Abstract: Text documents can be described by a number of abstract concepts such as semantic category, writing style, or sentiment. Machine learning (ML) models have been trained to automatically map documents to these abstract concepts, allowing to annotate very large text collections, more than could be processed by a human in a lifetime. Besides predicting the text’s category very accurately, it is also highly desirable to understand how and why the categorization process takes place. In this paper, we demonstrate that such understanding can be achieved by tracing the classification decision back to individual words using layer-wise relevance propagation (LRP), a recently developed technique for explaining predictions of complex non-linear classifiers. We train two word-based ML models, a convolutional neural network (CNN) and a bag-of-words SVM classifier, on a topic categorization task and adapt the LRP method to decompose the predictions of these models onto words. Resulting scores indicate how much individual words contribute to the overall classification decision. This enables one to distill relevant information from text documents without an explicit semantic information extraction step. We further use the word-wise relevance scores for generating novel vector-based document representations which capture semantic information. Based on these document vectors, we introduce a measure of model explanatory power and show that, although the SVM and CNN models perform similarly in terms of classification accuracy, the latter exhibits a higher level of explainability which makes it more comprehensible for humans and potentially more useful for other applications.

Abstract: What determines how languages categorize colors? We analyzed results of the World Color Survey (WCS) of 110 languages to show that despite gross differences across languages, communication of chromatic chips is always better for warm colors (yellows/reds) than cool colors (blues/greens). We present an analysis of color statistics in a large databank of natural images curated by human observers for salient objects and show that objects tend to have warm rather than cool colors. These results suggest that the cross-linguistic similarity in color-naming efficiency reflects colors of universal usefulness and provide an account of a principle (color use) that governs how color categories come about. We show that potential methodological issues with the WCS do not corrupt information-theoretic analyses, by collecting original data using two extreme versions of the color-naming task, in three groups: the Tsimane', a remote Amazonian hunter-gatherer isolate; Bolivian-Spanish speakers; and English speakers. These data also enabled us to test another prediction of the color-usefulness hypothesis: that differences in color categorization between languages are caused by differences in overall usefulness of color to a culture. In support, we found that color naming among Tsimane' had relatively low communicative efficiency, and the Tsimane' were less likely to use color terms when describing familiar objects. Color-naming among Tsimane' was boosted when naming artificially colored objects compared with natural objects, suggesting that industrialization promotes color usefulness.

Abstract: Text categorization, or text classification, is one of key tasks for representing the semantic information of documents. Multi-label text categorization is finer-grained approach to text categorization which consists of assigning multiple target labels to documents. It is more challenging compared to the task of multi-class text categorization due to the exponential growth of label combinations. Existing approaches to multi-label text categorization fall short to extract local semantic information and to model label correlations. In this paper, we propose an ensemble application of convolutional and recurrent neural networks to capture both the global and the local textual semantics and to model high-order label correlations while having a tractable computational complexity. Extensive experiments show that our approach achieves the state-of-the-art performance when the CNN-RNN model is trained using a large-sized dataset.

Abstract: Culturally diverse colleagues can be valuable sources for stimulating creativity at work, yet only if they decide to share their knowledge. Drawing on the social exchange theory, we propose that cross-cultural interactions among individuals from different national backgrounds can act as a salient contingency in the relationship between knowledge hiding and creativity (individual and team). We further suggest, based on the social categorization theory (e.g., the categorization process of “us” against “them” based on national differences), that cultural intelligence enhances the likelihood of high-quality social exchanges between culturally diverse individuals and, therefore, remedies the otherwise negative relationship between individual knowledge hiding and individual creativity. Two studies using field and experimental data offer consistent support for this argument. First, a field study of 621 employees nested among 70 teams revealed that individual knowledge hiding is negatively related to indi...

Abstract: This study introduces the Sentiment Analysis and Cognition Engine (SEANCE), a freely available text analysis tool that is easy to use, works on most operating systems (Windows, Mac, Linux), is housed on a user’s hard drive (as compared to being accessed via an Internet interface), allows for batch processing of text files, includes negation and part-of-speech (POS) features, and reports on thousands of lexical categories and 20 component scores related to sentiment, social cognition, and social order. In the study, we validated SEANCE by investigating whether its indices and related component scores can be used to classify positive and negative reviews in two well-known sentiment analysis test corpora. We contrasted the results of SEANCE with those from Linguistic Inquiry and Word Count (LIWC), a similar tool that is popular in sentiment analysis, but is pay-to-use and does not include negation or POS features. The results demonstrated that both the SEANCE indices and component scores outperformed LIWC on the categorization tasks.

Abstract: We present an approach to automatically classify clinical text at a sentence level. We are using deep convolutional neural networks to represent complex features. We train the network on a dataset providing a broad categorization of health information. Through a detailed evaluation, we demonstrate that our method outperforms several approaches widely used in natural language processing tasks by about 15%.

Abstract: It is widely recognized that language plays a key role in the transmission of human culture, but relatively little is known about the mechanisms by which language simultaneously encourages both cultural stability and cultural innovation. This paper examines this issue by focusing on the use of language to transmit categories, focusing on two universal devices: labels (e.g., shark, woman) and generics (e.g., "sharks attack swimmers"; "women are nurturing"). We propose that labels and generics each assume two key principles: norms and essentialism. The normative assumption permits transmission of category information with great fidelity, whereas essentialism invites innovation by means of an open-ended, placeholder structure. Additionally, we sketch out how labels and generics aid in conceptual alignment and the progressive "looping" between categories and cultural practices. In this way, human language is a technology that enhances and expands the categorization capacities that we share with other animals.

Abstract: The primary aim of this chapter is to provide a framework to understand and synthesize the processes of person construal—early perceptions that lead to initial ingroup/outgroup categorizations—with the processes involved in intergroup relations. To this end, we review research examining the initial perception and categorization of ingroup and outgroup members and its downstream consequences. We first discuss bottom-up processes in person construal based on visual features (e.g., facial prototypicality and bodily cues), and then discuss how top-down factors (e.g., beliefs, stereotypes) may influence these processes. Next, we examine how the initial categorization of targets as ingroup or outgroup members influences identification, stereotyping, and group-based evaluations, and the relations between these constructs. We also explore the implications of the activation of these constructs for a range of social judgments including emotion identification, empathy, and intergroup behaviors. Finally, we describe a variety of well established and more recent strategies to reduce intergroup bias that target the activation of category-based knowledge, including intergroup contact, approach orientations, evaluative conditioning, and perspective taking.

Abstract: We present an approach to automatically classify clinical text at a sentence level. We are using deep convolutional neural networks to represent complex features. We train the network on a dataset providing a broad categorization of health information. Through a detailed evaluation, we demonstrate that our method outperforms several approaches widely used in natural language processing tasks by about 15%.

Abstract: Background: In order to harness what people are tweeting about Zika, there needs to be a computational framework that leverages machine learning techniques to recognize relevant Zika tweets and, further, categorize these into disease-specific categories to address specific societal concerns related to the prevention, transmission, symptoms, and treatment of Zika virus. Objective: The purpose of this study was to determine the relevancy of the tweets and what people were tweeting about the 4 disease characteristics of Zika: symptoms, transmission, prevention, and treatment. Methods: A combination of natural language processing and machine learning techniques was used to determine what people were tweeting about Zika. Specifically, a two-stage classifier system was built to find relevant tweets about Zika, and then the tweets were categorized into 4 disease categories. Tweets in each disease category were then examined using latent Dirichlet allocation (LDA) to determine the 5 main tweet topics for each disease characteristic. Results: Over 4 months, 1,234,605 tweets were collected. The number of tweets by males and females was similar (28.47% [351,453/1,234,605] and 23.02% [284,207/1,234,605], respectively). The classifier performed well on the training and test data for relevancy (F1 score=0.87 and 0.99, respectively) and disease characteristics (F1 score=0.79 and 0.90, respectively). Five topics for each category were found and discussed, with a focus on the symptoms category. Conclusions: We demonstrate how categories of discussion on Twitter about an epidemic can be discovered so that public health officials can understand specific societal concerns within the disease-specific categories. Our two-stage classifier was able to identify relevant tweets to enable more specific analysis, including the specific aspects of Zika that were being discussed as well as misinformation being expressed. Future studies can capture sentiments and opinions on epidemic outbreaks like Zika virus in real time, which will likely inform efforts to educate the public at large. [JMIR Public Health Surveill 2017;3(2):e38]

Abstract: Multitask learning (MTL) is commonly used for jointly optimizing multiple learning tasks. To date, all existing MTL methods have been designed for tasks with feature-vector represented instances, but cannot be applied to structure data, such as graphs. More importantly, when carrying out MTL, existing methods mainly focus on exploring overall commonality or disparity between tasks for learning, but cannot explicitly capture task relationships in the feature space, so they are unable to answer important questions, such as what exactly is shared between tasks and what is the uniqueness of one task differing from others? In this paper, we formulate a new multitask graph learning problem, and propose a task sensitive feature exploration and learning algorithm for multitask graph classification. Because graphs do not have features available, we advocate a task sensitive feature exploration and learning paradigm to jointly discover discriminative subgraph features across different tasks. In addition, a feature learning process is carried out to categorize each subgraph feature into one of three categories: 1) common feature; 2) task auxiliary feature; and 3) task specific feature, indicating whether the feature is shared by all tasks, by a subset of tasks, or by only one specific task, respectively. The feature learning and the multiple task learning are iteratively optimized to form a multitask graph classification model with a global optimization goal. Experiments on real-world functional brain analysis and chemical compound categorization demonstrate the algorithm’s performance. Results confirm that our method can be used to explicitly capture task correlations and uniqueness in the feature space, and explicitly answer what are shared between tasks and what is the uniqueness of a specific task.

Abstract: Regulatory categorization can be a matter of life and death to firms, as it sets legal limitations on the production and sales of their product. In this paper we set out to uncover this critical pr...

Abstract: Computational thinking is an increasingly important focus in computer science or informatics curricula around the world, and ways of incorporating it into the school curricula are being sought. The Bebras contest on informatics, which originated 12 years ago and now involves around 50 countries, consists of short problemsolving tasks based on topics in informatics. Bebras tasks engender the development of computational thinking skills by incorporating abstraction, algorithmic thinking, decomposition, evaluation and generalization. Bebras tasks cover a range of informatics concepts including algorithms and data structures, programming, networking, databases and social and ethical issues. Having built up a substantial number of Bebras tasks over 12 years it is important to be able to categorise them so that they can be easily accessed by the Bebras community and teachers within schools. The categorization of tasks within Bebras is important as it ensures that tasks span a wide range of topics; there have been several categorization schemes suggested to date. In this paper we present a new twodimensional categorization system that takes account of computational thinking skills as well as content knowledge. Examples are given from recent tasks that illustrate the role that Bebras can play in the development of computational thinking skills.

Abstract: Language exerts a powerful influence on our concepts. We review evidence documenting the developmental origins of a precocious link between language and object categories in very young infants. This collection of studies documents a cascading process in which early links between language and cognition provide the foundation for later, more precise ones. We propose that, early in life, language promotes categorization at least in part through its status as a social, communicative signal. But over the first year, infants home in on the referential power of language and, by their second year, begin teasing apart distinct kinds of names (e.g. nouns, adjectives) and their relation to distinct kinds of concepts (e.g. object categories, properties). To complement this proposal, we also relate this evidence to several alternative accounts of language's effect on categorization, appealing to similarity ('labels-as-features'), familiarity ('auditory overshadowing'), and communicative biases ('natural pedagogy').

Abstract: During spoken language comprehension listeners transform continuous acoustic cues into categories (e.g., /b/ and /p/). While long-standing research suggests that phonetic categories are activated in a gradient way, there are also clear individual differences in that more gradient categorization has been linked to various communication impairments such as dyslexia and specific language impairments (Joanisse, Manis, Keating, & Seidenberg, 2000; Lopez-Zamora, Luque, Alvarez, & Cobos, 2012; Serniclaes, Van Heghe, Mousty, Carre, & Sprenger-Charolles, 2004; Werker & Tees, 1987). Crucially, most studies have used 2-alternative forced choice (2AFC) tasks to measure the sharpness of between-category boundaries. Here we propose an alternative paradigm that allows us to measure categorization gradiency in a more direct way. Furthermore, we follow an individual differences approach to (a) link this measure of gradiency to multiple cue integration, (b) explore its relationship to a set of other cognitive processes, and (c) evaluate its role in individuals' ability to perceive speech in noise. Our results provide validation for this new method of assessing phoneme categorization gradiency and offer preliminary insights into how different aspects of speech perception may be linked to each other and to more general cognitive processes. (PsycINFO Database Record

Abstract: The problem of partitioning a collection of objects based on their measurements on a set of dichotomous variables is a well-established problem in psychological research, with applications including clinical diagnosis, educational testing, cognitive categorization, and choice analysis. Latent class analysis and K-means clustering are popular methods for partitioning objects based on dichotomous measures in the psychological literature. The K-median clustering method has recently been touted as a potentially useful tool for psychological data and might be preferable to its close neighbor, K-means, when the variable measures are dichotomous. We conducted simulation-based comparisons of the latent class, K-means, and K-median approaches for partitioning dichotomous data. Although all 3 methods proved capable of recovering cluster structure, K-median clustering yielded the best average performance, followed closely by latent class analysis. We also report results for the 3 methods within the context of an application to transitive reasoning data, in which it was found that the 3 approaches can exhibit profound differences when applied to real data. (PsycINFO Database Record

Abstract: Cloud computing has been widely adopted by application service providers (ASPs) and enterprises to reduce both capital expenditures (CAPEX) and operational expenditures (OPEX). Applications and services previously running on private data centers are now being migrated to private or public clouds. Since most of the ASPs and enterprises have globally distributed user bases, their services need to be distributed across multiple clouds, spread across the globe which can achieve better performance in terms of latency, scalability and load balancing. The shift has eventually led the research community to study multi-cloud environments. However, the widespread acceptance of such environments has been hampered by major security concerns. Firewalls and traditional rule-based security protection techniques are not sufficient to protect user-data in multi-cloud scenarios. Recently, advances in machine learning techniques have attracted the attention of the research community to build intrusion detection systems (IDS) that can detect anomalies in the network traffic. Most of the research works, however, do not differentiate among different types of attacks. This is, in fact, necessary for appropriate countermeasures and defense against attacks. In this paper, we investigate both detecting and categorizing anomalies rather than just detecting, which is a common trend in the contemporary research works. We have used a popular publicly available dataset to build and test learning models for both detection and categorization of different attacks. To be precise, we have used two supervised machine learning techniques, namely linear regression (LR) and random forest (RF). We show that even if detection is perfect, categorization can be less accurate due to similarities between attacks. Our results demonstrate more than 99% detection accuracy and categorization accuracy of 93.6%, with the inability to categorize some attacks. Further, we argue that such categorization can be applied to multi-cloud environments using the same machine learning techniques.

Abstract: Decision-making is a high-level cognitive process based on cognitive processes like perception, attention, and memory. Real-life situations require series of decisions to be made, with each decision depending on previous feedback from a potentially changing environment. To gain a better understanding of the underlying processes of dynamic decision-making, we applied the method of cognitive modeling on a complex rule-based category learning task. Here, participants first needed to identify the conjunction of two rules that defined a target category and later adapt to a reversal of feedback contingencies. We developed an ACT-R model for the core aspects of this dynamic decision-making task. An important aim of our model was that it provides a general account of how such tasks are solved and, with minor changes, is applicable to other stimulus materials. The model was implemented as a mixture of an exemplar-based and a rule-based approach which incorporates perceptual-motor and metacognitive aspects as well. The model solves the categorization task by first trying out one-feature strategies and then, as a result of repeated negative feedback, switching to two-feature strategies. Overall, this model solves the task in a similar way as participants do, including generally successful initial learning as well as reversal learning after the change of feedback contingencies. Moreover, the fact that not all participants were successful in the two learning phases is also reflected in the modeling data. However, we found a larger variance and a lower overall performance of the modeling data as compared to the human data which may relate to perceptual preferences or additional knowledge and rules applied by the participants. In a next step, these aspects could be implemented in the model for a better overall fit. In view of the large interindividual differences in decision performance between participants, additional information about the underlying cognitive processes from behavioral, psychobiological and neurophysiological data may help to optimize future applications of this model such that it can be transferred to other domains of comparable dynamic decision tasks.

Abstract: Williams and O'Reilly (1998) published a seminal review of diversity research that has become a classic resource for researchers and practitioners alike. In the current review, we update the theoretical record by discussing traditional views of, as well as recent developments to, the 3 prominent frameworks used to understand diversity: social categorization, similarity-attraction, and information and decision-making. Furthermore, we propose a dual pathway model of diversity's effects that integrates all 3 frameworks. In our model, both positive and detrimental effects of diversity stem from processes of social categorization. Whereas these processes disrupt group functioning when intergroup bias is activated, when bias is not activated, we propose that social categorization and reduced attraction to dissimilar others will allow individuals in diverse rather than homogeneous groups to focus more on the task, anticipate differences in task-relevant opinions and perspectives, and engage in more effortful information processing. Finding the balance is key.

Abstract: People interact with entities in the environment in distinct and categorizable ways (e.g., kicking is making contact with foot). We can recognize these action categories across variations in actors, objects, and settings; moreover, we can recognize them from both dynamic and static visual input. However, the neural systems that support action recognition across these perceptual differences are unclear. Here, we used multivoxel pattern analysis of fMRI data to identify brain regions that support visual action categorization in a format-independent way. Human participants were scanned while viewing eight categories of interactions (e.g., pulling) depicted in two visual formats: (1) visually controlled videos of two interacting actors and (2) visually varied photographs selected from the internet involving different actors, objects, and settings. Action category was decodable across visual formats in bilateral inferior parietal, bilateral occipitotemporal, left premotor, and left middle frontal cortex. In most of these regions, the representational similarity of action categories was consistent across subjects and visual formats, a property that can contribute to a common understanding of actions among individuals. These results suggest that the identified brain regions support action category codes that are important for action recognition and action understanding.SIGNIFICANCE STATEMENT Humans tend to interpret the observed actions of others in terms of categories that are invariant to incidental features: whether a girl pushes a boy or a button and whether we see it in real-time or in a single snapshot, it is still pushing Here, we investigated the brain systems that facilitate the visual recognition of these action categories across such differences. Using fMRI, we identified several areas of parietal, occipitotemporal, and frontal cortex that exhibit action category codes that are similar across viewing of dynamic videos and still photographs. Our results provide strong evidence for the involvement of these brain regions in recognizing the way that people interact physically with objects and other people.

Abstract: The present study investigated whether the sequencing of exemplars affects learning of rule-based science categories. We taught undergraduate students chemical categories by showing them diagrams of organic chemical compounds accompanied by their category label in either an interleaved or blocked fashion. On a categorization test of novel exemplars administered 2 days later, interleaved presentation produced superior performance than blocked presentation for categories that were visually simple (Experiment 1) and relatively more complex (Experiment 2). The advantage of interleaved presentation was still observed when the diagnostic features of each category were highlighted during study (Experiments 3 and 4), suggesting that the benefit of interleaving may not be due solely to increasing the salience of contrastive features that differentiate the categories. These findings demonstrate that interleaved presentation can improve learning of educationally relevant categories, even when coupled with pedagogical techniques that emphasize the diagnostic features of each category.

Abstract: Research on how categories of object concepts are implemented in the human brain has focused primarily on the sorts of semantic structures that are found in English and a few other European languages. This paper provides a broader typological perspective by considering the multifarious categories of object concepts that are encoded by languages with nominal classification systems. In these languages, speakers must explicitly categorize objects at both basic and superordinate levels – indicating, for instance, that a particular entity is not just a pencil but an elongated thing. The following semantic parameters of nominal classification systems are discussed: animacy and related properties, shape and related properties, size, constitution, and interaction/function. For each parameter, cross-linguistically frequent and infrequent semantic distinctions are surveyed first, and then their relevance to cognitive neuroscience is considered. These analyses strongly suggest that the neural underpinnings o...

Abstract: This article presents a categorization theory of spatial voting, which postulates that voters perceive political stances through coarse classifications. Because voters think in terms of categories defined by the ideological center, their behavior deviates from standard models of utility maximization along ideological continua. Their preferences are characterized by discontinuities, rewarding parties on their side of the ideological space more than existing spatial models would predict. While this study concurs with prior studies suggesting that voters tend to use a proximity rule, it argues that this rule mainly serves to distinguish among parties of the same side. Overall, the results suggest that voters’ party evaluations are characterized by a nontrivial identity component, generating in-group biases not captured by the existing spatial models of voting.

Abstract: In this paper we address the problem of grounding distributional representations of lexical meaning. We introduce a new model which uses stacked autoencoders to learn higher-level representations from textual and visual input. The visual modality is encoded via vectors of attributes obtained automatically from images. We create a new large-scale taxonomy of 600 visual attributes representing more than 500 concepts and 700 K images. We use this dataset to train attribute classifiers and integrate their predictions with text-based distributional models of word meaning. We evaluate our model on its ability to simulate word similarity judgments and concept categorization. On both tasks, our model yields a better fit to behavioral data compared to baselines and related models which either rely on a single modality or do not make use of attribute-based input.

Abstract: In this paper, first we present an educational system that assists students in learning and tutors in teaching search algorithms, an artificial intelligence topic. Learning is achieved through a wide range of learning activities. Algorithm visualizations demonstrate the operational functionality of algorithms according to the principles of active learning. So, a visualization process can stop and request from a student to specify the next step or explain the way that a decision was made by the algorithm. Similarly, interactive exercises assist students in learning to apply algorithms in a step-by-step interactive way. Students can apply an algorithm to an example case, specifying the algorithm’s steps interactively, with the system’s guidance and help, when necessary. Next, we present assessment approaches integrated in the system that aim to assist tutors in assessing the performance of students, reduce their marking task workload and provide immediate and meaningful feedback to students. Automatic assessment is achieved in four stages, which constitute a general assessment framework. First, the system calculates the similarity between the student’s and the correct answer using the edit distance metric. In the next stage, it identifies the type of the answer, based on an introduced answer categorization scheme related to completeness and accuracy of an answer, taking into account student carelessness too. Afterwards, the types of errors are identified, based on an introduced error categorization scheme. Finally, answer is automatically marked via an automated marker, based on its type, the edit distance and the type of errors made. To assess the learning effectiveness of the system an extended evaluation study was conducted in real class conditions. The experiment showed very encouraging results. Furthermore, to evaluate the performance of the assessment system, we compared the assessment mechanism against expert (human) tutors. A total of 400 students’ answers were assessed by three tutors and the results showed a very good agreement between the automatic assessment system and the tutors.

Abstract: Exemplar theory assumes that people categorize a novel object by comparing its similarity to the memory representations of all previous exemplars from each relevant category. Exemplar theory has been the most prominent cognitive theory of categorization for more than 30 years. Despite its considerable success in providing good quantitative fits to a wide variety of accuracy data, it has never had a detailed neurobiological interpretation. This article proposes a neural interpretation of exemplar theory in which category learning is mediated by synaptic plasticity at cortical-striatal synapses. In this model, categorization training does not create new memory representations, rather it alters connectivity between striatal neurons and neurons in sensory association cortex. The new model makes identical quantitative predictions as exemplar theory, yet it can account for many empirical phenomena that are either incompatible with or outside the scope of the cognitive version of exemplar theory. (PsycINFO Database Record

Abstract: The popularity of research into categories has grown in recent decades and shows no sign of abating. This introductory article takes stock of the research into two facets of categorization, addressing it both as a cognitive and a social process. We advocate a rebalance toward the social process of categorization, paying more heed to the entity to be categorized, the actors involved, their acts, and the context and timing, which informs these activities. We summarize the contributions to the volume in relation to these dimensions and briefly discuss avenues for future research.

Abstract: Hashtags have become a powerful tool in social platforms such as Twitter to categorize and search for content, and to spread short messages across members of the social network. In this paper, we study temporal hashtag usage practices in Twitter with the aim of designing a cognitive-inspired hashtag recommendation algorithm we call BLLi,s. Our main idea is to incorporate the effect of time on (i) individual hashtag reuse (i.e., reusing own hashtags), and (ii) social hashtag reuse (i.e., reusing hashtags, which has been previously used by a followee) into a predictive model. For this, we turn to the Base-Level Learning (BLL) equation from the cognitive architecture ACT-R, which accounts for the time-dependent decay of item exposure in human memory. We validate BLLI,S using two crawled Twitter datasets in two evaluation scenarios. Firstly, only temporal usage patterns of past hashtag assignments are utilized and secondly, these patterns are combined with a content-based analysis of the current tweet. In both evaluation scenarios, we find not only that temporal effects play an important role for both individual and social hashtag reuse but also that our BLLI,S approach provides significantly better prediction accuracy and ranking results than current state-of-the-art hashtag recommendation methods.