IEEE Intelligent Systems

We've Come a Long Way, Maybe �

Despite all the major AI successes, when it comes to really understanding the human mind, we still know very little.

The Bigger Picture

Letters to the Editor.

In the News

In the first story, "Robot Swarms Tackle Big and Small Tasks," the author considers how a mass of small robots working together can do things that a larger robot could not. The second story, "AI Sound Recognition Makes Quiet Gains," looks at advances in sound recognition to identify distinct sounds.

Managing Household Wind-Energy Generation

Intelligent agent technology can be used to aggregate wind-energy generation installed at a large number of households and battery storage installed in similarly large numbers but not necessarily at the same households, creating a "virtual generator" that can be dispatched on the electricity grid in a similar manner to centralized generation. The purpose of aggregation is to sell renewable generation to the electricity network and market at a price commensurate with its true value. If this can provide a better return on investment in renewable energy, it will encourage more householders to buy their own generators and contribute to reducing greenhouse-gas emissions.

IT Strategies for Increased Rail Employee Satisfaction

Each week, 7,500 Canadian National Railway employees are assigned jobs on the basis of their preferences and seniority. CN seeks to leverage its implementation of an SAP enterprise system to improve employee satisfaction with the job assignment system. A team of Cornell University students investigated strategies to predict employee satisfaction, improve job assignment algorithms, enable selection of coworkers, and reduce the number of nights workers must spend away from home.

Natural Language Processing and the Web

This special issue focuses on applications that innovatively use the Web and Web-scale document collections to create useful resources or applications that let end users navigate the Web more easily. This article is part of a special issue on Natural Language Processing and the Web.

Mining the Web to Create Specialized Glossaries

Glossaries are helpful for integrating information, reducing semantic heterogeneity, and facilitating communication between information systems. Commercial publishers charge lexicographers with building glossaries, but this isn't appropriate when a domain's semantics are continuously evolving rather than precisely characterized, as in emerging Web communities and interest groups. In emerging domains, glossary building is the cooperative effort of a team of domain experts. It involves several steps, including identifying the domain-relevant terminology, defining each term, and harmonizing the results. This is a time-consuming, costly process that often requires support from a collaborative platform to facilitate shared decisions and validation. TermExtractor and GlossExtractor, two Web applications based on Web mining techniques, support this complete glossary-building procedure. The tools exploit the Web's evolving nature, allowing one to continually update the emerging community's vocabulary. TermExtractor and GlossExtractor, which were used in the European project Interop, are freely available and are being used in experiments in different domains across the world. This article is part of a special issue on Natural Language Processing and the Web.

Learning to Tag and Tagging to Learn: A Case Study on Wikipedia

Natural language technologies have long been envisioned to play a crucial role in developing a Semantic Web. Textual content's significance on the Web has increased with the rise of Web 2.0 and mass participation in content generation. Yet, natural language technologies face great challenges in dealing with Web content's heterogeneity: key among these is domain and task adaptation. To address this challenge, the authors consider the problem of semantically annotating Wikipedia. Specifically, they investigate a method for dealing with domain and task adaptation of semantic taggers in cases where parallel text and metadata are available. By creating a semantic mapping among vocabularies from two sources: Wikipedia and the original annotated corpus, they improve their tagger on Wikipedia. Moreover, by applying their tagger and mapping between sources, they significantly extend the metadata currently available in the DBpedia collection. This article is part of a special issue on Natural Language Processing and the Web.

Linking Documents to Encyclopedic Knowledge

Wikipedia can support the development of automatic methods for keyword extraction and word-sense disambiguation. The Wikify system combines these two methods to automatically enrich a text with links to Wikipedia content. The system identifies the important concepts in a given document and automatically links these concepts to the corresponding Wikipedia pages. An evaluation of the system using a Turing-like test shows that the automatic annotations are hardly distinguishable from manual annotations. A second evaluation in an educational environment shows that enriching educational materials with such annotations can improve the learning process by allowing faster access to background knowledge. This article is part of a special issue on Natural Language Processing and the Web.

Weighing Stars: Aggregating Online Product Reviews for Intelligent E-commerce Applications

The author identifies a new task in the ongoing research in text sentiment analysis: aggregating online product reviews in light of two orthogonal dimensions, namely, polarity/opinion extraction and usefulness scoring. The motivation is to build future review aggregation or ranking services that enable both online shoppers and vendors to better leverage information from multiple sources. Usefulness scoring is viewed as a regression problem. The author builds support-vector-regression models by incorporating a diverse set feature set computed from review text, which achieved promising performance on four Amazon product review collections. Findings also indicate that a product review's perceived usefulness is highly dependent on its linguistic style. Further rank correlation analyses on the Amazon data demonstrates the feasibility and advantage of the proposed review-aggregation framework, in the context of predicting market response to certain products. This article is part of a special issue on Natural Language Processing and the Web.

A Multiagent System for Coordinating Ambulances for Emergency Medical Services

To coordinate ambulances for emergency medical services, a multiagent system uses an auction mechanism based on trust. Results of tests using real data show that this system can efficiently assign ambulances to patients, thereby reducing transportation time.

Recognition of Complex Settings by Aggregating Atomic Scenes

Researchers have employed audio features to capture complex human settings. Most approaches model a complex setting as a monolithic scene; that is, they consider the stochastic property of the audio signal representing a setting as a whole, not an aggregation of distinct scenes. So, when some aspects of the training data are missing or are weakly represented in the test signal, recognition schemes trained to recognize the setting often make erroneous conclusions. Moreover, these approaches make it difficult to declaratively define new settings by combining scenes. A proposed conceptual architecture enables recognition of complex settings by combining scenes. The associated architecture and modeling approach help achieve human-like reasoning and improve recognition accuracy. The authors demonstrate their approach by modeling seven everyday settings with 27 atomic scenes.

Exploring Alternative Software Architecture Designs: A Planning Perspective

A software architecture is a blueprint that captures the main design decisions for a system. When designing an architecture, the architect normally evaluates multiple solutions, making a balance among architectural patterns that affect several quality attributes (for example, performance, modifiability, and so on). In this exploration of the design space, the architectural knowledge directs the search toward a good-enough solution. Although decision-making still relies on the architect's expertise, a novel architectural design theory has been recently developed to move more systematically from quality attributes to architectural decisions. In this content, we describe a framework called DesignBots to search for design alternatives, in which the concepts of that theory are mapped to a hierarchical and mixed-initiative planning model. Essentially, quality-attribute scenarios are considered as goals achievable by combinations of patterns that are instantiated by the planning engine. This approach fosters the development of proactive assistants to support architectural design activities.

Influencing versus Informing Design, Part 1: A Gap Analysis

The collaboration of cognitive systems engineers with systems engineers is motivated by the goal of creating human-centered systems. However, there can be a gap in this collaboration. In presentations at professional meetings about cognitive systems engineering projects, we often hear that one or another method of cognitive task analysis was employed in order to inform design. But what software developers need is designs. This is the first of two essays about the gap between the products of cognitive task analysis and the needs of the software engineers. We discuss a success story of cognitive systems engineering for a large-scale system, a project that coped with the practical constraints of time pressure and the challenge of designing for an envisioned world when system elements could not be fully specified in advance. This project relied on a particular product from cognitive task analysis, the abstraction-decomposition matrix, that speaks in a language that corresponds with the needs and goals of the software designers.

Web Semantics in the Clouds

Cloud Computing refers to the use of large-scale computer clusters often built from low-cost hardware and network equipment, where resources are allocated dynamically among users of the cluster. While the paradigm is not entirely novel, recent developments in software frameworks for Cloud Computing are making it increasingly easy for programmers to parallelize and thereby scale-up complex data-processing tasks. This article investigates how this trend is impacting the Semantic Web field and shows how Cloud Computing can be used to analyze, query, and reason with the massive amounts of metadata handled by semantic search engines.