Oscar: A Semantic-based Data Binning Approach

Matrix Factorisation is a recommendation approach that tries to understand what factors interest a user, based on his past ratings for items (products, movies, songs), and then use this factor information to predict future item ratings. A central limitation of this approach however is that it cannot capture how a user's tastes have evolved beforehand; thereby ignoring if a user's preference for a factor is likely to change. One solution to this is to include users' preferences for semantic (i.e. linked data) categories, however this approach is limited should a user be presented with an item for which he has not rated the semantic categories previously; so called cold-start categories. In this paper we present a method to overcome this limitation by transferring rated semantic categories in place of unrated categories through the use of vertex kernels; and incorporate this into our prior SemanticSVD + + model. We evaluated several vertex kernels and their effects on recommendation error, and empirically demonstrate the superior performance that we achieve over: (i) existing SVD and SVD + + models; and (ii) SemanticSVD + + with no transferred semantic categories.

BackgroundSemantic Category Disambiguation (SCD) is the task of assigning the appropriate semantic category to given spans of text from a fixed set of candidate categories, for example Protein to “Fibrin”. SCD is relevant to Natural Language Processing tasks such as Named Entity Recognition, coreference resolution and coordination resolution. In this work, we study machine learning-based SCD methods using large lexical resources and approximate string matching, aiming to generalise these methods with regard to domains, lexical resources and the composition of data sets. We specifically consider the applicability of SCD for the purposes of supporting human annotators and acting as a pipeline component for other Natural Language Processing systems.ResultsWhile previous research has mostly cast SCD purely as a classification task, we consider a task setting that allows for multiple semantic categories to be suggested, aiming to minimise the number of suggestions while maintaining high recall. We argue that this setting reflects aspects which are essential for both a pipeline component and when supporting human annotators. We introduce an SCD method based on a recently introduced machine learning-based system and evaluate it on 15 corpora covering biomedical, clinical and newswire texts and ranging in the number of semantic categories from 2 to 91.With appropriate settings, our system maintains an average recall of 99% while reducing the number of candidate semantic categories on average by 65% over all data sets.ConclusionsMachine learning-based SCD using large lexical resources and approximate string matching is sensitive to the selection and granularity of lexical resources, but generalises well to a wide range of text domains and data sets given appropriate resources and parameter settings. By substantially reducing the number of candidate categories while only very rarely excluding the correct one, our method is shown to be applicable to manual annotation support tasks and use as a high-recall component in text processing pipelines. The introduced system and all related resources are freely available for research purposes at: https://github.com/ninjin/simsem.

In this paper, a study on effective exploitation of geometrical features for classifying surveillance objects into a set of pre-defined semantic categories is presented. The geometrical features correspond to object’s motion, spatial location and velocity. The extraction of these features is based on object’s trajectory corresponding to object’s temporal evolution. These geometrical features are used to build a behaviour-based classifier to assign semantic categories to the individual blobs extracted from surveillance videos. The proposed classification framework has been evaluated against conventional object classifiers based on visual features extracted from semantic categories defined on AVSS 2007 surveillance dataset.KeywordsObject classificationgeometrical modelssurveillance videosobject trackingmotion features

Television news is an important medium to convey information to masses. This motivates several stakeholders to monitor and analyze the news broadcasts. Segmentation of streaming broadcast into programs and stories is a necessary first step for such analysis. Television news producers use predefined and unique presentation styles to create the channel content. Presentation styles vary with program and news story category, broadcast time, targeted audience etc. This motivated us to use presentation styles as features for segmenting news broadcasts. We propose a novel approach for characterization of spatio-temporal presentation styles. This involves characterization of spatial styles using a set of (presentation style specific) semantic shot categories derived from LSCOM-Lite Ontology. We also identify features and classifiers to automate the process of shot labeling for spatial style characterization. Further, the temporal presentation styles of shots are modeled using conditional random fields. This spatio-temporal modeling of presentation styles is used for segmenting the broadcast into programs and stories. We have also contributed a 360 hours broadcast video dataset acquired from three Indian English news channels with ground-truth marked semantic shot categories, program genres and story boundaries. Experimentations on this dataset have shown the utility of our proposal for news broadcast video segmentation.

The Internet of Things (IoT) is a paradigm that can connect an enormous number of intelligent objects, share large amounts of data, and produce new services. However, it is a challenge to select the proper sensors for a given request due to the number of devices in use, the available resources, the restrictions on resource utilization, the nature of IoT networks, and the number of similar services. Previous studies have suggested how to best address this challenge, but suffer from low accuracy and high execution times. We propose a new distributed model to efficiently deal with heterogeneous sensors and select accurate ones in a dynamic IoT environment. The model’s server uses and manages multiple gateways to respond to the request requirements. First, sensors were grouped into three semantic categories and several semantic sensor network types in order to define the space of interest. Second, each type’s sensors were clustered using the Whale-based Sensor Clustering (WhaleCLUST) algorithm according to the context properties. Finally, the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) was improved to search and select the most adequate sensor matching users’ requirements. Experimental results from real data sets demonstrate that our proposal outperforms state-of-the-art approaches in terms of accuracy (96%), execution time, quality of clustering, and scalability of clustering.

The Internet of Things (IoT) is a paradigm that can connect an enormous number of intelligent objects, share large amounts of data, and produce new services. However, it is a challenge to select the proper sensors for a given request due to the number of devices in use, the available resources, the restrictions on resource utilization, the nature of IoT networks, and the number of similar services. Previous studies have suggested how to best address this challenge, but suffer from low accuracy and high execution times. We propose a new distributed model to efficiently deal with heterogeneous sensors and select accurate ones in a dynamic IoT environment. The model’s server uses and manages multiple gateways to respond to the request requirements. First, sensors were grouped into three semantic categories and several semantic sensor network types in order to define the space of interest. Second, each type’s sensors were clustered using the Whale-based Sensor Clustering (WhaleCLUST) algorithm according to the context properties. Finally, the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) was improved to search and select the most adequate sensor matching users’ requirements. Experimental results from real data sets demonstrate that our proposal outperforms state-of-the-art approaches in terms of accuracy (96%), execution time, quality of clustering, and scalability of clustering.

We consider the problem of semantic segmentation, i.e. assigning each pixel in an image to a set of pre-defined semantic object categories. State-of-the-art semantic segmentation algorithms typically consist of three components: a local appearance model, a local consistency model and a global consistency model. These three components are generally integrated into a unified probabilistic framework. While it enables at training time a joint estimation of the model parameters and while it ensures at test time a globally consistent labeling of the pixels, it also comes at a high computational cost. We propose a simple approach to semantic segmentation where the three components are decoupled (this journal submission is an extended version of the following conference paper: G. Csurka and F. Perronnin, "A simple high performance approach to semantic segmentation", BMVC, 2008). For the local appearance model, we make use of the Fisher kernel. While this framework was shown to lead to high accuracy for image classification, to our best knowledge this is its first application to the segmentation problem. The semantic segmentation process is then guided by a low-level segmentation which enforces local consistency. Finally, to enforce image-level consistency we use global image classifiers: if an image as a whole is unlikely to contain an object class, then the corresponding class is not considered in the segmentation pipeline. The decoupling of the components makes our system very efficient both at training and test time. An efficient training enables to estimate the model parameters on large quantities of data. Especially, we explain how our system can leverage weakly labeled data, i.e. images for which we do not have pixel-level labels but either object bounding boxes or even only image-level labels. We believe that an important contribution of this paper is to show that even a simple decoupled system can provide state-of-the-art performance on the PASCAL VOC 2007, PASCAL VOC 2008 and MSRC 21 datasets.

Blending presents unique and unusual attractive words in mass media because of its creativity (Beliaeva, 2019). The study aims to: 1) identify types of blends, 2) investigate the semantic relationships of blends, and 3) examine the contribution of splinters in the formation of blends in The Manhattan Fish Market advertisements. Mattiello (2013, 2017) comprehensively classified blends into morphotactic (total and partial), morphological and graphic (overlapping and non-overlapping), and morphosemantic (attributive and coordinative). This study uses Bauer’s (2006, 2012) model to study the semantic relationship between paradigmatic and syntagmatic based on the headedness and meanings of the blend. The Manhattan Fish Market (MFM) contributes many blends inspired by other splinters (e.g., sea-, -licious, -tastic). This study analyses 142 blends to classify the syntactic and semantic categories using a descriptive qualitative research design. The findings show that MFM uses overlapping blends and syntagmatic blends to attract readers. The paradigmatic blends share the same lexical class, while syntagmatic blends share different lexical categories. The subtypes of syntagmatic blends are right-headed (modifier + head) and left-headed (head + modifier). The findings help identify the semantic relations based on the headedness in blends and the contribution of splinters in the formation of blends. The paper concludes that headedness determines the semantic types and function of blends.Keywords: blends, Manhattan Fish Market, headedness, syntagmatic, paradigmatic, splinter

Dynamic categorization of clinical research eligibility criteria by hierarchical clustering

The study explores the fusion of semantic roles and the different semantic fusion types, aiming at establishing a semantic fusion model to explain the cognitive alignment of events in Chinese and English simple sentence constructions containing two verbs. In total, 20,280 simple sentence constructions containing two verbs are collected from Chinese literary works, Peking University Chinese Corpus, and English classic literary works. The semantic fusion in the collected simple sentence constructions containing two verbs is classified into five major semantic fusion categories, which appear with different occurrence frequencies in the two languages. The semantic fusion model of event alignment is comprehensively supported by linguistic research in Chinese and English. From a cognitive linguistic perspective, it is found that the double semantic profiles of the same syntactic element N (noun) make N psychologically activated twice and enable it to enter two processes profiled by the two verbs as a participant. The two processes are combined into one event, which designates a cognitive occurrence of any degree of complexity. N’s entry into the two subevents is realized by its double semantic profiles that enable it to fuse two semantic roles into one syntactic element and explain the relationship between N’s double syntactic identities and double semantic roles. The semantic fusion model was used to explore event alignment in simple sentence constructions containing two verbs, and it was discovered that the fusion of two semantic roles is universal in languages and is a common psychological and cognitive behavior deeply rooted in the mental conceptualization of language users. The empirical discussion of simple sentence constructions containing two verbs proves that semantic fusion as an important psychological passage in event alignment has solid psychological reality and verifies the applicability of the semantic fusion model in the explanation of event alignment.

Should research institutions from developing countries provide open access to their data? Who would benefit? These are highly relevant questions for environmental research. In the backdrop of increasing discussions regarding open data, attention should be paid to guarantee that research institutions from developing countries do not become mere data providers. At the same time, restricted access to data should not become a barrier for advancing our knowledge of critical environmental issues. Open access policies for environmental data must be developed not only for advancing science, but also to provide growing opportunities for researchers in developing countries.

To describe processes used in the Healthy Brain project to manage data collection, coding, and data distribution in a large qualitative project, conducted by researchers at 9 universities in 9 states. Project management protocols included: (a) managing audiotapes and surveys to ensure data confidentiality, data tracking and distribution; (b) managing qualitative data to ensure the accuracy and confidentiality of transcription; (c) training in qualitative methods and use of qualitative software; and (d) managing participant survey data and analysis. The project team coded and managed qualitative and survey data for 69 focus groups with more than 500 participants. Multiple interactive training sessions in qualitative data analysis and use of qualitative software (ATLAS.ti) were conducted. To develop a codebook, 2 teams used an open-coding process to identify codes and develop definitions; 2 team members integrated and conceptually organized these results into an initial codebook. For the audio-recordings from each research site, 2 or 3 team members hand coded 1 transcript and calculated interrater agreement (.80 or higher). Using clear protocols, participatory training sessions, team-based coding, and frequent communication among team members via e-mail and regular in-person meetings promotes effective management of large-scale qualitative research projects.

Tell me, I'll forget,Show me, I may remember,Involve me, I'll understand. - ConfuciusThis proverb describes the essence of our paper and the motivation behind the development of the Interactive Parallelization Tool (IPT) that can transform serial applications into multiple parallel variants. The end-users of IPT must have an understanding of the basic concepts involved in parallel programming (e.g., data distribution and data gathering). After developing an understanding of the basic parallel programming concepts, IPT can be used by its target audience (domain-experts and students) to semi-automatically generate parallel programs using multiple parallel programming paradigms (MPI, OpenMP, and CUDA), and learn about these paradigms through observation and comparison. This IPT-based personalized learning approach complements the traditional methods of learning and training that usually emphasize the syntax and semantics of one or more programming standards. The main benefit of IPT is that it provides a jumpstart to the domain-experts in using modern HPC platforms for their research and development needs, and hence lowers the adoption barriers to HPC.

The DAta Food NEtworking (DAFNE) initiative was conceived in the 1980s and aims at using data already collected in the context of household budget surveys (HBSs) for the assessment of dietary patterns across Europe. HBSs are country-representative surveys conducted, with similar methodology, at regular time intervals by the National Statistical Offices of almost all European countries. DAFNE focuses at the designation of comparable, between European countries, categories of food and socio-demographic data, and the estimation of daily per capita food availability. With food and socio-demographic data classified into common categories, comparisons of food habits between European countries become possible. Fifteen European countries with one or more HBS all undertaken during the period 1981-1999. Representative population samples in 15 European countries. DAFNE HBS data are comparable and allow a wide range of dietary analyses. Preliminary data document differences in food habits among European countries and provide insights into the socio-demographic determinants of food preferences. The DAFNE databank, if properly expanded and exploited, could become a valuable tool for national food and nutrition policy planning across Europe and for the identification of groups at higher risk for developing nutrition-related conditions.

Scarce data are currently available about the distribution of the Mediterranean medicinal leech Hirudo verbana in Italy, and most of the known occurrence localities are based on records collected in the nineteenth and the first half of the twentieth century, which were not confirmed in the last decades, mostly due to a lack of surveys. Accordingly, the available knowledge on the distribution of the species is far from being updated and representative, although a significant negative trend of H. verbana throughout the country is supposed. The lack of sound distribution data is a significant shortfall, which hinders the implementation of the monitoring of the species as required by the Article 17 of the “Habitats Directive” on the species of Union concern. To provide recent, validated distributional data for the Mediterranean medicinal leech in Italy to be used as baseline data for further surveys and monitoring, we present herein a set of unpublished observations collected in the last decades in peninsular Italy, Sicily, and Sardinia. Moreover, we report observation of H. verbana feeding on amphibians, a feeding habit to date poorly documented for the Mediterranean medicinal leech. Based on both published and novel data, H. verbana is characterised by a large but fragmented distribution in Italy. Therefore, dedicated monitoring programs and the collection of validated occasional observations are highly desirable to get a clearer picture of the real distribution of the species.