Search

Many studies have exposed the massive collection of personal data in the digital ecosystem through, for instance, websites, mobile apps, or smart devices. This fact goes unnoticed by most users, who are also unaware that the collectors are sharing their personal data with many different organizations around the globe. This paper assesses techniques available in the state of the art to identify the organizations receiving this personal data. Based on our findings, we propose Receiver Organization Identifier (ROI), a fully automated method that combines different techniques to achieve a 95.71% precision score in identifying an organization receiving personal data. We demonstrate our method in the wild by evaluating 10,000 Android apps and exposing the organizations that receive users' personal data. We further assess the transparency of these data-sharing practices by analyzing the apps' privacy policies. The results reveal a concerning lack of transparency in almost 78% of apps, suggesting the need for regulators to take action. [ABSTRACT FROM AUTHOR]

The splitting of words into stressed and unstressed syllables is the foundation for the scansion of poetry, a process that aims at determining the metrical pattern of a line of verse within a poem. Intricate language rules and their exceptions, as well as poetic licenses exerted by the authors, make calculating these patterns a nontrivial task. Some rhetorical devices shrink the metrical length, while others might extend it. This opens the door for interpretation and further complicates the creation of automated scansion algorithms useful for automatically analyzing corpora on a distant reading fashion. In this paper, we compare the automated metrical pattern identification systems available for Spanish, English, and German, against fine-tuned monolingual and multilingual language models trained on the same task. Despite being initially conceived as models suitable for semantic tasks, our results suggest that transformers-based models retain enough structural information to perform reasonably well for Spanish on a monolingual setting, and outperforms both for English and German when using a model trained on the three languages, showing evidence of the benefits of cross-lingual transfer between the languages. [ABSTRACT FROM AUTHOR]

In the present era, there is a large amount of new data available readily from different sources to collect and store. One of the main problems is to label these new data for various machine learning applications correctly. The active learning approach presents a unique case of machine learning which is widely used to solve the above problem by significantly minimizing the need for labeled data. It aims to select the most appropriate samples from the unlabeled data to be correctly labeled by the oracle and is passed to train the active learner incrementally. There are several different query sampling strategies that exist using which the appropriate samples are selected. One of the main problems with the active learning approach is that it is very time-consuming. So in this research work, a new multi-core-based algorithm is proposed to speed up the active learning approach, which can utilize the complete computational resources present in the system. The experiments have been performed for the problem of named entity recognition which deals with labeling the sequences of words in an unstructured text by classifying them into pre-existing categories. The proposed algorithm is evaluated in terms of both: the performance and execution time over three named entity recognition corpus of distinct biomedical domains. The evaluation results shows considerable improvement in terms of execution time for the proposed active learning algorithm than the existing active learning approach. [ABSTRACT FROM AUTHOR]

Training machine learning models in a meaningful order, from the easy samples to the hard ones, using curriculum learning can provide performance improvements over the standard training approach based on random data shuffling, without any additional computational costs. Curriculum learning strategies have been successfully employed in all areas of machine learning, in a wide range of tasks. However, the necessity of finding a way to rank the samples from easy to hard, as well as the right pacing function for introducing more difficult data can limit the usage of the curriculum approaches. In this survey, we show how these limits have been tackled in the literature, and we present different curriculum learning instantiations for various tasks in machine learning. We construct a multi-perspective taxonomy of curriculum learning approaches by hand, considering various classification criteria. We further build a hierarchical tree of curriculum learning methods using an agglomerative clustering algorithm, linking the discovered clusters with our taxonomy. At the end, we provide some interesting directions for future work. [ABSTRACT FROM AUTHOR]

Peer review is one of the important procedures to determine which research proposals are to be funded and to evaluate the quality of scientific research. How to find suitable reviewers for scientific research proposals is an important task for funding agencies. Traditional methods for reviewer recommendation focus on the relevance of the proposal and knowledge of candidate reviewers by mainly matching the keywords or disciplines. However, the sparsity of keyword space and the broadness of disciplines lead to inaccurate reviewer recommendations. To overcome these limitations, this paper introduces a reviewer recommendation method (RRM) for scientific research proposals. This research applies word embedding to construct vector representation for terms, which provides a semantic and syntactic measurement. Further, we develop representation models for reviewers' knowledge and proposals, and recommend reviewers by matching two representation models incorporating ranking fusions. The proposed method is implemented and tested by recommending reviewers for scientific research proposals of the National Natural Science Foundation of China. This research invites reviewers to provide feedback, which works as the benchmark for evaluation. We construct three evaluation metrics, Precision, Strict-precision, and Recall. The results show that the proposed reviewer recommendation method highly improves the accuracy. Research results can provide feasible options for the decision-making of the committee, and improve the efficiency of funding agencies. [ABSTRACT FROM AUTHOR]

Using software applications or services, which provide word or even word pattern recommendation service has become part of our lives. Those services appear in many form in our daily basis, just think of our smartphones keyboard, or Google search suggestions and this list can be continued. With the help of these tools, we can not only find the suitable word that fits into our sentence, but we can also express ourselves in a much more nuanced, diverse way. To achieve this kind of recommendation service, we use an algorithm which is capable to recommend word by word pattern queries. Word pattern queries, can be expressed as a combination of words, part-of-speech (POS) tags and wild card words. Since there are a lot of possible patterns and sentences, we use Big Data frameworks to handle this large amount of data. In this paper, we compared two popular framework Hadoop and Spark with the proposed algorithm and recommend some enhancement to gain faster word pattern generation.

This paper proposes an approach to measure the credibility of solutions of inverse problems with incomplete or missing data, encountered in some physical problems. In such problems, the same set of data can produce multiple solutions, depending on the constraints or assumptions used to compensate for the missing information. The actual “true” solution, against which the quality of a constrained solution can be measured, is not usually available. In this work, we propose to obtain a complete but coarse reference solution that reliably possesses attributes of the actual solution. This is done by solving an over-complete problem with the same set of data but with a coarser structure. A number of residual, composition and fidelity metrics are then used to quantitatively measure the quality of a solution obtained from incomplete data, against the coarse complete reference. The approach was tested with various degrees of incompleteness and different levels of uncertainty (noise) in data in the linear problem of image reconstruction in computed-tomography. The results show that the approach was successful in identifying credible solutions, as well as those that were not trustworthy.

Quantification of knowledge technologies has long posed a challenge to the measurement of cognitive proximity. This paper proposes a method to measure cognitive proximity by mining patent description text with the LDA topic model. With the patent-topic distribution got from the LDA topic model, the cognitive proximity is measured between enterprises or within cities, which could make up for the shortage of existing measurement methods limited by the rigid IPC, industry classification system, or non-standard interview data. Our empirical studies on the ICT industry indicate that the 20 topics obtained through the topic model have a good correspondence with the technologies involved in this industry's leading products and services. And we dig out the knowledge and technology information in the patent text to depict the technology landscape, including mining the changes of technology topics over time, the difference of distribution in various cities, and the development trend of the urban innovation network. This method's effectiveness is also proved in the model that compares different measurement methods when revealing the relationship between cognitive proximity and patent productivity. Last, researchers can use this approach to delve deeper into urban innovation issues, and policymakers can use it to figure out further innovation. [ABSTRACT FROM AUTHOR]

This paper explores a multi-strategy technique that aims at enriching text documents for improving clustering quality. We use a combination of entity linking and document summarization in order to determine the identity of the most salient entities mentioned in texts. To effectively enrich documents without introducing noise, we limit ourselves to the text fragments mentioning the salient entities, in turn, belonging to a knowledge base like Wikipedia, while the actual enrichment of text fragments is carried out using WordNet. To feed clustering algorithms, we investigate different document representations obtained using several combinations of document enrichment and feature extraction. This allows us to exploit ensemble clustering, by combining multiple clustering results obtained using different document representations. Our experiments indicate that our novel enriching strategies, combined with ensemble clustering, can improve the quality of classical text clustering when applied to text corpora like The British Broadcasting Corporation (BBC) NEWS.

In recent internet era, micro-blogging sites produce enormous amount of short textual information, which appears in the form of opinions or sentiments of users. Sentiment analysis is a challenging task in short text, due to use of formal language, misspellings, and shortened forms of words, which leads to high dimensionality and sparsity. In order to deal with these challenges, this paper proposes a novel, simple, and yet effective feature selection method, to select frequently distributed features related to each class. In this paper, the feature selection method is based on class-wise information, to identify the relevant feature related to each class. We evaluate the proposed feature selection method by comparing with existing feature selection methods like chi-square ( χ2), entropy, information gain, and mutual information. The performances are evaluated using classification accuracy obtained from support vector machine, K nearest neighbors, and random forest classifiers on two publically available datasets viz., Stanford Twitter dataset and Ravikiran Janardhana dataset. In order to demonstrate the effectiveness of the proposed feature selection method, we conducted extensive experimentation by selecting different feature sets. The proposed feature selection method outperforms the existing feature selection methods in terms of classification accuracy on the Stanford Twitter dataset. Similarly, the proposed method performs competently equally in terms of classification accuracy compared to other feature selection methods in most of the feature subsets on Ravikiran Janardhana dataset.

Music information retrieval has lately become an important field of information retrieval, because by profound analysis of music pieces important information can be collected: genre labels, mood prediction, artist identification, just to name a few. The lack of large-scale music datasets containing audio features and metadata has lead to the construction and publication of the Million Song Dataset (MSD) and its satellite datasets. Nonetheless, mainly because of licensing limitations, no freely available lyrics datasets have been published for research.

With the tremendous growth in the number of scientific papers being published, searching for references while writing a scientific paper is a time-consuming process. A technique that could add a reference citation at the appropriate place in a sentence will be beneficial. In this perspective, the context-aware citation recommendation has been researched for around two decades. Many researchers have utilized the text data called the context sentence, which surrounds the citation tag, and the metadata of the target paper to find the appropriate cited research. However, the lack of well-organized benchmarking datasets, and no model that can attain high performance has made the research difficult. In this paper, we propose a deep learning-based model and well-organized dataset for context-aware paper citation recommendation. Our model comprises a document encoder and a context encoder. For this, we use graph convolutional networks layer, and bidirectional encoder representations from transformers, a pre-trained model of textual data. By modifying the related PeerRead dataset, we propose a new dataset called FullTextPeerRead containing context sentences to cited references and paper metadata. To the best of our knowledge, this dataset is the first well-organized dataset for a context-aware paper recommendation. The results indicate that the proposed model with the proposed datasets can attain state-of-the-art performance and achieve a more than 28% improvement in mean average precision and recall@k. [ABSTRACT FROM AUTHOR]

Research and development (R&D) in many technological areas is characterized by growing complexity. In biomedical engineering, too, interdisciplinary collaboration is regarded as a promising way to master this challenge. Therefore, identifying suitable experts becomes crucial, which is currently being researched, amongst others, by analyzing semantic data. However, previous approaches lack clarity and traceability of the mechanisms for compiling top-n lists of recommended experts, as domain specificity in profiling is insufficient. Moreover, these recommenders are mainly based on scientific publications, while patents are rarely considered as an important outcome of R&D. Thus, we study the feasibility of profiling 16 biomedical engineering experts using both publications and patents. These documents are automatically labeled according to a three-dimensional domain model by machine learning-based classifiers. On this basis, we created various activity-based representations, including author-contribution-weighting. We evaluated the profiling through self- and external-assessments and tested the recommendation compared to scientometric measures in three case studies. All interviewed experts identify themselves among 10 pseudonymous profiles and 96% of all 51 external-assignments are correct. The recommendation over three case studies reaches a high mean average precision of 89% and contrasts with the use of scientometric measures (41%). Moreover, the activity based on patents primarily corresponds to that of publications but patents also introduce new activities. The author-contribution-weighting improves the performance. In conclusion, our findings show that exploiting publications and patents enables comprehensible profiling of biomedical engineering experts that allows visual comparisons and clear selection and ranking of potential R&D collaboration partners along the translational value chain. [ABSTRACT FROM AUTHOR]

The analysis of scientific and technical documents is crucial in the process of establishing science and technology strategies. One popular method for such analysis is for field experts to manually classify each scientific or technical document into one of several predefined technical categories. However, not only is manual classification error-prone and expensive, but it also requires extended efforts to handle frequent data updates. In contrast, machine learning and text mining techniques enable cheaper and faster operations, and can alleviate the burden on human resources. In this paper, we propose a method for extracting embedded feature vectors by applying a neural embedding approach for text features in patent documents and automatically clustering the embedding features by utilizing a deep embedding clustering method. [ABSTRACT FROM AUTHOR]

With the increase of translation demand, the advancement of information technology, the development of linguistic theories and the progress of natural language understanding models in artificial intelligence research, machine translation has gradually gained worldwide attention. However, at present, machine translation research still has problems such as insufficient bilingual data and lack of effective feature representation, which affects the further improvement of key modules of machine translation such as word alignment, sequence adjustment and translation modelling. The effect of machine translation is still unsatisfactory. As a new machine learning method, deep neural network can automatically learn abstract feature representation and establish a complex mapping relationship between input and output signals, which provides a new idea for statistical machine translation research. Firstly, the multi-layer neural network and the undirected probability graph model are combined, and the similarity and context information of vocabulary are effectively utilized to model the word alignment more fully, and the word alignment model named NNWAM is constructed. Secondly, the low dimension will be used. The feature representation is combined with other features into a linearly ordered pre-ordering model to construct the pre-ordering model named NNPR. Finally, the word alignment model and the pre-ordering model are combined in the same deep neural network framework to form DNNAPM, a statistical machine translation model based on deep neural networks. The experimental results show that the statistical machine translation model based on deep neural network has better effect, faster convergence and better reliability than the comparison model algorithm. [ABSTRACT FROM AUTHOR]

Knowing driving factors and understanding researcher behaviors from the dynamics of collaborations over time offer some insights, i.e. help funding agencies in designing research grant policies. We present longitudinal network analysis on the observed collaborations through co-authorship over 15 years. Since co-authors possibly influence researchers to have interest changes, by focusing on researchers who could become the influencer, we propose a stochastic actor-oriented model of bipartite (two-mode) author-topic networks from article metadata. Information of scientific fields or topics of article contents, which could represent the interests of researchers, are often unavailable in the metadata. Topic absence issue differentiates this work with other studies on collaboration dynamics from article metadata of title-abstract and author properties. Therefore, our works also include procedures to extract and map clustered keywords as topic substitution of research interests. Then, the next step is to generate panel-waves of co-author networks and bipartite author-topic networks for the longitudinal analysis. The proposed model is used to find the driving factors of co-authoring collaboration with the focus on researcher behaviors in interest changes. This paper investigates the dynamics in an academic social network setting using selected metadata of publicly-available crawled articles in interrelated domains of "natural language processing" and "information extraction". Based on the evidence of network evolution, researchers have a conformed tendency to co-author behaviors in publishing articles and exploring topics. Our results indicate the processes of selection and influence in forming co-author ties contribute some levels of social pressure to researchers. Our findings also discussed on how the co-author pressure accelerates the changes of interests and behaviors of the researchers. [ABSTRACT FROM AUTHOR]

Diabetic is becoming a very serious disease today for the most of people all over the world due to the unhealthy food habits. For predicting the diabetes, we introduce a new diabetic diagnosis system which combines a newly proposed temporal feature selection and temporal fuzzy ant miner tree (TFAMT) classifier for effective decision making in type-2 diabetes analysis. Moreover, a new temporal weighted genetic algorithm is proposed in this work for enhancing the detection accuracy by preprocessing the text and image data. Moreover, intelligent fuzzy rules are extracted from the weighted temporal capabilities with ant miner fuzzy decision tree classifier, and then fuzzy rule extractor is used to reduce the variety of functions in the extracted regulations. We empirically evaluated the effectiveness of the proposed TFAMT-TWGA model using the UCI Repository dataset and the collected retinopathy image dataset. The outcomes are analyzed and as compared with other exiting works. Furthermore, the detection accuracy is proven by way of using the ten-fold cross validation. [ABSTRACT FROM AUTHOR]

The manuscript provides a novel method for language identification using the texture analysis of the script. The method consists of mapping each letter from the text with certain script type. It is made according to characteristics concerning the position of the letter in the baseline area. In order to extract features, the co-occurrence matrix is computed. Then, the texture features are calculated. Extracted measures show meaningful differences due to dissimilarities in the script and language characteristics. It represents a basis in a decision-making process of the language identification. Feature classification is performed by the extension of a state-of-the-art method called genetic algorithms image clustering for document analysis. The proposed method is tested on an example of documents given in English, French, Slovenian and Serbian languages and compared to other well-known classification methods and feature representations in the state of the art. The results of experiments show the superiority of the proposed approach. [ABSTRACT FROM AUTHOR]

We present a new quantitative method for detecting changes in 3D medical images. The dissimilarity between shapes is quantified as a measure of the effort it takes to deform one 3D region into another. Our main tool is an assessment of conformal and isometric distortions of mappings between volumes. Unlike most existing techniques for shape comparison, our algorithm operates both on triangular and tetrahedral meshes, and therefore can be applied both for closed simply connected surfaces, as well as for volumetric domains homeomorphic to a ball, with geometrically complicated boundaries. Furthermore we extend our main geometric distortion measure to higher dimensions, in a manner that allows for the dealing with spatial data at the maximal, as well as at all the lower dimensions. [ABSTRACT FROM AUTHOR]

Scientific communities are clusters of researchers and play important roles in modern science. Studying different forms of scientific communities that either physically or virtually exist is a feasible way to disclose underlying mechanisms of science. From the perspective of complex networks, topology-based communities and topic-based communities reflect scientific collaboration and topical features of science respectively. However, the two features are not isolated but intertwined in scientific practice. This study proposes an approach to detect Topical Scientific Communities (TSCs) with both topology and topic features by applying machine learning techniques and network theory. As an example, the TSCs of the informetrics field are detected, and then the characteristics of these TSCs are analyzed. It is shown that collaboration patterns on the topic level can be revealed by analyzing the static network structure and dynamics of TSCs. Furthermore, cross-topic collaborations at multiple levels could be investigated through TSCs. In addition, TSCs can effectively organize researchers in terms of productivity. Future work will further explore and generalize characteristics of TSCs, and the applications of TSCs to other tasks of studying science. [ABSTRACT FROM AUTHOR]

Currently, there exists a big gap between formal computer-understandable mathematics and informal mathematics, as written by humans. When looking more closely, there are two important subproblems: making documents written by humans at least syntactically understandable for computers, and the formal verification of the actual mathematics in the documents. In this paper, we will focus on the first problem. For the time being, most authors use , , or one of its graphical front-ends in order to write documents with many mathematical formulas. In the past decade, we have developed an alternative wysiwyg system GNU , which is not based on . All these systems are only adequate for visual typesetting and do not carry much semantics. Stated in the MathML jargon, they concentrate on presentation markup, not content markup. In recent versions of , we have started to integrate facilities for the semantic editing of formulas. In this paper, we will describe these facilities and expand on the underlying motivation and design choices. To go short, we continue to allow the user to enter formulas in a visually oriented way. In the background, we continuously run a packrat parser, which attempts to convert (potentially incomplete) formulas into content markup. As long as all formulas remain sufficiently correct, the editor can then operate both on a visual or semantic level, independently of the low-level representation being used. An important related topic, which will also be discussed at length, is the automatic correction of syntax errors in existing mathematical documents. In particular, the syntax corrector that we have implemented enables us to upgrade existing documents and test our parsing grammar on various books and papers from different sources. We will provide a detailed analysis of these experiments. [ABSTRACT FROM AUTHOR]