed an alternative formulation called interval association rules which is more appropriate for the task of inference, and developed algorithms and pruning strategies for generating interval rules. In this paper we present some theoretical and experimental analyses demonstrating the differences between the two formulations, and show how each of the two approaches can be beneficial under different circumstances.


Flairs03-071.pdf

AAAI Conferences

Association mining explores algorithms capable of detecting frequently cooccurring items in transactions. A transaction can be identified with a market basket--a list of items a customer pays for at the checkout desk. In this paper, we explore a framework for the detection of changes in the buying patterns, as affected by fashion, season, or the introduction of a new product. We present several versions of our algorithm and experimentally examine their behaviors in domains with gradually changing domains.


Learning from Reinforcement and Advice Using Composite Reward Functions

AAAI Conferences

Reinforcement learning has become a widely used methodology for creating intelligent agents in a wide range of applications. However, its performance deteriorates in tasks with sparse feedback or lengthy inter-reinforcement times. This paper presents an extension that makes use of an advisory entity to provide additional feedback to the agent. The agent incorporates both the rewards provided by the environment and the advice to attain faster learning speed, and policies that are tuned towards the preferences of the advisor while still achieving the underlying task objective. The advice is converted to "tuning" or user rewards that, together with the task rewards, define a composite reward function that more accurately defines the advisor's perception of the task. At the same time, the formation of erroneous loops due to incorrect user rewards is avoided using formal bounds on the user reward component. This approach is illustrated using a robot navigation task.


Optimizing F-Measure with Support Vector Machines

AAAI Conferences

Support vector machines (SVMs) are regularly used for classification of unbalanced data by weighting more heavily the error contribution from the rare class. This heuristic technique is often used to learn classifiers with high F-measure, although this particular application of SVMs has not been rigorously examined. We provide significant and new theoretical results that support this popular heuristic. Specifically, we demonstrate that with the right parameter settings SVMs approximately optimize F-measure in the same way that SVMs have already been known to approximately optimize accuracy. This finding has a number of theoretical and practical implications for using SVMs in F-measure optimization.


MDL-Based Context-Free Graph Grammar Induction

AAAI Conferences

We present an algorithm for the inference of context-free graph grammars from examples. The algorithm builds on an earlier system for frequent substructure discovery, and is biased toward grammars that minimize description length. Grammar features include recursion, variables and relationships. We present an illustrative example, demonstrate the algorithm's ability to learn in the presence of noise, and show real-world examples.


Subgoal Discovery for Hierarchical Reinforcement Learning Using Learned Policies

AAAI Conferences

Reinforcement learning addresses the problem of learning to select actions in order to maximize an agent's performance in unknown environments. To scale reinforcement learning to complex real-world tasks, agent must be able to discover hierarchical structures within their learning and control systems. This paper presents a method by which a reinforcement learning agent can discover subgoals with certain structural properties. By discovering subgoals and including policies to subgoals as actions in its action set, the agent is able to explore more effectively and accelerate learning in other tasks in the same or similar environments where the same subgoals are useful. The agent discovers the subgoals by searching a learned policy model for state that exhibits certain structural properties. This approach is illustrated using gridworld tasks.