Kalyan Moy Gupta

Exploiting Taxonomic and Causal Relations in Conversational Case Retrieval

Conversational case-based reasoning (CCBR) systems engage their users in a series of questions and answers and present them with cases that are most applicable to their decision problem. In previous research, we introduced the Taxonomic CCBR methodology, an extension of standard CCBR that improved performance by organizing features related by abstraction into taxonomies. We recently extended this methodology to include causal relations between taxonomies and claimed that it could yield additional performance gains. In this paper, we formalize the causal extension of Taxonomic CCBR, called Causal CCBR, and empirically assess its benefits using a new methodology for evaluating CCBR performance. Evaluation of Taxonomic and Causal CCBR systems in troubleshooting and customer support domains demonstrates that they significantly outperform the standard CCBR approach. In addition, Causal CCBR outperforms Taxonomic CCBR to the extent causal relations are incorporated in the case bases.

Taxonomic Conversational Case-Based Reasoning

Conversational Case-Based Reasoning (CCBR) systems engage a user in a series of questions and answers to retrieve cases that solve his/her current problem. Help-desk and interactive troubleshooting systems are among the most popular implementations of the CCBR methodology. As in traditional CBR systems, features in a CCBR system can be expressed at varying levels of abstraction. In this paper, we identify the sources of abstraction and argue that they are uncontrollable in applications typically targeted by CCBR systems. We contend that ignoring abstraction in CCBR can cause representational inconsistencies, adversely affect retrieval and conversation performance, and lead to case indexing and maintenance problems. We propose an integrated methodology called Taxonomic CCBR that uses feature taxonomies for handling abstraction to correct these problems. We describe the benefits and limitations of our approach and examine issues for future research.

Knowledge-Based Project Planning

Project management is a business process for successfully delivering one-of-a kind products and services under real-world time and resource constraints. Developing a project plan, a crucial element of project management, is a difficult task that requires significant experience and expertise. Interestingly, artificial intelligence researchers have developed both mixed-initiative and automated hierarchical planning systems for reducing planning effort and increasing plan evaluation measures. However, they have thus far not been used in project planning, in part because the relationship between project planning and hierarchical planning has not been established, hi this chapter, we identify this relationship and explain how project planning representations called work breakdown structures (WBS) are similar to plan representations employed by hierarchical planners. We exploit this similarity and apply well-known hierarchical planning techniques, including an integrated (case-based) plan retrieval module, to assist a project planner efficiently create WBSs. Our approach uses stored episodes (i.e., cases) of previous project planning experiences to support the development of new plans. We present an architecture for knowledge-based project planning system that implements this approach.

Case-Based Collective Inference for Maritime Object Classification

Maritime assets such as merchant and navy ships, ports, and harbors, are targets of terrorist attacks as evidenced by the USS Cole bombing. Conventional methods of securing maritime assets to prevent attacks are manually intensive and error prone. To address this shortcoming, we are developing a decision support system that shall alert security personnel to potential attacks by automatically processing maritime surveillance video. An initial task that we must address is to accurately classify maritime objects from video data, which is our focus in this paper. Object classification from video images can be problematic due to noisy outputs from image processing. We approach this problem with a novel technique that exploits maritime domain characteristics and formulates it as a graph of spatially related objects. We then apply a case-based collective classification algorithm on the graph to classify objects. We evaluate our approach on river traffic video data that we have processed. We found that our approach significantly increases classification accuracy in comparison with a conventional (i.e., non-relational) alternative.

Soft computing techniques for web services brokering

To enhance and improve the interoperability of meteorological Web Services, we are currently developing an Integrated Web Services Brokering System (IWB). IWB uses a case-based classifier to automatically discover Web Services. In this paper, we explore the use of rough set techniques for selecting features prior to classification. We demonstrate the effectiveness of this feature technique by comparing it with a leading non-rough set (Information Gain) feature selection technique.

Rough set feature selection algorithms for textual case-based classification

Feature selection algorithms can reduce the high dimensionality of textual cases and increase case-based task performance. However, conventional algorithms (e.g., information gain) are computationally expensive. We previously showed that, on one dataset, a rough set feature selection algorithm can reduce computational complexity without sacrificing task performance. Here we test the generality of our findings on additional feature selection algorithms, add one data set, and improve our empirical methodology. We observed that features of textual cases vary in their contribution to task performance based on their part-of-speech, and adapted the algorithms to include a part-of-speech bias as background knowledge. Our evaluation shows that injecting this bias significantly increases task performance for rough set algorithms, and that one of these attained significantly higher classification accuracies than information gain. We also confirmed that, under some conditions, randomized training partitions can dramatically reduce training times for rough set algorithms without compromising task performance.

Rough set feature selection methods for case-based categorization of text documents

Textual case bases can contain thousands of features in the form of tokens or words, which can inhibit classification performance. Recent developments in rough set theory and its applications to feature selection offer promising approaches for selecting and reducing the number of features. We adapt two rough set feature selection methods for use on n-ary class text categorization problems. We also introduce a new method for selecting features that computes the union of features selected from randomly-partitioned training subsets. Our comparative evaluation of our method with a conventional method on the Reuters-21578 data set shows that it can dramatically decrease training time without compromising classification accuracy. Also, we found that randomized training set partitions dramatically reduce training time.

A comparative evaluation of anomaly detection algorithms for maritime video surveillance

A variety of anomaly detection algorithms have been applied to surveillance tasks for detecting threats with some success. However, it is not clear which anomaly detection algorithms should be used for domains such as ground-based maritime video surveillance. For example, recently introduced algorithms that use local density techniques have performed well for some tasks, but they have not been applied to ground-based maritime video surveillance. Also, the reasons for the performance differences of anomaly detection algorithms on problems of varying difficulty are not well understood. We address these two issues by comparing families of global and local anomaly detection algorithms on tracks extracted from ground-based maritime surveillance videos. Obtaining maritime anomaly data can be difficult or even impractical. Therefore, we use a generative approach to vary and control the difficulty of anomaly detection tasks and to focus on borderline and difficult situations in our empirical comparison studies. We report that global algorithms outperform local algorithms when tracks have large and unstructured variations, while they perform equally well when the tracks have only minor variations.

Adaptive maritime video surveillance

Maritime assets such as ports, harbors, and vessels are vulnerable to a variety of near-shore threats such as small-boat attacks. Currently, such vulnerabilities are addressed predominantly by watchstanders and manual video surveillance, which is manpower intensive. Automatic maritime video surveillance techniques are being introduced to reduce manpower costs, but they have limited functionality and performance. For example, they only detect simple events such as perimeter breaches and cannot predict emerging threats. They also generate too many false alerts and cannot explain their reasoning. To overcome these limitations, we are developing the Maritime Activity Analysis Workbench (MAAW), which will be a mixed-initiative real-time maritime video surveillance tool that uses an integrated supervised machine learning approach to label independent and coordinated maritime activities. It uses the same information to predict anomalous behavior and explain its reasoning; this is an important capability for watchstander training and for collecting performance feedback. In this paper, we describe MAAWs functional architecture, which includes the following pipeline of components: (1) a video acquisition and preprocessing component that detects and tracks vessels in video images, (2) a vessel categorization and activity labeling component that uses standard and relational supervised machine learning methods to label maritime activities, and (3) an ontology-guided vessel and maritime activity annotator to enable subject matter experts (e.g., watchstanders) to provide feedback and supervision to the system. We report our findings from a preliminary system evaluation on river traffic video.

Case-based classification alternatives to ontologies for automated web service discovery and integration

Web Services are becoming the standard technology used to share data for many Navy and other DoD operations. Since Web Services technologies provide for discoverable, self-describing services that conform to common standards, this paradigm holds the promise of an automated capability to obtain and integrate data. However, automated integration of applications to access and retrieve data from heterogeneous sources in a distributed system such as the Internet poses many difficulties. Assimilation of data from Web-based sources means that differences in schema and terminology prevent simple querying and retrieval of data. Thus, machine understanding of the Web Services interface is necessary for automated selection and invocation of the correct service. Service availability is also an issue that needs to be resolved. There have been many advances on ontologies to help resolve these difficulties to support the goal of sharing knowledge for various domains of interest. In this paper we examine the use of case-based classification as an alternative/supplement to using ontologies for resolving several questions related to knowledge sharing. While ontologies encompass a formal definition of a domain of interest, case-based reasoning is a problem solving methodology that retrieves and reuses decisions from stored cases to solve new problems, and case-based classification involves applying this methodology to classification tasks. Our approach generalizes well in sparse data, which characterizes our Web Services application. We present our study as it relates to our work on development of the Advanced MetOc Broker, whose objective is the automated application integration of meteorological and oceanographic (MetOc) Web Services.