Abstract: Although an ordered 3D structure is generally considered to be a necessary pre-condition for protein functionality, there are disordered counter examples found to have biological activity. The objectives of our data mining project are: (1) to generalize from the limited set of counter examples and then apply this knowledge to large data bases of amino acid sequence in order to estimate commonness of disordered protein regions in nature, and (2) to determine whether there are different types of protein disorder. For general disorder estimation, a neural network based predictor was designed and tested on data built from several public domain data banks through a nontrivial search, statistical analysis and data dimensionality reduction. In addition, predictors for identification of family-specific disorder were developed by extracting knowledge from databases generated through multiple sequence alignments of a known disordered sequence to other highly related proteins. Family-specific predictors were also integrated to test quality of general protein disorder identification from such hybrid prediction systems. Out-of-sample cross validation performance of several predictors was computed first, followed by tests on an unrelated database of proteins with long disordered regions, and the application of few selected predictors to two large protein data banks: Nrll3D, currently containing more than 10,000 protein fragments of known 3D structure, and Swiss Protein, having almost 60,000 protein sequences. The obtained results provide evidence that long disordered regions are common in nature, with an estimate that 11% of all the residues in the Swiss Protein data bank belong to disordered regions of length 40 or greater. The hypothesis that different protein disorder types exist is supported by high specificity/low sensitivity results of two family-specific predictors, by hybrid systems outperforming general models on a two-family test, and by existence of significant gaps in Swiss Protein vs. Nrll3D disorder frequency estimates for both families. These findings prompt the need for a revision in the current understanding of protein structure and function, as well as for the developing of improved disorder predictors that should have important uses in biotechnology applications.

Abstract: This paper describes the knowledge sources and methods of knowledge acquisition for USCSH (University of South Carolina SHell). USCSH is an active intelligent assistance system for Unix. The system operates in two modes, the active mode and the intelligent mode. In the active mode USCSH monitors the user's interactions with the system, and at appropriate times makes suggestions on how the user may better utilize the system to perform tasks. In the intelligent mode the system accepts questions in natural language and responds to them, taking into consideration the ability of the user and the context of the question.

Abstract: This paper presents the PlanMine sequence mining algorithm to extract patterns of events that predict failures in databases of plan executions. New techniques were needed because previous data mining algorithms were overwhelmed by the staggering number of very frequent, but entirely unpredictive patterns that exist in the plan database. This paper combines several techniques for pruning out unpredictive and redundant patterns which reduce the size of the returned rule set by more than three orders of magnitude. PlanMine has also been fully integrated into two real-world planning systems. We experimentally evaluate the rules discovered by PlanMine, and show that they are extremely useful for understanding and improving plans, as well as for building monitors that raise alarms before failures happen.

Abstract: The paper presents a set of requirements for a data mining system for mining remotely sensed satellite data based on a number of taxonomies that characterize mining of such data. The first of these taxonomies is based on knowledge of the mining objectives and mining algorithms. The second is based on various relationships that are found in data, including those between different types of data, different spatial locations of the data and different times of data capture. The paper then describes the ADaM data mining system, which was developed to address these requirements. The paper describes several data mining techniques that have been applied to remotely sensed data. The first type is target independent mining, which mines data for transients and trends, with mined results representing a highly concentrated form of the original data. The second type is the mining of vectors (representing multi-spectral or fused data) for association rules representing relationships between the various types of data represented by the elements of the vector. The third type mines data for association rules that characterize the texture of the data.

Abstract: An empirical study undertaken on a cross-section of UNIX users at an academic site reveals a role for an active form of help system, rather than the more usual passive kind. Sample scripts supporting this view are presented and the kind of aid required for these examples is discussed. It is then proposed that to provide such aid requires the construction and maintenance of an individual model of each user.

Abstract: The SINIX Consultant is an intelligent help system for the SINIX operating system which answers natural language questions about SINIX concepts and commands and gives unsolicited advice to a user as well. In this paper the representation of domain knowledge in the SINIX Consultant will be discussed, i.e. the representation of concepts of the SINIX operating system in the so-called SINIX Knowledge Base. The SINIX Knowledge Base is a taxonomical hierarchy of SINIX concepts which are divided into objects and actions operating on these objects. A single concept in the knowledge base is described by a set of attributes reflecting structural or syntactical features, the use, application and purpose of the command or object, and additional information for explaining the concept to the user. The description of commands in the SINIX Knowledge Base which is mainly used in order to generate tutorial explanations and advice for the user, and the main ideas of describing SINIX objects are outlined in this paper.

Abstract: Plan recognition in a dialogue system is the process of explaining why an utterance was made, in terms of the plans and goals that its speaker was pursuing in making the utterance. I present a theory of how such an explanation of an utterance may be judged as to its merits as an explanation. I propose three criteria for making such judgments: applicability, grounding, and completeness. The first criterion is the applicability of the explanation to the needs of the system that will use it. The second criterion is the grounding of the explanation in what is already known of the speaker and of the dialogue. Finally, the third criterion is the completeness of the explanation's coverage of the goals that motivated the production of the utterance. An explanation of an utterance is a good explanation of that utterance to the extent that it meets these three criteria. In addition to forming the basis of a method for evaluating the merit of an explanation, these criteria are useful in designing and evaluating a plan recognition algorithm and its associated knowledge base.

Abstract: The language used to describe technical domains like UNIX is filled with metaphor. An approach to metaphor, based on the explicit representation of knowledge about metaphors, has been developed. MIDAS (Metaphor Interpretation, Denotation, and Acquisition System) is a computer program that that has been developed based upon this approach. MIDAS can be used to represent knowledge about conventional metaphors, interpret metaphoric language by applying this knowledge, and dynamically learn new metaphors as they are encountered during normal processing.

Abstract: An intelligent help system needs to take into account the user's knowledge when formulating answers. This allows the system to provide more concise answers, because it can avoid telling users things that they already know. Since these concise answers concentrate exclusively on pertinent new information, they are also easier to understand. Information about the user's knowledge also allows the system to take advantage of the user's prior knowledge in formulating explanations. The system can provide better answers by referring to the user's prior knowledge in the explanation (e.g., through use of similes). This process of refining answers is called answer expression. The process of answer expression has been implemented in the UCExpress component of UC (UNIX Consultant), a natural language system that helps the user solve problems in using the UNIX operating system. UCExpress separates answer expression into two phases: pruning and formatting. In the pruning phase, subconcepts of the answer are pruned by being marked as already known by the user (and hence do not need to be generated), or marked as candidates for generating anaphora or ellipsis (since they are part of the conversational context). In the formatting phase, UCExpress uses information about the user's prior domain knowledge to select among specialized expository formats, such as similes and examples, for expressing information to the user. These formats allow UCExpress to present different types of information to the user in a clear, concise manner. The result of UCExpress' answer expression process is an internal form that a tactical level generator can easily use to produce good English.

Abstract: This paper deals with the problem of assigning meaning to the interaction of a user with a command-driven system such as UNIX This research is part of the construction of an active intelligent help system that monitors users in order to offer spontaneous help when they are facing problems In order to ensure this, the help system must build and maintain a model of the user We describe a mechanism that is used by the user modelling component of such a help system This mechanism makes explicit assumptions about the user which account for different hypotheses about what the user is actually thinking at every stage of the interaction The consistency of these assumptions is managed by an Assumption-based Truth Maintenance System (ATMS) The selection between a number of different user models is based on the information which is extracted from the feedback that UNIX provides in response to user actions

Abstract: OSCON (Operating System CONsultant) is an operating system consultant which gives English answers to English queries about computer operating systems. The program currently answers queries for over 40 commands from the UNIX and MS-DOS operating systems. OSCON answers a wide variation of queries that users may wish to ask. OSCON is intended to be a consultant for various types of users who may ask vague and detailed queries. OSCON is programmed in Quintus Prolog and answers queries in less than 2.5 seconds. An empirical study with the Wizard-of-Oz technique provides important data for the further development of OSCON.

Abstract: Yucca-a is a consultation system which is designed to provide the UNIX user, through a friendly interface, with detailed expert advice on the use of the UNIX command language. One of the principal design goals of this system is the ability to provide correct responses to technically complex queries whose solution may involve the interconnection of several commands, each with multiple options. The realization of such a goal requires two things. First, representation of dynamic knowledge about command behavior at a sufficient level of detail to support solution of the query is needed. Second, a planning mechanism capable of interconnecting such knowledge into a cohesive solution must be provided. This paper first develops the command dynamics representation techniques employed in Yucca-a. It then examines in detail the plan generation mechanism which is used to solve complex dynamic queries. Particular emphasis is placed upon those aspects of the problem which are unique to this particular domain.

Abstract: Some user interfaces, such as that of Unix, are difficult for novices to use, and this paper suggests a possible solution to such problems. The results of a study of Unix users enabled the development of a taxonomy of error types so that users' errors can be classified. This information is encapsulated as production rules within a knowledge base and forms the basis for the design and development of an intelligent interface to Unix. The prototype makes inferences about users' mental models and uses these to select appropriate tutorial advice. Performance of users of the prototype intelligent interface was compared with that of users of the usual Unix interface. The prototype users were found to make fewer errors, exhibit fewer misconceptions and take less time to complete a standard set of tasks.

Abstract: We have realized the help system SINIX Consultant (SC) for SINIX1 users. The system is capable of answering – in German – natural language questions concerning SINIX commands, objects, and concepts. But not only does this help system react to inquiries – additionally, the system is capable of activating itself. If the user employs a sequence of SINIX commands (a plan) in order to reach a specific goal, the help system proposes a sequence which reaches the same goal, but, with fewer commands. In this paper, a brief survey of the SINIX Consultant and the realized plan recognizer REPLIX is first given. Then, an initial attempt of a theoretical treatment of plan recognition is presented. This is done within the logical framework. We show how we can use an interval-based logic of time to describe actions, atomic plans, non-atomic plans, action execution, and simple plan recognition. We also show that the recognition of inserted sub-plans managed by REPLIX can be handled as well. Then, we present a problem which cannot be treated in the formalism. Thus, in this paper, we don't present a full developed theory but nevertheless, a step towards it.

Abstract: Intelligent help systems cannot merely respond passively to the user's commands and queries. They need to be able to volunteer information, correct user misconceptions, and reject unethical requests when appropriate. In order to do these things, a system must be designed as an intelligent agent. That is, a system needs to have its own goals and then plan for these goals. A system which did not have its own goals would never refuse to help users perform unethical actions. Such an intelligent agent has been implemented in the UCEgo component of UC (Wilensky et al. 1984s Wilensky et al. 1988) (UNIX Consultant), a natural language system that helps the user solve problems in using the UNIX operating system. UCEgo provides UC with its own goals and plans. By adopting different goals in different situations, UCEgo creates and executes different plans, enabling it to interact appropriately with the user. UCEgo adopts goals when it notices that the user either lacks necessary knowledge, or has incorrect beliefs. In these cases, UCEgo plans to volunteer information or correct the user's misconception as appropriate. These plans are pre-stored skeletal plans that are indexed under the types of situations in which they are typically useful. Plan suggestion situations include the goal which the plan is used to achieve, the preconditions of the plan, and appropriateness conditions for the plan. Indexing plans by situations improves efficiency and allows UC to respond appropriately to the user in real time. Detecting situations in which a plan should be suggested or a goal adopted is implemented using if-detected daemons. These daemons provide a single mechanism which can be used both for detecting goals and suggesting plans. Different methodologies for the efficient implementation of if-detected daemons are discussed.

Abstract: Novice unix users have many incorrect beliefs about unix commands. An intelligent advisory system for unix should provide explanatory responses that correct these mistaken beliefs. To do so, the system must be able to understand how the user is justifying these beliefs, and it must be able to provide justifications for its own beliefs. These tasks not only require knowledge about specific unix-related plans but also abstract knowledge about how beliefs can be justified. This paper shows how this knowledge can be represented and sketches how it can be used to form justifications for advisor beliefs and to understand justifications given for user beliefs. Knowledge about belief justification is captured by justification patterns, domain-independent knowledge structures that are similar to the abstract knowledge structures used to understand the point behind a story. These justification patterns allow the advisor to understand and formulate novel belief justifications, giving the advisor the ability to recognize and respond to novel misconceptions.

Abstract: This is the last of a series of three special issues on intelligent help systems for UNIX.1 This issue addresses natural language dialogue whereas the previous issues focussed on computational models and systems and planning and knowledge representation, respectively. In this collection, the focus is not upon issues of parsing and production per se, even though several of the systems described here in have significant capacities in this dimension. Instead, work here has evolved more intimately within the context of consultation systems, a topic seldom dealt with by other natural language systems. Nevertheless, the issues discussed are of general concern within natural language processing.

Abstract: This is the second of a series of three special issues on intelligent help systems for UNIX.1 This issue addresses planning and knowledge representation whereas the first issue focussed on computational models and systems and the next will be on natural language dialogue. The papers in this issue are concerned with discovering what the user wants to do, and figuring out a way to do it as well as representing the knowledge needed to do so.

Abstract: This is the first in a series of three special issues focussed on intelligent help systems for UNIX, each with its own emphasis: (1) computational models and systems, (2) planning and knowledge representation and (3) natural language dialogue. In this first issue focussing on computational models and systems there are five papers, one addressing empirical foundations, another virtues and problems, with the final three describing comprehensive implemented systems.

Abstract: We describe CHAMP (CHurn Analysis, Modeling, and Prediction), an automated system for modeling cellular subscriber churn that is predicting which customers will discontinue cellular phone service. We describe various issues related to developing and deploying this system including automating data access from a remote data warehouse, preprocessing, feature selection, model validation, and optimization to reflect business tradeoffs. Using data from GTE's data warehouse for cellular phone customers, CHAMP is capable of developing churn models customized by region for over one hundred GTE cellular phone markets totaling over 5 million customers. Every month churn factors are identified for each geographic region and models are updated to generate churn scores predicting who is likely to churn in the short term. Learning methods such as decision trees and genetic algorithms are used for feature selection and a cascade neural network is used for predicting churn scores. In addition to producing churn scores, CHAMP also produces qualitative results in the form of rules and comparison of market trends that are disseminated through a web based interface.

Abstract: In this paper we describe a data mining framework for constructing intrusion detection models. The first key idea is to mine system audit data for consistent and useful patterns of program and user behavior. The other is to use the set of relevant system features presented in the patterns to compute inductively learned classifiers that can recognize anomalies and known intrusions. In order for the classifiers to be effective intrusion detection models, we need to have sufficient audit data for training and also select a set of predictive system features. We propose to use the association rules and frequent episodes computed from audit data as the basis for guiding the audit data gathering and feature selection processes. We modify these two basic algorithms to use axis attribute(s) and reference attribute(s) as forms of item constraints to compute only the relevant patterns. In addition, we use an iterative level-wise approximate mining procedure to uncover the low frequency but important patterns. We use meta-learning as a mechanism to make intrusion detection models more effective and adaptive. We report our extensive experiments in using our framework on real-world audit data.

Abstract: ``Customer Retention'' is an increasingly pressing issue in today's ever-competitive commercial arena. This is especially relevant and important for sales and services related industries. Motivated by a real-world problem faced by a large company, we proposed a solution that integrates various techniques of data mining, such as feature selection via induction, deviation analysis, and mining multiple concept-level association rules to form an intuitive and novel approach to gauging customer loyalty and predicting their likelihood of defection. Immediate action triggered by these ``early-warnings'' resulting from data mining is often the key to eventual customer retention.