Advances in Case-Based Reasoning

Knowledge-intensive CBR assumes that cases are enriched with general domain knowledge. In CREEK, there is a very strong coupling between cases and general domain knowledge, in that cases are embedded within a general domain model. This increases the knowledge-intensiveness of the cases themselves. A knowledge-intensive CBR method calls for powerful knowledge acquisition and modeling techniques, as well as machine learning methods that take advantage of the general knowledge represented in the system. The focusing theme of the paper is on cases as knowledge within a knowledge-intensive CBR method. This is made concrete by relating it to the CREEK architecture and system, both in general terms, and through a set of example projects where various aspects of this theme have been studied.

The development of knowledge management applications for business environments requires balancing the needs of various parties within the client’s organization. While the end users desire to have a system that operates as effectively and efficiently as possible without changing their existing workflow, the business unit may be interested in capturing, verifying and endorsing corporate policies. The IT department of the client will be interested in applications that fit into their standard deployment environment and will certainly not appreciate the need for the business unit to modify some of the knowledge containers occasionally. Balancing these requirements without endangering the long term success of an application can become challenging.

Over the years, many successful applications of case-based reasoning (CBR) systems have been developed in different areas. The performance of CBR systems depends on several factors, including case representation, similarity measure, and adaptation. Achieving good performance requires careful design, implementation, and continuous optimization of these factors. In this paper, we propose a maintenance technique that integrates an ensemble of CBR classifiers with spectral clustering and logistic regression to improve the classification accuracy of CBR classifiers on (ultra) high-dimensional biological data sets.Our proposed method is applicable to any CBR system; however, in this paper, we demonstrate the improvement achieved by applying the method to a computational framework of a CBR system called $\mathit{TA3}$. We have evaluated the system on two publicly available microarray data sets that cover leukemia and lung cancer samples. Our maintenance method improves the classification accuracy of $\mathit{TA3}$ by approximately 20% from 65% to 79% for the leukemia and from 60% to 70% for the lung cancer data set.

We present an object-oriented framework in Java for building CBR systems that is an evolution of previous work on knowledge intensive CBR [8,9]. JColibri is a software artifact that promotes software reuse for building CBR systems, integrating the application of well proven Software Engineering techniques with a knowledge level description that separates the problem solving method, that defines the reasoning process, from the domain model, that describes the domain knowledge. Framework instantiation is supported by a graphical interface that guides the configuration of a particular CBR system, alleviating the steep learning curve typical for these type of systems.

Mémoire is a framework for the sharing and distribution of case bases and case based reasoning in biology and medicine. Based on the fact that semantics account for the success of biomedical case based reasoning systems, this paper defends the suitability of a semantic approach similar to the semantic Web for sharing and distributing case bases and case based reasoning in biology and medicine. Mémoire will permit to bridge the gap between the multiple case based reasoning systems dedicated to a single domain, and make available to agents and Web services on the Web the case based competency of the CBR systems adopting its interchange language. This paper presents the components of Mémoire for the representation of cases, case structure, and case based ontologies in biology and medicine. The approach could be extended to other application domains of CBR.

A fundamental problem for case-based reasoning systems is how to select relevant prior cases. Numerous strategies have been developed for determining the similarity of prior cases, given full descriptions of the problem at hand, and situation assessment methods have been developed for formulating appropriate initial case descriptions. However, in real-world applications, attempting to determine all relevant features of a new problem before retrieval may be impractical or impossible. Consequently, how to guide retrieval based on partial problem descriptions is an important question for CBR. This paper examines the problem of assessing similarity in partially-described cases. It proposes a set of similarity assessment strategies for handling missing information, evaluates their performance and efficiency on sample data sets, and discusses their tradeoffs.

Two key objectives of conversational case-based reasoning (CCBR) systems are (1) eliciting case facts in a manner that minimizes the user’s burden in terms of resources such as time, information cost, and cognitive load, and (2) integrating CBR with other problem solving modalities. This paper proposes an architecture that addresses both these goals by integrating CBR with a discourse-oriented dialogue engine. The dialogue engine determines when CBR or other problem-solving techniques are needed to achieve pending discourse goals. Conversely, the CBR component has the full resources of a dialogue engine to handle topic changes, interruptions, clarification questions by either the user or the system, and other speech acts that arise in problem-solving dialogues.

Hybrid recommender systems combine recommendation components of different types to achieve improved performance. Many such hybrids have been built but recent studies show that hybrids using case- based recommendation are rare. This paper shows how a range of different hybrids can be constructed using a case-based recommender as one component, and describes a series of experiments in which 20 different hybrids are built and evaluated. Cascade and feature augmentation hybrids are shown to have the highest accuracy over a range of different profile sizes.

The case-based reasoning (CBR) methodology can be augmented with the ability to determine the confidence in the correctness of individual solutions. A confidence calculation can be added to the REUSE portion of the CBR methodology. The confidence calculation takes confidence indicators, like “number of cases retrieved with best solution” and “average similarity of cases which suggest an alternative solution,” and generates a confidence value. The information gain algorithm C4.5 can be used to select the best confidence indicators by evaluating their usefulness in historical cases. A genetic algorithm can be used to optimize and maintain the confidence calculation.

As Case-Based Reasoning has matured as a discipline; the need for a standard means of representing case-based knowledge has come to the fore. While proposals exist for representing the vocabulary and the case-base knowledge containers, there are still no proposed standards for representing similarity or adaptation knowledge. In this paper we present extensions for representing similarity knowledge to CBML, an XML-based CBR language.

Because of the volume of spam email and its evolving nature, any deployed Machine Learning- based spam filtering system will need to have procedures for case-base maintenance. Key to this will be procedures to edit the case-base to remove noise and eliminate redundancy. In this paper we present a two stage process to do this. We present a new noise reduction algorithm called Blame-Based Noise Reduction that removes cases that are observed to cause misclassification. We also present an algorithm called Conservative Redundancy Reduction that is much less aggressive than the state-of-the-art alternatives and has significantly better generalisation performance in this domain. These new techniques are evaluated against the alternatives in the literature on four datasets of 1000 emails each (50% spam and 50% non spam).

Automatic construction of story plots has always been a longed-for utopian dream in the entertainment industry, especially in the more commercial genres that are fuelled by a large number of story plots with only a medium threshold on plot quality, such as TV series or video games. We propose a Knowledge Intensive CBR (KI-CBR) approach to the problem of generating story plots from a case base of existing stories analyzed in terms of Propp functions. A CBR process is defined to generate plots from a user query specifying an initial setting for the story, using an ontology to measure the semantical distance between words and structures taking part in the texts.

This paper is based on the observation that the nearest neighbour in a case-based prediction system may not be the best case to explain a prediction. This observation is based on the notion of a decision surface (i.e. class boundary) and the idea that cases located between the target case and the decision surface are more convincing as support for explanation. This motivates the idea of explanation utility, a metric that may be different to the similarity metric used for nearest neighbour retrieval. In this paper we present an explanation utility framework and present detailed examples of how it is used in two medical decision-support tasks. These examples show how this notion of explanation utility sometimes select cases other than the nearest neighbour for use in explanation and how these cases are more convincing as explanations.

The definition of similarity measures – one core component of every CBR application – leads to a serious knowledge acquisition problem if domain and application specific requirements have to be considered. To reduce the knowledge acquisition effort, different machine learning techniques have been developed in the past. In this paper, enhancements of our framework for learning knowledge-intensive similarity measures are presented. The described techniques aim to restrict the search space to be considered by the learning algorithm by exploiting available background knowledge. This helps to avoid typical problems of machine learning, such as overfitting the training data.

The complexity of software systems makes design reuse a necessary task in the software development process. CASE tools can provide cognitive assistance in this task, helping the software engineers to select designs to be reused. In this paper, we propose an approach for case indexing and retrieval based on Bayesian Networks, Case-Based Reasoning and WordNet. This approach is integrated in a CASE tool that reuses UML class diagrams, providing cognitive help for the software design phase.

Case adaptation has always been a difficult process to engineer within the case-based reasoning (CBR) cycle. To combat the difficulties of CBR adaptation, such as its domain dependency, computational cost and the inability to produce novel cases to solve new problems, genetic algorithms (GAs) have been applied to CBR adaptation. As the quality of cases stored in a case library has a significant effect on the solutions produced by a case-based reasoner, it is important to investigate the impact of the quality and quantity of cases injected into a GA initial population for adapting fitter solutions to new problems. This work explores a method applying a GA to CBR adaptation, where a learning mechanism is applied to feed knowledge back from the CBR revision stage into the reuse stage, allowing the GA to learn which mutations result in invalid solutions. In collaboration with this learning mechanism, the number of cases to be injected, and the fitness of cases to be injected from retrieval into reuse is explored. The fitness of adapted cases and their response to our developed learning feedback is also trialled through varying the size and quality of the GA initial population.

Taxonomic case retrieval systems significantly outperform standard conversational case retrieval systems. However, their feature taxonomies, which are the principal reason for their superior performance, must be manually developed. This is a laborious and error prone process. In an earlier paper, we proposed a framework for automatically acquiring features and organizing them into taxonomies to reduce the taxonomy acquisition effort. In this paper, we focus on the second part of this framework: automated feature organization. We introduce TAXIND, an algorithm for inducing taxonomies from a given set of features; it implements a step in our FACIT framework for knowledge extraction. TAXIND builds taxonomies using a novel bottom up procedure that operates on a matrix of asymmetric similarity values. We introduce measures for evaluating taxonomy induction performance and use them to evaluate TAXIND’s learning performance on two case bases. We investigate both a knowledge poor and a knowledge rich variant of TAXIND. While both outperform a baseline approach that does not induce taxonomies, there is no significant performance difference between the TAXIND variants. Finally, we discuss how a more comprehensive representation for features should improve measures on TAXIND’s learning and performance tasks.

Maintenance of Case-Based Reasoning (CBR) systems became an important area since applications of CBR technologies were established in different real-world domains. Maintenance issues cover all aspects that help to keep a running CBR system in a usable state of high quality. Concepts and techniques that were developed for maintenance of CBR systems range from methodologies and frameworks that particularly define phases, steps, and tasks necessary to integrate maintenance into the CBR process up to specific programs that enable CBR engineers to carry out the maintenance activities. In this paper, we exemplify this range of research on maintenance of CBR systems by brief characterizations of the Siam methodology, the MaMa maintenance manual, and the MaSh maintenance shell. The overall goal of this paper is then to conclude areas for further research in maintenance for CBR systems from the experience of the work on Siam, MaMa, MaSh, and related approaches.

The case-based reasoning approach to email response consists of reusing past messages to synthesize new responses to incoming requests. This task presents various challenges due to the nature of the messages: Textual descriptions, multiple topics, heterogeneous content, variable text length and varying recurrence of the statements. In this paper, we address the problem of determining which portions of past cases are reusable. Our scheme consists of identifying parts of a past message and declaring them variable, optional or reusable. This formulation of case reuse corresponds, from an application point of view, to the dynamic creation of a response template from antecedent messages. We describe and compare two strategies for selecting the messages portions to be reused: Case grouping and condensation models. Our results indicate that the case grouping strategy is a better choice. We also describe some of our experiments for identifying variable parts, based on named entity extraction techniques.

This paper describes ProCHiP, a planner that combines CBR with the techniques of decision-theoretic planning and HTN planning in order to deal with uncertain, dynamic large-scale real-world domains. We explain how plans are represented, generated and executed. Unlike in regular HTN planning, ProCHiP can generate plans in domains where there is no complete domain theory by using cases instead of methods for task decomposition. ProCHiP generates a variant of a HTN – a kind of AND/OR tree of probabilistic conditional tasks – that expresses all the possible ways to decompose an initial task network. As in Decision-Theoretic planning, the expected utility of alternative plans is computed, although in ProCHiP this happens beforehand at the time of building the HTN. ProCHiP is used by agents inhabiting multi-agent environments. We present an experiment carried out to evaluate the role of the size of the case-base on the performance of the planner. We verified that the CPU time increases monotonically with the case-base size while effectiveness is improved only up to a certain case-base size.

Exploration involves selecting and executing sequences of actions so that the knowledge of the environments is acquired. In this paper we address the problem of exploring unknown, dynamic environments populated with both static and non-static entities (objects and agents) by an autonomous agent. The agent has a case-base of entities and another of plans. This case-base of plans is used for a case-based generation of goals and plans for visiting the unknown entities or regions of the environment. The case-base of entities is used for a case-based generation of expectations for missing information in the agent’s perception. Both case- bases are continuously updated: the case-base of entities is updated as new entities are perceived or visited, while the case-base of plans is updated as new sequences of actions for visiting entities/regions are executed successfully. We present and discuss the results of an experiment conducted in a simulated environment in order to evaluate the role of the size of the case-base of entities on the performance of exploration.

Most CBR systems try to solve problems in one shot neglecting the sequential behavior of most real world domains and the simultaneous occurrence of interleaved problems proper to multi-agent settings. This article provides a first answer to the following question: how can the CBR paradigm be enriched to support the analysis of unsegmented sequences of observational data stemming from multiple coincidental sources? We propose Ceaseless CBR, a new model that considers the CBR task as on-going rather than one-shot and aims at finding the best explanation of an unsegmented sequence of alerts with the purpose of pinpointing whether undesired situations have occurred or not and, if so, indicating the multiple responsible sources or at least which ones are the most plausible.

Case-based Reasoning (CBR) is a mature technology for building knowledge-based systems that are capable to produce useful results even if no answer matches the query exactly. Often the result sets presented to users are ordered by means of similarity and utility. However, for complex applications with knowledge intensive domains we have discovered that results sets enriched by calculated similarity values for particular answers are not sufficient. Users have a demand for additional information and explanations making the proposed results more transparent. By presenting additional explanations to them, their confidence in the result set increases and possible deficiencies, e.g., in the weight model, can be revealed and corrected. This paper presents a realized explanation service that combines several existing and new explanation technologies into one system.

We begin by examining the limitations of precedent-based explanations of the predicted outcome in case-based reasoning (CBR) approaches to classification and diagnosis. By failing to distinguish between features that support and oppose the predicted outcome, we argue, such explanations are not only less informative than might be expected, but also potentially misleading. To address this issue, we present an evidential approach to explanation in which a key role is played by techniques for the discovery of features that support or oppose the predicted outcome. Often in assessing the evidence provided by a continuous attribute, the problem is where to “draw the line” between values that support and oppose the predicted outcome. Our approach to the selection of such an evidence threshold is based on the weights of evidence provided by values above and below the threshold. Examples used to illustrate our evidential approach to explanation include a prototype CBR system for predicting whether or not a person is over the legal blood alcohol limit for driving based on attributes such as units of alcohol consumed.

Increasingly in case-based reasoning (CBR) approaches to product recommendation, some or all of the user’s requirements are treated, at least initially, as constraints that the retrieved cases must satisfy. We present a mixed-initiative approach to recovery from the retrieval failures that occur when there is no case that satisfies all the user’s requirements. The recovery process begins with an explanation of the retrieval failure in which the user’s attention is drawn to combinations of constraints in her query for which there are no matching cases. The user is then guided in the selection of the most useful attribute, and associated constraint, to be eliminated from her query at each stage of an incremental relaxation process. If not prepared to compromise on the attribute suggested for elimination at any stage, the user can select another attribute to be eliminated. On successful completion of the recovery process, the retrieved cases involve only compromises that the user has chosen, in principle, to accept.

A CBR system needs a good case retention strategy to decide which cases to incorporate into the case base in order to maximize the performance of the system. In this work we present a collaborative case retention strategy, designed for multiagent CBR systems, called the Collaborative Case Bargaining strategy. The CCB strategy is a bargaining mechanism in which each CBR agent tries to maximize the utility of the cases it retains. We will present a case utility measure called the Justification-based Case Utility (JCU) based upon the ability of the individual CBR agents to provide justifications of their own results. An empirical evaluation of the CCB strategy shows the benefits for CBR agents to use this strategy: individual and collective accuracy are increased while the size of the case bases is decreased.

Good similarity functions are at the heart of effective case-based reasoning. However, the similarity functions that have been designed so far have been mostly linear, weighted-sum in nature. In this paper, we explore how to handle case retrieval when the case base is nonlinear in similarity measurement, in which situation the linear similarity functions will result in the wrong solutions. Our approach is to first transform the case base into a feature space using kernel computation. We perform correlation analysis with maximum correlation criterion(MCC) in the feature space to find the most important features through which we construct a feature-space case base. We then solve the new case in the feature space using the traditional similarity-based retrieval. We show that for nonlinear case bases, our method results in a performance gain by a large margin. We show the theoretical foundation and empirical evaluation to support our observations.

Model-based object recognition is a well-known task in Computer Vision. Usually, one object that can be generalized by a model should be detected in an image based on this model. Biomedical applications have the special quality that one object can have a great variation in appearance. Therefore the appearance of this object cannot be generalized by one model. A set of cases of the appearance of this object (sometimes 50 cases or more) is necessary to detect this object in an image. The recognition method is rather a case-based object recognition than a model-based object recognition. Case-based object recognition is a challenging task. It puts special requirements to the similarity measure and needs a matching algorithm that can work fast on a large number of cases. In this paper we describe the chosen case representation, the similarity measure and the recent matching algorithm. Finally, we give results on the performance of the system.

By design, Case-Based Reasoning (CBR) systems do not need deep general knowledge. In contrast to (rule-based) expert systems, CBR systems can already be used with just some initial knowledge. Further knowledge can then be added manually or learned over time. CBR systems are not addressing a special group of users. Expert systems, on the other hand, are intended to solve problems similar to human experts. Because of the complexity and difficulty of building and using expert systems, research in this area addressed generating explanations right from the beginning. But for knowledge-intensive CBR applications, the demand for explanations is also growing. This paper is a first pass on examining issues concerning explanations produced by CBR systems from the knowledge containers perspective. It discusses what naturally can be explained by each of the four knowledge containers (vocabulary, similarity measures, adaptation knowledge, and case base) in relation to scientific, conceptual, and cognitive explanations.

The concept of generalized cases has been proven useful when searching for configurable and flexible products, for instance, reusable components in the area of electronic design automation. This paper addresses the similarity assessment and retrieval problem for case bases consisting of traditional and generalized cases. While approaches presented earlier were restricted to continuous domains, this paper addresses generalized cases defined over mixed, continuous and discrete, domains. It extends the view on the similarity assessment as a nonlinear optimization problem (NLP) towards a mixed integer nonlinear optimization problem (MINLP), which is an actual research topic in mathematical optimization. This is an important step because most real world applications require mixed domains for the case description. Furthermore, we introduce two optimization-based retrieval methods that operate on a previously created index structure, which restricts the retrieval response time significantly.

An application of relational case-based learning to the task of expressive music performance is presented. We briefly recapitulate the relational case-based learner DISTALL and empirically show that DISTALL outperforms a straightforward propositional k-NN on the music task. A set distance measure based on maximal matching – incorporated in DISTALL – is discussed in more detail and especially the problem associated with its ‘penalty part’: the distance between a large and a small set is mainly determined by their difference in cardinality. We introduce a method for systematically varying the influence of the penalty on the overall distance measure and experimentally test different variants of it. Interestingly, it turns out that the variants with high influence of penalty clearly perform better than the others on our music task.

In this paper we propose an architecture for an adaptive workflow management system (WFMS) and present the research prototype CBRFlow. CBRFlow extends workflow execution with conversational case-based reasoning (CCBR) to adapt the predefined workflow model to changing circumstances and to provide the WFMS with learning capabilities. Business rules within the predefined workflow model are annotated during run-time with context-specific information in the form of cases using the CCBR sub-system. When case reuse becomes frequent, the cases are manually refactored into rules to foster automatic execution. This feedback supports continuous process improvement, resulting in more manageable and more efficient business processes over time.

We present in this paper a hybrid neuro-symbolic system called “CASEP2”, which combines the case-based reasoning with an adequate artificial neural network “M-SOM-ART” for sequence classification or prediction task. In CASEP2, we present a new case modelling by dynamic covariance matrices. This model takes into account the temporal dynamics contained in the sequences and allows to avoid problems related to the comparison of different length sequences. In the CBR cycle, one neural network is used during the retrieval phase for indexing the case base and another is used during the reuse phase in order to provide the target case solution.

The development of industrial case-based reasoning systems that have to operate within a continually evolving environment, is a challenging problem. Industrial applications require of robust and competent systems. When the problem domain is evolving, the solutions provided by the system can easily become wrong. In this paper we present an algorithm for dealing with real-world domains where case solutions are evolving along the time. Specifically, the algorithm deals with what we call the innovation problem: the continuous improvements on the components that are part of case solutions. We will show how the use of the proposed algorithm improves significantly the quality of solutions in a deployed engineering design system.

When events such as severe weather or congestion interfere with the normal flow of air traffic, air traffic controllers may implement plays that reroute one or more traffic flows. Currently, plays are assessed and selected based on controllers’ experience using the National Playbook, a collection of plays that have worked in the past. This paper introduces PlayMaker, a CBR prototype that replicates the Playbook and models how controllers select plays. This paper describes the PlayMaker design, a model validation, and discusses developments necessary for a full-scale CBR tool for this application.

Web search is typically memory-less, in the sense that each new search query is considered afresh and ‘solved’ from scratch. We believe that this reflects the strong information retrieval bias that has influenced the development of Web search engines. In this paper we argue for the value of a fresh approach to Web search, one that is founded on the notion of reuse and that seeks to exploit past search histories to answer future search queries. We describe a novel case-based technique and evaluate it using live-user data. We show that it can deliver significant performance benefits when compared to alternative strategies including meta-search.

This paper describes P–Truck Curing, a Case Based Reasoning system supporting the design of the curing phase for truck tyre production. The design of this process provides a trade–off between an optimal curing degree, to avoid imperfections in the final product, and the reduction of costs, related to thermal energy employed in the curing. Expert curing process designers store information about past episodes and exploit it to define new ones, without starting from scratch. A CBR system is thus a suitable approach to model this problem solving method: case structure, similarity and adaptation functions and a general system overview will be described. This work has been developed in the context of the P–Truck project, whose goal is the development of an integrated Knowledge Management (KM) system to support the Business Unit Truck of Pirelli Tyres in the design and manufacture of truck tyres.

Traditional search techniques frequently fail the average user in their quest for online information. Recommender systems attempt to address this problem by discovering the context in which the search occurs. Though effective, these systems are often hampered by the brevity of typical user queries. In this paper we describe CASPER, an online recruitment search engine which combines similarity-based search with a client-side personalisation technique. In particular we argue thatCASPER’s personalisation strategy is effective in determining retrieval relevance in the face of incomplete queries.

By improving accuracy in the quantification of the ocean’s CO₂ budget, a more precise estimation can be made of the terrestrial fraction of global CO₂ budget and its subsequent effect on climate change. First steps have been taken towards this from an environmental and economic point of view, by using an instance based reasoning system, which incorporates a novel clustering and retrieval method – a Cooperative Maximum Likelihood Hebbian Learning model (CoHeL). This paper reviews the problems of measuring the ocean’s CO₂ budget and presents the CoHeL model developed and outlines the IBR system developed to resolve the problem.

In this paper we present an agent-based application of a wireless tourist guide that combines the Beliefs-Desires-Intentions approach with learning capabilities of Case Base Reasoning techniques. This application shows how to develop adaptive agents with a goal driven design and a decision process built on a CBR architecture. The resulting agent architecture has been validated by real users who have used the tourist guide application, on a mobile device, and can be generalized for the development of other personalized services.

The ranking of offers is an issue in e-commerce that has received a lot of attention in Case-Based Reasoning research. In the absence of a sales assistant, it is important to provide a facility that will bring suitable products and services to the attention of the customer. In this paper we present such a facility that is part of a Personal Travel Assistant (PTA) for booking flights online. The PTA returns a large number of offers (24 on average) and it is important to rank them to bring the most suitable to the fore. This ranking is done based on similarity to previously accepted offers. It is a characteristic of this domain that the case-base of accepted offers will be small, so the learning of appropriate feature weights is a particular challenge. We describe a process for learning personalised feature weights and present an evaluation that shows its effectiveness.

Textual case-based reasoning (TCBR) provides the ability to reason with domain-specific knowledge when experiences exist in text. Ideally, we would like to find an inexpensive way to automatically, efficiently, and accurately represent textual documents as cases. One of the challenges, however, is that current automated methods that manipulate text are not always useful because they are either expensive (based on natural language processing) or they do not take into account word order and negation (based on statistics) when interpreting textual sources. Recently, Schenker et al. [1] introduced an algorithm to convert textual documents into graphs that conserves and conveys the order and structure of the source text in the graph representation. Unfortunately, the resulting graphs cannot be used as cases because they do not take domain knowledge into consideration. Thus, the goal of this study is to investigate the potential benefit, if any, of this new algorithm to TCBR. For this purpose, we conducted an experiment to evaluate variations of the algorithm for TCBR. We discuss the potential contribution of this algorithm to existing TCBR approaches.

Although Case-based Reasoning is supposed to alleviate the well known knowledge acquisition bottleneck for knowledge-based systems, case acquisition remains an expensive process. In this paper we present a semiautomatic methodology for building an ontology-based organization of the Case Base and to populate it with cases extracted from structured documents. The methodology is analyzed through the case study of a help desk system.

In this paper, we describe a CBR system for applying musically acceptable tempo transformations to monophonic audio recordings of musical performances. Within the tempo transformation process, the expressivity of the performance is adjusted in such a way that the result sounds natural for the new tempo. A case base of previously performed melodies is used to infer the appropriate expressivity. Tempo transformation is one of the audio post-processing tasks manually done in audio-labs. Automatizing this process may, therefore, be of industrial interest.

Model-based image recognition requires a general model of the object that should be detected in an image. In many applications such models are not known a-priori instead of they must be learnt from examples. Real world applications such as the recognition of biological objects in images cannot be solved by one general model but a lot of different models are necessary in order to handle the natural variations of the appearance of the objects of a certain class. Therefore we are talking about case-based object recognition. In this paper we describe how the shape of an object can be extracted from images and input into a case description. These acquired cases we mine for more general shapes so that at the end a case base of shapes can be constructed and applied for case-based object recognition.

Most project-based industries such as construction, shipbuilding, and software development etc. should generate and manage project network for successful project planning. We suggest a set of criteria of good project network generator such as network generation efficiency, quality of network, and economics of system development. For the efficiency of the planning, the first criterion, we decided to take a CBR approach. However, using only previous cases is insufficient to generate a proper network for a new project. By embedding rules and constraints in the case-based system, we could improve the quality of the project network: the second criterion. The integration of CBR approach and the knowledge-based approach makes feasible the development of the project network generator and improves the quality of the network by mutual enhancement through crosschecking the knowledge and cases in the development and maintenance stages. For some complex project network planning, a single-case assumed project network generation methodology is refined into Dynamic Leveled Multiple Case approach. The methodology contributes again the efficiency and effectiveness of project network generation and reduces the efforts of the system development.

In the steel industry, quality designing is related to the determination of mechanical properties of the final products and operational conditions according to the specifications that a customer requests. It involves the utilization of metallurgical knowledge and field experience in the industry. On the other hand, the production scheduling for steel making is a large-scale, multi-objective, grouping and sequencing problem with various restrictions. Traditionally, these two problems have been handled separately. However, the rapid development of information techniques has enabled the simultaneous solution of these two problems. In this paper, we develop an integrated case based reasoning framework for quality designing and scheduling. As proposed, the case base is established with proper case representation scheme, similar cases are retrieved and selected using fuzzy techniques, and finally the selected cases are put into the production process using the scheduling technique. The experimental results show good performance to the quality designing and scheduling of steel products. The framework developed is expected to be applied to other process industries.

In this paper, we present a case-based retrieval system called Rhene (Retrieval of HEmodialysis in NEphrological disorders) working in the domain of patients affected by nephropatologies and treated with hemodialysis. Defining a dialysis session as a case, retrieval of past similar cases has to operate both on static and on dynamic (time-dependent) features, since most of the monitoring variables of a dialysis session are time series. In Rhene, retrieval relies upon a multi-step procedure. In particular, a preliminary grouping/classification step, based on static features, reduces the retrieval search space. Intra-class retrieval then takes place by considering time-dependent features, and is articulated as follows: (1) “locally” similar cases (considering one feature at a time) are extracted and the intersection of the retrieved sets is computed; (2) “global” similarity is computed – as a weighted average of local distances – and the best cases are listed. The main goal of the paper is to present an approach for efficiently implementing step (2), by taking into account specific information regarding the final application. We concentrate on a classical dimensionality reduction technique for time series allowing for efficient indexing, namely Discrete Fourier Transform (DFT). Thanks to specific index structures (i.e. k-d trees) range queries (on local feature similarity) can be efficiently performed on our case base; as mentioned above, results of such local queries are then suitably combined, allowing the physician to examine the most similar stored dialysis sessions with respect to the current one and to assess the quality of the overall hemodialysis service.

Respiratory Sinus Arrhythmia has until now been analysed manually by reviewing long time series of heart rate measurements. Patterns are identified in the analysis of the measurements. We propose a design for a classification system of Respiratory Sinus Arrhythmia by time series analysis of heart and respiration measurements. The classification uses Case-Based Reasoning and Rule-Based Reasoning in a Multi-Modal architecture. The system is in use as a research tool in psychophysiological medicine, and will be available as a decision support system for treatment personnel.

In industrial manufacturing rigorous testing is used to ensure that the delivered products meet their specifications. Mechanical maladjustment or faults often show their presence through abnormal acoustic signals. This is the same case in robot assembly – the application domain addressed in this paper. Manual diagnosis based on sound requires extensive experience, and usually such experience is acquired at the cost of reduced production efficiency or degraded product quality due to mistakes in judgments. The acquired experience is also difficult to preserve and transfer and it often gets lost if the corresponding personnel leave the task of testing. We propose herein a Case-Based Reasoning approach to collect, preserve and reuse the available experience for robot diagnosis. This solution enables fast experience transfer and more reliable and informed testing. Sounds from normal and faulty robots are recorded and stored in a case library together with their diagnosis results. Given an unclassified sound signal, the relevant cases are retrieved from the case library as reference for deciding the fault class of the new case. Adding new classified sound profiles to the case library improves the system’s performance. So far the developed system has been applied to the testing environment for industrial robots. The preliminary results demonstrate that our system is valuable in this application scenario in that it can preserve and transfer the related experience among technicians and shortens the overall testing time.

Advances in technology for digital image capture and storage have caused an information overload problem in the geo-sciences. This has compounded existing image retrieval problems whereby most image matching is performed using content-based image retrieval techniques. The biggest problem in this field is the so-called semantic gap – the mismatch between the capabilities of current content-based image retrieval systems and the user needs. One way of addressing this problem is to develop context-based image retrieval methods. Context-based retrieval relies on knowledge about why image contents are important in a particular area and how specific images have been used to address particular tasks. We are developing a case-based knowledge- management retrieval system that employs a task-centric approach to capturing and reusing user context. This is achieved through image annotation and adaptive content presentation. In this paper we present an extension of a previous implementation of our approach and a thorough evaluation of our application.

Initial successes in the area of recommender systems have led to considerable early optimism. However as a research community, we are still in the early days of our understanding of these applications and their capabilities. Evaluation metrics continue to be refined but we still need to account for the relative contributions of the various knowledge elements that play a part in the recommendation process. In this paper, we make a fine-grained analysis of a successful case-based recommendation approach, providing an ablation study of similarity knowledge and similarity metric contributions to improved system performance. In particular, we extend our earlier analyses to examine how measures of interestingness can be used to identify and analyse relative contributions of segments of similarity knowledge. We gauge the strengths and weaknesses of knowledge components and discuss future work as well as implications for research in the area.

Satellite health monitoring is a specialized task usually carried out by human experts. In this paper, we address the task of monitoring by defining it as an anomaly and event detection task cast in Case Based Reasoning framework. We discuss how each CBR step is achieved in a time series domain such as the Satellite health monitoring. In the process, we define the case structure in a time series domain, discuss measures of distance between cases and address other issues such as building initial Case Base and determining similarity threshold. We briefly describe the system that we have built, and end the paper with a discussion on possible extensions to current work.

There are plenty of methods proposed for analog electronic circuit diagnosis, but the most popular ones are the fault dictionary techniques. Admitting more cases in a fault dictionary can be seen as a natural development towards a CBR system. The proposal of this paper is to extend the fault dictionary towards a Case Based Reasoning system. The case base memory, retrieval, reuse, revise and retain tasks are described. Special attention to the learning process is taken. An application example on a biquadratic filter is shown. The faults considered are parametric, permanent, independent and simple, although the methodology could be extrapolated for catastrophic and multiple fault diagnosis. Also, the method is focused and tested only on passive faulty components. Nevertheless, it can be extended to cover active devices as well.

Critiquing is a powerful style of feedback for case-based recommender systems. Instead of providing detailed feature values, users indicate a directional preference for a feature. For example, a user might ask for a ‘less expensive’ restaurant in a restaurant recommender; ‘less expensive’ is a critique over the price feature. The value of critiquing is that it is generally applicable over a wide range of domains and it is an effective means of focusing search. To date critiquing approaches have usually been limited to single-feature critiques, and this ultimately limits the degree to which a given critique can eliminate unsuitable cases. In this paper we propose extending the critiquing concept to cater for the possibility of compound critiques – critiques over multiple case features. We describe a technique for automatically generating useful compound critiques and demonstrate how this can significantly improve the performance of a conversational recommender system. We also argue that this generalised form of critiquing offers explanatory benefits by helping the user to better understand the structure of the recommendation space.

Helping software designers in their task implies the development of tools with intelligent capabilities. One such capability is the integration of natural language understanding in CASE tools, thus improving the designer/tool communication. In this paper, we present a CBR approach for the generation of UML class diagrams from natural language text. This approach is implemented in a CASE tool, with the goal of helping the software designer create the first system model. We describe the natural language conversion module and give an overview of the tool in which it is integrated. Experimental results for retrieval and adaptation mechanisms are also presented.

This paper focuses on one critical element, indexing – retaining and representing knowledge in an applied case-based reasoning (CBR) model for supporting strategic manufacturing vision development (CBRM). Manufacturing vision (MV) is a kind of knowledge management concept and process concerned with the competence improvement of an enterprise’s manufacturing system. There are two types of cases within the CBRM – an event case (EC) and a general supportive case (GSC). We designed one set of indexing vocabulary for the two types of cases, but a different indexing representation structure for each of them. In this paper, after the background introduction of the MV, the CBRM and the indexing challenges of the MV cases, we present the structure and content of the index vocabulary and the two indexing representation structures, then illustrate briefly the indexing of cases with two examples. We also summarize the methods, primary conclusions of test runs with the indexing scheme. Further research work to refine the index vocabulary is discussed as well.

Textual CBR systems solve problems by reusing experiences that are in textual form. Knowledge-rich comparison of textual cases remains an important challenge for these systems. However mapping text data into a structured case representation requires a significant knowledge engineering effort. In this paper we look at automated acquisition of the case indexing vocabulary as a two step process involving feature selection followed by feature generalisation. Boosted decision stumps are employed as a means to select features that are predictive and relatively orthogonal. Association rule induction is employed to capture feature co-occurrence patterns. Generalised features are constructed by applying these rules. Essentially, rules preserve implicit semantic relationships between features and applying them has the desired effect of bringing together cases that would have otherwise been overlooked during case retrieval. Experiments with four textual data sets show significant improvement in retrieval accuracy whenever generalised features are used. The results further suggest that boosted decision stumps with generalised features to be a promising combination.