Search

Copyright of JMD: Journal für Mathematik-Didaktik is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In this paper, we explore different possible explanations for research misconduct (especially falsification and fabrication), and investigate whether they are compatible. We suggest that to explain research misconduct, we should pay attention to three factors: (1) the beliefs and desires of the misconductor, (2) contextual affordances, (3) and unconscious biases or influences. We draw on the three different narratives (individual, institutional, system of science) of research misconduct as proposed by Sovacool to review six different explanations. Four theories start from the individual: Rational Choice theory, Bad Apple theory, General Strain Theory and Prospect Theory. Organizational Justice Theory focuses on institutional factors, while New Public Management targets the system of science. For each theory, we illustrate the kinds of facts that must be known in order for explanations based on them to have minimal plausibility. We suggest that none can constitute a full explanation. Finally, we explore how the different possible explanations interrelate. We find that they are compatible, with the exception of explanations based on Rational Choice Theory and Prospect Theory respectively, which are incompatible with one another. For illustrative purposes we examine the case of Diederik Stapel. [ABSTRACT FROM AUTHOR]

This paper discusses a class of mechanistic explanations employed in engineering science where the activities and organization of nonstandard entities (voids, cracks, pits…) are cited as core factors responsible for failures. Given the use of mechanistic language by engineers and the manifestly mechanistic structure of these explanations, I consider several interpretations of these explanations within the new mechanical framework (among others: voids should be considered as shorthand expressions for other entities, voids should be reduced to lower-level mechanisms, or the explanations are simply abstract mechanistic explanations). I argue that these interpretations fail to solve several philosophical problems and propose an account of fictional mechanism explanations instead. According to this account, fictional mechanism explanations provide descriptions of fictional mechanisms that enable the tracking of counterfactual dependencies of the physical system they model by capturing system constraints. Engineers use these models to learn about and understand properties of materials, to build computational simulations of their behaviour, and to design new materials. [ABSTRACT FROM AUTHOR]

In this paper we argue that transparency of machine learning algorithms, just as explanation, can be defined at different levels of abstraction. We criticize recent attempts to identify the explanation of black box algorithms with making their decisions (post-hoc) interpretable, focusing our discussion on counterfactual explanations. These approaches to explanation simplify the real nature of the black boxes and risk misleading the public about the normative features of a model. We propose a new form of algorithmic transparency, that consists in explaining algorithms as an intentional product, that serves a particular goal, or multiple goals (Daniel Dennet's design stance) in a given domain of applicability, and that provides a measure of the extent to which such a goal is achieved, and evidence about the way that measure has been reached. We call such idea of algorithmic transparency "design publicity." We argue that design publicity can be more easily linked with the justification of the use and of the design of the algorithm, and of each individual decision following from it. In comparison to post-hoc explanations of individual algorithmic decisions, design publicity meets a different demand (the demand for impersonal justification) of the explainee. Finally, we argue that when models that pursue justifiable goals (which may include fairness as avoidance of bias towards specific groups) to a justifiable degree are used consistently, the resulting decisions are all justified even if some of them are (unavoidably) based on incorrect predictions. For this argument, we rely on John Rawls's idea of procedural justice applied to algorithms conceived as institutions. [ABSTRACT FROM AUTHOR]

The offline evaluation of recommender systems is typically based on accuracy metrics such as the Mean Absolute Error and the Root Mean Squared Error for error rating prediction and Precision and Recall for measuring the quality of the top-N recommendations. However, it is difficult to reproduce the results since there are various libraries that can be used for running experiments and also within the same library there are many different settings that if not taken into consideration when replicating the results might vary. In this paper, we show that within the use of the same library an explanation-based approach can be used to assist in the reproducibility of experiments. Our proposed approach has been experimentally evaluated using a wide range of recommendation algorithms ranging from collaborative filtering to complicated fuzzy recommendation approaches that can solve the filter bubble problem, a real dataset, and the results show that it is both practical and effective. [ABSTRACT FROM AUTHOR]

The need for transparency of predictive systems based on Machine Learning algorithms arises as a consequence of their ever-increasing proliferation in the industry. Whenever black-box algorithmic predictions influence human affairs, the inner workings of these algorithms should be scrutinised and their decisions explained to the relevant stakeholders, including the system engineers, the system's operators and the individuals whose case is being decided. While a variety of interpretability and explainability methods is available, none of them is a panacea that can satisfy all diverse expectations and competing objectives that might be required by the parties involved. We address this challenge in this paper by discussing the promises of Interactive Machine Learning for improved transparency of black-box systems using the example of contrastive explanations—a state-of-the-art approach to Interpretable Machine Learning. Specifically, we show how to personalise counterfactual explanations by interactively adjusting their conditional statements and extract additional explanations by asking follow-up "What if?" questions. Our experience in building, deploying and presenting this type of system allowed us to list desired properties as well as potential limitations, which can be used to guide the development of interactive explainers. While customising the medium of interaction, i.e., the user interface comprising of various communication channels, may give an impression of personalisation, we argue that adjusting the explanation itself and its content is more important. To this end, properties such as breadth, scope, context, purpose and target of the explanation have to be considered, in addition to explicitly informing the explainee about its limitations and caveats. Furthermore, we discuss the challenges of mirroring the explainee's mental model, which is the main building block of intelligible human–machine interactions. We also deliberate on the risks of allowing the explainee to freely manipulate the explanations and thereby extracting information about the underlying predictive model, which might be leveraged by malicious actors to steal or game the model. Finally, building an end-to-end interactive explainability system is a challenging engineering task; unless the main goal is its deployment, we recommend "Wizard of Oz" studies as a proxy for testing and evaluating standalone interactive explainability algorithms. [ABSTRACT FROM AUTHOR]

People often offer an excuse or an apology after they do something wrong. In this paper, we examine how giving an excuse, an apology, or no explanation after arriving late to a meeting influences the attitudes and behavioral intentions others form toward the late arrival. Additionally, we examined how a group-related factor (complaining) and the late arrival's history with coming late affected participant judgments. Across two studies using complementary experimental and survey methods, we found that an excuse is better than no explanation, but that the difference between apology and no explanation and apology and excuse is not always clear. Furthermore, we found that common distinctions between explanation types used in the literature may not fully exist in non-laboratory social interactions. Implications of these findings and future directions are discussed. [ABSTRACT FROM AUTHOR]

This paper makes two contributions towards enabling a robot to provide explanatory descriptions of its decisions, the underlying knowledge and beliefs, and the experiences that informed these beliefs. First, we present a theory of explanations comprising (i) claims about representing, reasoning with, and learning domain knowledge to support the construction of explanations; (ii) three fundamental axes to characterize explanations; and (iii) a methodology for constructing these explanations. Second, we describe an architecture for robots that implements this theory and supports scalability to complex domains and explanations. We demonstrate the architecture's capabilities in the context of a simulated robot (a) moving target objects to desired locations or people; or (b) following recipes to bake biscuits. [ABSTRACT FROM AUTHOR]

The reuse of Learning Designs can bring significant advantages to the educational community such as the diffusion of best teaching practices and the improvement of teaching quality and learning outcomes. Although various tools, including Recommender Systems, have been developed to implement the notion of reusing Learning Designs, their adoption by teachers falls short of expectations. This paper investigates the results of providing explanations for Learning Design recommendations to teachers. To this end, we designed and implemented an explanatory mechanism incorporated into a Recommender System, which propose pre-existing Learning Designs to teachers. We then conducted a user-centric evaluation experiment. Overall, this study provides evidence that explanations should be incorporated into Recommender Systems that propose Learning Designs, as a way of improving the teacher-perceived experience and promoting their wider adoption by teachers. The more teachers accept and adopt Recommender Systems that propose Learning Designs, the more the educational community gains the benefits of reusing Learning Designs. [ABSTRACT FROM AUTHOR]

Even with the lack of consensus on the nature of an argument, the thesis that explanations and arguments are distinct is near orthodoxy in well-known critical thinking texts and in the more advanced argumentation literature. In this paper, I reconstruct two rationales for distinguishing arguments from explanations. According to one, arguments and explanations are essentially different things because they have different structures. According to the other, while some explanations and arguments may have the same structure, they are different things because explanations are used for different purposes than arguments. I argue that both rationales fail to motivate a distinction between arguments and explanations. Since these are the only rationales for distinguishing arguments from explanations that I am prepared to take seriously, I don't see why we should exclude explanations from being arguments. [ABSTRACT FROM AUTHOR]

When recommender systems present items, these can be accompanied by explanatory information. Such explanations can serve seven aims: effectiveness, satisfaction, transparency, scrutability, trust, persuasiveness, and efficiency. These aims can be incompatible, so any evaluation needs to state which aim is being investigated and use appropriate metrics. This paper focuses particularly on effectiveness (helping users to make good decisions) and its trade-off with satisfaction. It provides an overview of existing work on evaluating effectiveness and the metrics used. It also highlights the limitations of the existing effectiveness metrics, in particular the effects of under- and overestimation and recommendation domain. In addition to this methodological contribution, the paper presents four empirical studies in two domains: movies and cameras. These studies investigate the impact of personalizing simple feature-based explanations on effectiveness and satisfaction. Both approximated and real effectiveness is investigated. Contrary to expectation, personalization was detrimental to effectiveness, though it may improve user satisfaction. The studies also highlighted the importance of considering opt-out rates and the underlying rating distribution when evaluating effectiveness. [ABSTRACT FROM AUTHOR]

This paper addresses the theoretical notion of a game as it arises across scientific inquiries, exploring its uses as a technical and formal asset in logic and science versus an explanatory mechanism. While games comprise a widely used method in a broad intellectual realm (including, but not limited to, philosophy, logic, mathematics, cognitive science, artificial intelligence, computation, linguistics, physics, economics), each discipline advocates its own methodology and a unified understanding is lacking. In the first part of this paper, a number of game theories in formal studies are critically surveyed. In the second part, the doctrine of games as explanations for logic is assessed, and the relevance of a conceptual analysis of games to cognition discussed. It is suggested that the notion of evolution plays a part in the game-theoretic concept of meaning. [ABSTRACT FROM AUTHOR]

When justifying a mathematical statement, textbooks often interweave several instances of justification, creating, what we term, paths of justification. This study aims to characterize the paths of justification of mathematical statements that are offered in eight 7th grade Israeli mathematics textbooks. The study attended to context and compared paths of justification in different mathematical domains (algebra and geometry) and in textbooks intended for students with different achievement levels. Analysis focused on the number of instances of justification in a path, the types of justification of those instances, and their order. Comparing paths of justification between algebra and geometry, and between textbooks intended for students with different levels of achievements, revealed several similarities and differences. The findings portray a complex picture of textbook justifications, underlining the need to be sensitive to context when studying textbook justifications. The study results are discussed in light of relevant literature, highlighting potential factors involved in the way a textbook path of justification of a specific mathematical statement is structured. [ABSTRACT FROM AUTHOR]

One of the most well-known and widely used local search techniques for solving optimization problems in Constraint Programming is the Large Neighborhood Search (LNS) algorithm. Such a technique is, by nature, very flexible and can be easily integrated within standard backtracking procedures. One of its drawbacks is that the relaxation process is quite often problem dependent. Several works have been dedicated to overcome this issue through problem independent parameters. Nevertheless, such generic approaches need to be carefully parameterized at the instance level. In this paper, we demonstrate that the issue of finding a problem independent neighborhood generation technique for LNS can be addressed using explanation-based neighborhoods. An explanation is a subset of constraints and decisions which justifies a solver event such as a domain modification or a conflict. We evaluate our proposal for a set of optimization problems. We show that our approach is at least competitive with or even better than state-of-the-art algorithms and can be easily combined with state-of-the-art neighborhoods. Such results pave the way to a new use of explanation-based approaches for improving search. [ABSTRACT FROM AUTHOR]

Few empirical studies have examined how people understand counterfactual explanations for other people's decisions, for example, "if you had asked for a lower amount, your loan application would have been approved". Yet many current Artificial Intelligence (AI) decision support systems rely on counterfactual explanations to improve human understanding and trust. We compared counterfactual explanations to causal ones, i.e., "because you asked for a high amount, your loan application was not approved", for an AI's decisions in a familiar domain (alcohol and driving) and an unfamiliar one (chemical safety) in four experiments (n = 731). Participants were shown inputs to an AI system, its decisions, and an explanation for each decision; they attempted to predict the AI's decisions, or to make their own decisions. Participants judged counterfactual explanations more helpful than causal ones, but counterfactuals did not improve the accuracy of their predictions of the AI's decisions more than causals (Experiment 1). However, counterfactuals improved the accuracy of participants' own decisions more than causals (Experiment 2). When the AI's decisions were correct (Experiments 1 and 2), participants considered explanations more helpful and made more accurate judgements in the familiar domain than in the unfamiliar one; but when the AI's decisions were incorrect, they considered explanations less helpful and made fewer accurate judgements in the familiar domain than the unfamiliar one, whether they predicted the AI's decisions (Experiment 3a) or made their own decisions (Experiment 3b). The results corroborate the proposal that counterfactuals provide richer information than causals, because their mental representation includes more possibilities. [ABSTRACT FROM AUTHOR]

basic goal in human-robot interaction is to establish such a communication mode between the two parties that the humans perceive it as effective and natural; effective in the sense of being responsive to the information needs of the humans, and natural in the sense of communicating information in modes familiar to humans. This paper sets the framework for a robot guide to visitors in art collections and other assistive environments, which incorporates the principles of effectiveness and naturalness. The human-robot interaction takes place in natural language in the form of a dialogue session during which the robot describes exhibits, but also recommends exhibits that might be of interest to the visitors. It is also possible for the robot to explain its reasoning to the visitors, with a view to increasing transparency and consequently trust in the robot's suggestions. Furthermore, the robot leads the visitors to the location of the desired exhibit. The framework is general enough to be implemented in different hardware, including portable computational devices. The framework is based on a cognitive model comprised of four modules: a reactive, a deliberative, a reflective and an affective one. An initial implementation of a dialogue system realising this cognitive model is presented. main ontology. [ABSTRACT FROM AUTHOR]

The purpose of this exploratory study is to examine the use of an ethical intervention strategy – counterexplanation – on individuals’ ethical decision-making. As opposed to providing reasons to support a decision in the case of explanation, counterexplanation is the provision of reasons that either speak against or provide evidence against a chosen course of action. The number of explanations and/or counterexplanations provided by the participants is expected to have a significant effect on ethical evaluation and intention. The number of explanations is expected to be negatively related to ethical decision-making while the number of counterexplanations is expected to be positively related to ethical decision-making. The experiment, that made use of five ethical vignettes, manipulated four treatment groups – explanation, counterexplanation, explanation/counterexplanation, and counterexplanation/explanation. Participants were randomly assigned to one of the four reatments. They performed the requirements of their treatment before recording their ethical evaluations and intentions. As expected, larger numbers of explanations led to less ethical decision-making and larger numbers of counterexplanations led to more ethical decision-making. However, when both types of explanations are required, the order of counterexplaining before explaining is more desirable as it leads to more ethical decision-making. The study also reports that individuals with high social desirability bias (a tendency to present oneself in a culturally acceptable manner) may generate less counterexplanations. Implications of the findings are explained in the paper. [ABSTRACT FROM AUTHOR]

Identifying structures in a given combinatorial problem is often a key step for designing efficient search heuristics or for understanding the inherent complexity of the problem. Several Operations Research approaches apply decomposition or relaxation strategies upon such a structure identified within a given problem. The next step is to design algorithms that adaptively integrate that kind of information during search. We claim in this paper, inspired by previous work on impact-based search strategies for constraint programming, that using an explanation-based constraint solver may lead to collect invaluable information on the intimate dynamically revealed and static structures of a problem instance. Moreover, we discuss how dedicated OR solving strategies (such as Benders decomposition) could be adapted to constraint programming when specific relationships between variables are exhibited. [ABSTRACT FROM AUTHOR]

There are several possible approaches in which dynamic computerized environments play a significant, and possibly unique, role in supporting innovative learning trajectories in mathematics in general, and particularly in geometry. These approaches are influenced by the way one views mathematics and mathematical activity. In this paper we briefly describe an approach based on a problem situation and our experiences using it with students and teachers. This leads naturally to a discussion of some of the ways in which parts of the mathematics curriculum, classroom practice, and student learning may differ from the traditional approach. [ABSTRACT FROM AUTHOR]

Reciprocal Recommender Systems (RRS) aim to recommend relevant matches to users based on the mutual agreement of their preferences. Explainability of reciprocal recommendations is important for developing a persuasive reciprocal recommender system, since it can improve the effectiveness and credibility of the reciprocal recommendation results. Explainable RRS provide an explanation highlighting why a recommendation would be relevant to the user. Explaining the rationale behind predictions with textual or visual artifacts help in increasing trustworthiness and transparency of the system which is crucial especially for models that are used in critical decision making. In this work, XSiameseBiGRU-UCB, a deep learning contextual bandits framework with post-hoc argumentation based explanations for RRS is proposed. XSiameseBiGRU-UCB is an explainable Siamese neural network-based framework that provides explanations to justify the generated reciprocal recommendations for both the parties involved. In RRS, dilemma between exploitation and exploration requires identifying the best possible recommendation from known information or collecting more information about the environment while generating reciprocal recommendations. To tackle this, we propose to use a contextual bandit policy with upper confidence bound, which adaptively exploits and explores user interests to achieve increased rewards in the long run. Experimental studies conducted with four real-world datasets validate the efficacy of the proposed approach. [ABSTRACT FROM AUTHOR]

Expert systems have had little impact as computing artifacts. In this paper we argue that the reason for this stems from the underlying assumption of most builders of expert systems that an expert system needs to acquire information and to control the interaction between the human user and itself. We show that this assumption has serious linguistic and usability flaws which diminish the likelihood of producing socially acceptable expert systems. We propose a reversal of this paradigm, for the design of expert systems, by assuming that it is the human user who needs to acquire information and to control the interaction between themselves and the system. [ABSTRACT FROM AUTHOR]

The introduction of machine learning to small molecule research– an inherently multidisciplinary field in which chemists and data scientists combine their expertise and collaborate - has been vital to making screening processes more efficient. In recent years, numerous models that predict pharmacokinetic properties or bioactivity have been published, and these are used on a daily basis by chemists to make decisions and prioritize ideas. The emerging field of explainable artificial intelligence is opening up new possibilities for understanding the reasoning that underlies a model. In small molecule research, this means relating contributions of substructures of compounds to their predicted properties, which in turn also allows the areas of the compounds that have the greatest influence on the outcome to be identified. However, there is no interactive visualization tool that facilitates such interdisciplinary collaborations towards interpretability of machine learning models for small molecules. To fill this gap, we present CIME (ChemInformatics Model Explorer), an interactive web-based system that allows users to inspect chemical data sets, visualize model explanations, compare interpretability techniques, and explore subgroups of compounds. The tool is model-agnostic and can be run on a server or a workstation. [ABSTRACT FROM AUTHOR]

Due to the prominent role of recommender systems in our daily lives, it is increasingly important to inform users why certain items are recommended and personalize these explanations to the user. In this study, we explored how explanations in a music recommender system should be designed to fit the preference of different personal characteristics. More specifically, we investigated three personal characteristics that influence the perception of explanations in music recommender system interfaces: need for cognition, musical sophistication, and openness. For each of these personal characteristics, we designed explanations for users with lower and higher levels of the personal characteristic. Afterward, we conducted for each personal characteristic a within-subject user study in which we compared the two explanations. Based on the results of these user studies, we provide design suggestions to adapt explanations to different levels of these three personal characteristics. In general, we suggest providing explanations up-front for all recommendations at once. For users low in need for cognition, displaying these explanations must be optional. To support users with low musical sophistication, we suggest providing brief explanations that do not require domain knowledge. For users with low openness, we suggest providing explanations with a lower number of explanation elements. [ABSTRACT FROM AUTHOR]

The pervasive use of computer simulations in the sciences brings novel epistemological issues discussed in the philosophy of science literature since about a decade. Evolutionary biology strongly relies on such simulations, and in relation to it there exists a research program (Artificial Life) that mainly studies simulations themselves. This paper addresses the specificity of computer simulations in evolutionary biology, in the context (described in Sect. 1) of a set of questions about their scope as explanations, the nature of validation processes and the relation between simulations and true experiments or mathematical models. After making distinctions, especially between a weak use where simulations test hypotheses about the world, and a strong use where they allow one to explore sets of evolutionary dynamics not necessarily extant in our world, I argue in Sect. 2 that (weak) simulations are likely to represent in virtue of the fact that they instantiate specific features of causal processes that may be isomorphic to features of some causal processes in the world, though the latter are always intertwined with a myriad of different processes and hence unlikely to be directly manipulated and studied. I therefore argue that these simulations are merely able to provide candidate explanations for real patterns. Section 3 ends up by placing strong and weak simulations in Levins' triangle, that conceives of simulations as devices trying to fulfil one or two among three incompatible epistemic values (precision, realism, genericity). [ABSTRACT FROM AUTHOR]

Generative learning theory posits that learners engage more deeply and produce better learning outcomes when they engage in selecting, organizing, and integrating processes during learning. The present experiments examine whether the generative learning activity of generating explanations can be extended to online multimedia lessons and whether prompts to engage in this generative learning activity work better than more passive instruction. Across three experiments, college students learned about greenhouse gasses from a 4-part online lesson involving captioned animations and subsequently took a posttest. After each part, learners were asked to generate an explanation (write-an-explanation), write an explanation using provided terms (write-a-focused-explanation), rewrite a provided explanation (rewrite-an-explanation), read a provided explanation (read-an-explanation), or simply move on to the next part (no-activity). Overall, students in the write-an-explanation group (Experiments 2 and 3), write-a-focused-explanation group (Experiment 2), and rewrite-an-explanation group (Experiment 3) performed significantly better on a delayed posttest than the no-activity group, but the groups did not differ significantly on an immediate posttest (Experiment 1). These results are consistent with generative learning theory and help identify generative learning strategies that improve online multimedia learning, thereby priming active learning with passive media. [ABSTRACT FROM AUTHOR]

Conflict detection is used in various scenarios ranging from interactive decision making (e.g., knowledge-based configuration) to the diagnosis of potentially faulty models (e.g., using knowledge base analysis operations). Conflicts can be regarded as sets of restrictions (constraints) causing an inconsistency. Junker's QuickXPlain is a divide-and-conquer based algorithm for the detection of preferred minimal conflicts. In this article, we present a novel approach to the detection of such conflicts which is based on speculative programming. We introduce a parallelization of QuickXPlain and empirically evaluate this approach on the basis of synthesized knowledge bases representing feature models. The results of this evaluation show significant performance improvements in the parallelized QuickXPlain version. [ABSTRACT FROM AUTHOR]

A range of explanation engines assist data analysts by performing feature selection over increasingly high-volume and high-dimensional data, grouping and highlighting commonalities among data points. While useful in diverse tasks such as user behavior analytics, operational event processing, and root-cause analysis, today's explanation engines are designed as stand-alone data processing tools that do not interoperate with traditional, SQL-based analytics workflows; this limits the applicability and extensibility of these engines. In response, we propose the DIFF operator, a relational aggregation operator that unifies the core functionality of these engines with declarative relational query processing. We implement both single-node and distributed versions of the DIFF operator in MB SQL, an extension of MacroBase, and demonstrate how DIFF can provide the same semantics as existing explanation engines while capturing a broad set of production use cases in industry, including at Microsoft and Facebook. Additionally, we illustrate how this declarative approach to data explanation enables new logical and physical query optimizations. We evaluate these optimizations on several real-world production applications and find that DIFF in MB SQL can outperform state-of-the-art engines by up to an order of magnitude. [ABSTRACT FROM AUTHOR]

The initial analysis of any large data set can be divided into two phases: (1) the identification of common trends or patterns and (2) the identification of anomalies or outliers that deviate from those trends. We focus on the goal of detecting observations with novel content, which can alert us to artifacts in the data set or, potentially, the discovery of previously unknown phenomena. To aid in interpreting and diagnosing the novel aspect of these selected observations, we recommend the use of novelty detection methods that generate explanations. In the context of large image data sets, these explanations should highlight what aspect of a given image is new (color, shape, texture, content) in a human-comprehensible form. We propose DEMUD-VIS, the first method for providing visual explanations of novel image content by employing a convolutional neural network (CNN) to extract image features, a method that uses reconstruction error to detect novel content, and an up-convolutional network to convert CNN feature representations back into image space. We demonstrate this approach on diverse images from ImageNet, freshwater streams, and the surface of Mars. Finally, we evaluate the utility of the visual explanations with a user study. [ABSTRACT FROM AUTHOR]

Review texts constitute a valuable source for making system-generated recommendations both more accurate and more transparent. Reviews typically contain statements providing argumentative support for a given item rating that can be exploited to explain the recommended items in a personalized manner. We propose a novel method called Aspect-based Transparent Memories (ATM) to model user preferences with respect to relevant aspects and compare them to item properties to predict ratings, and, by the same mechanism, explain why an item is recommended. The ATM architecture consists of two neural memories that can be viewed as arrays of slots for storing information about users and items. The first memory component encodes representations of sentences composed by the target user while the second holds an equivalent representation for the target item based on statements of other users. An offline evaluation was performed with three datasets, showing advantages over two baselines, the well-established Matrix Factorization technique and a recent competitive representative of neural attentional recommender techniques. [ABSTRACT FROM AUTHOR]

We propose a technique for producing 'visual explanations' for decisions from a large class of Convolutional Neural Network (CNN)-based models, making them more transparent and explainable. Our approach—Gradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of any target concept (say 'dog' in a classification network or a sequence of words in captioning network) flowing into the final convolutional layer to produce a coarse localization map highlighting the important regions in the image for predicting the concept. Unlike previous approaches, Grad-CAM is applicable to a wide variety of CNN model-families: (1) CNNs with fully-connected layers (e.g.VGG), (2) CNNs used for structured outputs (e.g.captioning), (3) CNNs used in tasks with multi-modal inputs (e.g.visual question answering) or reinforcement learning, all without architectural changes or re-training. We combine Grad-CAM with existing fine-grained visualizations to create a high-resolution class-discriminative visualization, Guided Grad-CAM, and apply it to image classification, image captioning, and visual question answering (VQA) models, including ResNet-based architectures. In the context of image classification models, our visualizations (a) lend insights into failure modes of these models (showing that seemingly unreasonable predictions have reasonable explanations), (b) outperform previous methods on the ILSVRC-15 weakly-supervised localization task, (c) are robust to adversarial perturbations, (d) are more faithful to the underlying model, and (e) help achieve model generalization by identifying dataset bias. For image captioning and VQA, our visualizations show that even non-attention based models learn to localize discriminative regions of input image. We devise a way to identify important neurons through Grad-CAM and combine it with neuron names (Bau et al. in Computer vision and pattern recognition, 2017) to provide textual explanations for model decisions. Finally, we design and conduct human studies to measure if Grad-CAM explanations help users establish appropriate trust in predictions from deep networks and show that Grad-CAM helps untrained users successfully discern a 'stronger' deep network from a 'weaker' one even when both make identical predictions. Our code is available at https://github.com/ramprs/grad-cam/, along with a demo on CloudCV (Agrawal et al., in: Mobile cloud visual media computing, pp 265–290. Springer, 2015) (http://gradcam.cloudcv.org) and a video at http://youtu.be/COjUB9Izk6E. [ABSTRACT FROM AUTHOR]

A layoff is a threatening yet common event which employees might face at some point in their working lives. In two scenario-based experiments (total N = 344), we investigated which actions of a layoff agent (i.e., who delivers the layoff notice) during a dismissal notification meeting may contribute to laid-off employees' fairness judgments and negative attitudes toward the employer. In general, the extent to which layoff victims were treated with respect was consistently found to increase perceptions of interpersonal and procedural fairness and to mitigate negative attitudes toward the employer. Further results showed that layoff victims preferred to be given an adequate (vs. inadequate) explanation of the reasons for the layoff and to receive notice from the direct supervisor (vs. an external consultant). Relationships between the layoff agent's actions and layoff victims' negative attitudes toward the employer were mediated by perceptions of procedural fairness. In addition, delegating the layoff agent's task to an external consultant increased perceived psychological contract breach. Our findings have important implications for organizational justice research and for the managerial practice of implementing fair layoffs. In particular, small actions, such as treating employees with respect, might be of benefit both to humans and organizations. [ABSTRACT FROM AUTHOR]

One line of research on student understanding of complex systems has tended to emphasize discontinuities between common misconceptions and relatively more sophisticated understandings. Other work has focused on instruction while acknowledging the existence of other ways of understanding complex systems, but less emphasis has been on examining the knowledge structures within these intermediate ways of understanding. This study takes a microgenetic approach to examining student’s explanations for the behavior of complex systems. Using the Knowledge-in-Pieces epistemological perspective, the analysis documents a continuity of reasoning patterns across less prototypically centralized and more prototypically decentralized (a more sophisticated causality) explanations while explaining the movement of sand dunes. The first analysis examines 31 interviews and shows that many reflected a general reasoning pattern that encompassing some combination of an initial centralized explanation, a final decentralized explanation, and transitional explanations. A second analysis examines a single student’s reasoning pattern and finds that the activation of relevant intuitive knowledge pieces (p-prims) and transitional explanations function as threads of continuity across the continuum of reasoning patterns. These findings suggest that students are able to exhibit a continuity of reasoning patterns across centralized to decentralized causality and are able to access productive intuitive knowledge about complex systems that are applicable to both the macro and micro levels of sand dune movement. Implications suggest that future research investigate these transitional explanations along with the mechanisms of shifting explanations that can account for this robust continuum. [ABSTRACT FROM AUTHOR]

This article describes several approaches to assessing student understanding using written explanations that students generate as part of a multiple-document inquiry activity on a scientific topic (global warming). The current work attempts to capture the causal structure of student explanations as a way to detect the quality of the students' mental models and understanding of the topic by combining approaches from Cognitive Science and Artificial Intelligence, and applying them to Education. First, several attributes of the explanations are explored by hand coding and leveraging existing technologies (LSA and Coh-Metrix). Then, we describe an approach for inferring the quality of the explanations using a novel, two-phase machine-learning approach for detecting causal relations and the causal chains that are present within student essays. The results demonstrate the benefits of using a machine-learning approach for detecting content, but also highlight the promise of hybrid methods that combine ML, LSA and Coh-Metrix approaches for detecting student understanding. Opportunities to use automated approaches as part of Intelligent Tutoring Systems that provide feedback toward improving student explanations and understanding are discussed. [ABSTRACT FROM AUTHOR]

I critically examine Adams and Garrison's proposed necessary condition for the mark of the cognitive (Adams and Garrison in Minds Mach 23(3):339-352, ). After a brief presentation of their position, I argue not only that their proposal is in need of additional support, but also that it is too restrictive. [ABSTRACT FROM AUTHOR]

Recommender systems usually provide explanations of their recommendations to better help users to choose products, activities or even friends. Up until now, the type of an explanation style was considered in accordance to the recommender system that employed it. This relation was one-to-one, meaning that for each different recommender systems category, there was a different explanation style category. However, this kind of one-to-one correspondence can be considered as over-simplistic and non generalizable. In contrast, we consider three fundamental resources that can be used in an explanation: users, items and features and any combination of them. In this survey, we define (i) the Human style of explanation, which provides explanations based on similar users, (ii) the Item style of explanation, which is based on choices made by a user on similar items and (iii) the Feature style of explanation, which explains the recommendation based on item features rated by the user beforehand. By using any combination of the aforementioned styles we can also define the Hybrid style of explanation. We demonstrate how these styles are put into practice, by presenting recommender systems that employ them. Moreover, since there is inadequate research in the impact of social web in contemporary recommender systems and their explanation styles, we study new emerged social recommender systems i.e. Facebook Connect explanations (HuffPo, Netflix, etc.) and geo-social explanations that combine geographical with social data (Gowalla, Facebook Places, etc.). Finally, we summarize the results of three different user studies, to support that Hybrid is the most effective explanation style, since it incorporates all other styles. [ABSTRACT FROM AUTHOR]

The global constraint was proposed for modelling cumulative resources in scheduling problems for finite domain (FD) propagation. Since that time a great deal of research has investigated new stronger and faster filtering techniques for , but still most of these techniques only pay off in limited cases or are not scalable. Recently, the 'lazy clause generation' hybrid solving approach has been devised which allows a finite domain propagation engine possible to take advantage of advanced SAT technology, by 'lazily' creating a SAT model of an FD problem as computation progresses. This allows the solver to make use of SAT explanation and autonomous search capabilities. In this article we show how, once we use lazy clause generation, modelling the constraint by decomposition creates a highly competitive version of . Using decomposition into component parts automatically makes the propagator incremental and able to explain itself. We then show how, using the insights from the behaviour of the decomposition, we can create global constraints that explain their propagation. We compare these approaches to explaining the constraint on resource constrained project scheduling problems. All our methods are able to close a substantial number of open problems from the well-established PSPlib benchmark library of resource-constrained project scheduling problems. [ABSTRACT FROM AUTHOR]

What makes a good explanation? We examine the function of latent scope, i.e., the number of unobserved phenomena that an explanation can account for. We show that individuals prefer narrow latent scope explanations-those that account for fewer unobserved effects-to broader explanations. In Experiments 1a-d, participants found narrow latent scope explanations to be both more satisfying and more likely. In Experiment 2 we directly manipulated base rate information and again found a preference for narrow latent scope explanations. Participants in Experiment 3 evaluated more natural explanations of unexpected observations, and again displayed a bias for narrow latent scope explanations. We conclude by considering what this novel bias tells us about how humans evaluate explanations and engage in causal reasoning. [ABSTRACT FROM AUTHOR]

Conventions consist of conventionally accepted reasons for dereliction, deviation, distinction, or good fortune. Their use facilitates coordination of interpersonal effort through appeal to shared understandings that emerge from the push and pull of social interaction, but then constrain further rounds of that interaction. The use of conventions differs from three other well defined and widely used forms of reason giving: codes, technical accounts, and stories. All four do relational work, but conventions operate most easily and effectively when participants in social relations are simply confirming the character of those relations rather than establishing them anew, contesting them, terminating them, or transforming them. In those cases, participants are more likely to employ codes, technical accounts, or stories. [ABSTRACT FROM AUTHOR]

Volunteer non-major chemistry students taking an introductory university chemistry course ( n = 17) were interviewed about their understanding of a variety of chemical diagrams. All the students’ interviewed appreciated that diagrams of laboratory equipment were useful to show how to set up laboratory equipment. However students’ ability to explain specific diagrams at either the macroscopic or sub-microscopic level varied greatly. The results highlighted the poor level of understanding that some students had even after completing both exercises and experiments using the diagrams. The connection between the diagrams of the macroscopic level (equipment, chemicals), the sub-microscopic level (molecular) and the symbolic level (equations) was not always considered explicitly by students. The results indicate a need for chemical diagrams to be used carefully and more explicitly to ensure learner understanding. Correspondingly, students need to interpret visual chemical diagrams using meta-visualization skills linking the various levels of representation, and appreciating the role of the diagrams in explanations need to be developed. [ABSTRACT FROM AUTHOR]

Previous research on peer tutoring has found that students sometimes benefit academically from tutoring other students. In this study we combined quantitative and qualitative analyses to explore how untrained peer tutors learned via explaining and responding to tutee questions in a non-reciprocal tutoring setting. In support of our hypotheses, we found that tutors learned most effectively when their instructional activities incorporated reflective knowledge-building in which they monitored their own understanding, generated inferences to repair misunderstandings, and elaborated upon the source materials. However, tutors seemed to adopt a knowledge-telling bias in which they primarily summarized the source materials with little elaboration. Tutors’ reflective knowledge-building activities, when they occurred, were more frequently elicited by interactions with their tutee. In particular, when tutees asked questions that contained an inference or required an inferential answer, tutors’ responses were more likely to be elaborative and metacognitive. Directions for future research are also discussed. [ABSTRACT FROM AUTHOR]