Semantic Kernel Research Articles

A K‐Net‐based deep learning framework for automatic rock quality designation estimation

AbstractRock quality designation (RQD) plays a crucial role in the design and analysis of rock engineering. The traditional method of measuring RQD relies on manual logging by geologists, which is often labor‐intensive and time‐consuming. Thus, this study presents an autonomous framework for expeditious RQD estimation based on two‐dimensional corebox photographs. The scale‐invariant feature transform (SIFT) algorithm is employed for rapid image calibration. A K‐Net‐based model with dynamic semantic kernels, conditional on their actual activations, is proposed for rock core segmentation. It surpasses other prevalent models with a mean intersection over union of 95.43%. The automatic RQD estimation error of our proposed framework is only 1.46% compared to manual logging results, demonstrating its exceptional reliability and effectiveness. The robustness of the framework is then validated on an additional test set, proving its potential for widespread adoption in geotechnical engineering practice.

Integrating Web Applications with Azure OpenAI Services: A Focus on Semantic Kernel

Integrating Web Applications with Azure OpenAI Services: A Focus on Semantic Kernel

A K-Net-based hybrid semantic segmentation method for extracting lake water bodies

Lakes have a crucial impact on natural disaster prevention, resource recycling, maintenance of agricultural production and daily life. The traditional way of acquiring lake water body information lacks efficiency, is dangerous, and is not suitable for lake water body information acquisition and real-time monitoring. For this reason, the automated lake water body extraction method based on deep learning semantic segmentation model is gradually becoming a mainstream method. However, most of the semantic segmentation models used for lake extraction today express features through static semantics, while ignoring the extraction relationships of different convolutional kernels for these features. In order to better extract lake water bodies from remote sensing images, this paper proposes a hybrid semantic segmentation method based on K-Net, which achieves high accuracy extraction of lake water bodies by introducing dynamic semantic kernels to iteratively refine the feature information. The superiority of the K-Net-based hybrid model on a Google remote sensing image dataset of lakes is validated. The experimental results show that (1) the hybrid model is able to achieve accurate extraction of lake water bodies, with the UperNet + K-Net model using Swin-l performing the best among all six evaluation metrics, with mean intersection over union (mIoU) reaching 97.77%; and that (2) after incorporating the K-Net module, all tested models obtain a larger mIoU than before.

О методологических трудностях работы с понятием «идеология»

The article discusses the methodological issues involved in using the concept of “ideology”, which arise from its complex character. “Ideology” cannot be easily reduced to one single meaning, as far as it bears trails of multiple (and sometimes opposite) meanings that it had got through its long intellectual his­tory. As a result, this concept has a common semantic kernel and varying pe­riphery. While using this concept, one needs not only to stick firmly on the se­lected definition, but also to be aware about the theoretical debates that shaped this particular configuration of meanings. The article argues that the definition of ideology as “a system of ideas”, that was proposed by A.V. Rubtsov, is in­sufficient to address the major theoretical junctions of interpreting ideology. In particular, it does not take into account the social and political functions of ideology and exaggerates its systemic characteristics. The article supports Rubtsov’s idea of differentiating ideology as a system of ideas and as a body of institutions and proposes an alternative concept, that also focuses on the ways ideas function in social contexts characterized by relationships of power and inequality, but is not that much loaded with contradictory meanings – the concept of symbolic politics.

Lightweight Salient Object Detection in Optical Remote-Sensing Images via Semantic Matching and Edge Alignment

Recently, relying on convolutional neural networks (CNNs), many methods for salient object detection in optical remote-sensing images (ORSI-SOD) are proposed. However, most methods ignore the number of parameters and computational cost brought by CNNs, and only a few pay attention to portability and mobility. To facilitate practical applications, in this article, we propose a novel lightweight network for ORSI-SOD based on semantic matching and edge alignment, termed SeaNet. Specifically, SeaNet includes a lightweight MobileNet-V2 for feature extraction, a dynamic semantic matching module (DSMM) for high-level features, an edge self-alignment module (ESAM) for low-level features, and a portable decoder for inference. First, the high-level features are compressed into semantic kernels. Then, semantic kernels are used to activate salient object locations in two groups of high-level features through dynamic convolution operations in DSMM. Meanwhile, in ESAM, cross-scale edge information extracted from two groups of low-level features is self-aligned through <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$L_{2}$ </tex-math></inline-formula> loss and used for detail enhancement. Finally, starting from the highest level features, the decoder infers salient objects based on the accurate locations and fine details contained in the outputs of the two modules. Extensive experiments on two public datasets demonstrate that our lightweight SeaNet not only outperforms most state-of-the-art lightweight methods, but also yields comparable accuracy with state-of-the-art conventional methods, while having only 2.76 M parameters and running with 1.7 G floating point operations (FLOPs) for <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$288 \times 288$ </tex-math></inline-formula> inputs. Our code and results are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/MathLee/SeaNet</uri> .

EXAMPLE OF USE OF VIRTUAL PROMOTION MAP

The paper presents an example of using new technology to solve the problem of promoting goods and services according to the 7P concept. The technology was called virtual promotion. The reason for its occurrence is a decrease in the effectiveness of the currently classical search engine optimization methodology on the Internet. The task of the latter is to bring the given corporate WEB resource to the first place in the search server’s answers to the requests of potential buyers. Virtual promotion is based on two ideas. The first is based on marketing theory when to sell a product or service, you need to build a marketing channel. The second is that a search server is an intelligent machine. Then, to get the first position, we need to teach the virtual space, which consists of Internet nodes, to activate in such a way that the number of links to the WEB page with the desired product is maximum. To implement these ideas, two objects are synthesized as part of the virtual promotion: a semantic kernel and a promotion map. A semantic kernel is a description of a product or service or a training sequence of keywords. A map is a set of nodes on the Internet, where the semantic core should be placed to attract the maximum number of buyers of the product. The map is also a variant of marketing channels on the Internet. The paper considers an example of a situation where a semantic core and a minimal map have already been created using the classical approach (search optimization). In this case, the classic approach gave an increase in the number of visits to the WEB site, but there was not a single buyer of the product for six months. That is, the classical approach improved the key performance indicator, but it did not affect the sales performance in any way. The paper demonstrates how virtual promotion forms such a modification of the map that gives a given result in a given period. At the same time, it is not enough to synthesize a new map. It is necessary to apply a special mechanism for its implementation. For this purpose, the paper shows a real task of inter-functional coordination, which makes it possible to implement a promotion map. Real data for 2021, which was used to promote the product on the Ukrainian market, is given. The introduction of the new version of the map resulted in the first two orders within the first two weeks.

METHOD OF SOLVING THE PROBLEM OF SITUATIONAL MANAGEMENT OF THE SEMANTIC KERNEL OF WEB CONTENT

Research conducted over the past ten years in the field of search engine optimization on the Internet shows the creation of a new phenomenon - virtual promotion [1]. Its main goal is to increase the level of sales of goods through technologies that exist in virtual space. In terms of properties, virtual promotion is similar to a logistics channel, but with the characteristics of a marketing channel. That is, virtual promotion can be described as the establishment of a channel for the transmission of a special message from the company to a potential buyer of its goods or services. This article proposes to consider the problem of creating and managing such a message in cyberspace. To do this, it is proposed to solve the problem of situational management of the semantic kernel of web content. At the first stage the review of properties of a semantic kernel is carried out. This analysis allows us to formulate the problem of situational management of the semantic kernel. The main idea of ​​this problem is that the current situation is a variant of the semantic kernel of web content. Then the paper proposes a mathematical model for estimating the current situation and choosing the target situation. Conditions are also proposed to determine the need for transition between situations. The article also describes the algorithmic support for the implementation of the method of solving the problem of situational management. The result of the algorithm is K-applicant, ie the semantic kernel that describes the class of needs, which was identified at the stage of clustering. The prospect for further research is the software implementation of this algorithm on the NodeJS platform.

MDSCMF: Matrix Decomposition and Similarity-Constrained Matrix Factorization for miRNA-Disease Association Prediction.

MicroRNAs (miRNAs) are small non-coding RNAs that are related to a number of complicated biological processes, and numerous studies have demonstrated that miRNAs are closely associated with many human diseases. In this study, we present a matrix decomposition and similarity-constrained matrix factorization (MDSCMF) to predict potential miRNA–disease associations. First of all, we utilized a matrix decomposition (MD) algorithm to get rid of outliers from the miRNA–disease association matrix. Then, miRNA similarity was determined by utilizing similarity kernel fusion (SKF) to integrate miRNA function similarity and Gaussian interaction profile (GIP) kernel similarity, and disease similarity was determined by utilizing SKF to integrate disease semantic similarity and GIP kernel similarity. Furthermore, we added L2 regularization terms and similarity constraint terms to non-negative matrix factorization to form a similarity-constrained matrix factorization (SCMF) algorithm, which was applied to make prediction. MDSCMF achieved AUC values of 0.9488, 0.9540, and 0.8672 based on fivefold cross-validation (5-CV), global leave-one-out cross-validation (global LOOCV), and local leave-one-out cross-validation (local LOOCV), respectively. Case studies on three common human diseases were also implemented to demonstrate the prediction ability of MDSCMF. All experimental results confirmed that MDSCMF was effective in predicting underlying associations between miRNAs and diseases.

Запечатленная память: социологический анализ практик коммеморации

Article is devoted to judgment of a phenomenon of cultural memory as steady system of the significant ideas of the past broadcast in generally - symbolical forms and generating social valuable behavioral models. Cultural memory acts as the central semantic kernel of a large number of both foreign, and domestic researches which all variety can be divided into three types conditionally: “archival” (memory understanding as conservative of socially demanded information in a format of social experience); “active” (memory understanding as mechanism of the translator of social experience); “integrative” (the uniting both approaches to interpretation of essence of memory). One of forms of an objectification of content of cultural memory is the commemoration understood as set public collective the practician, the values and behavior models directed to formation through ritually issued deduction and reproduction in the relevant culture of the significant for group, symbolically expressed ideas of the past. Intrinsic lines of a commemoration it is possible to call: collective character, publicity (representation of interests of a certain group), institutionality (purposeful organization), emotional involvement of participants into memorial action. Structurally the commemoration can be presented as follows: commemorative kernel (reason for the act of commemoration), commemorative symbol (valuable “loop” of a commemorative kernel) and commemorative function (target vector of a commemoration). Functions of integration, maintenance of identity, socialization, visualization, compensation act as commemorative functions. The variety commemorative practician can be systematized by their tipologization on various to the bases: on existence duration: short-term and long-term; on subject of a commemorative kernel (personalized, it is event - procedural, topographical); in a form (a parade, a procession, assignment of a name, opening of a commemorative plaque, erection of a monument, etc.); on chronological localization of a commemorative kernel (retrofocused and updated); by origin (authentic and borrowed); on distribution scale (global, local, regional, subcultural); on an emotional message (constructive and destructive); on the nature of action (procedural and static); by the origin nature (natural and artificial), etc. The sociological analysis three commemorative practician - names of streets, opening of commemorative plaques and monuments on the example of Chelyabinsk is presented in this article (the analyzed period from 1991 to 2016). Authors compare the territorial and chronological bases of names, types of their representation (the personality, an event, the place), reveal the most popular subject and figurative and symbolical designs of commemoration. As the general conclusion following the results of a research it is possible to note the following. In terms of territorial localization commemorative practicians are mainly connected with local history (the regional focused character); temporal - with the XX century, mainly Soviet period (the retro focused type); substantial - with reflection of achievements of outstanding persons (the personal focused type).

METRICS OF VIRTUAL PROMOTION OF A PRODUCT

An approach to the mathematical description of the criterion for the effectiveness of a new object of research – virtual promotion is presented in thepaper. The emergence of this new object of research is connected, on the one hand, with the classical theory of marketing, and on the other withmodern Internet technologies. Marketing is based on the 4P principle: product, price, location and promotion. Promotion is a component of thisprinciple. But in modern conditions, this phenomenon is changing under the influence of the Internet. Now this 4P component is becoming a fullyvirtual instrument. The traditional scheme of promotion functioning is as follows. A message is created to a potential buyer and the delivery channel ofthis message undergoes a change. It is based on the principle: money – goods – money. While the new sales scheme is described by the scheme: weattract a client, make money on a client, we spend money. In the new scheme, we deal with product knowledge in the form of the so-called semanticcore of web content. Knowledge describes for a potential client how a given product can cover his need for something. Using the logistic principles ofthe transfer of goods, this semantic core is loaded into the specified Internet nodes. That is, virtual promotion is formed as two channels: logistics andmarketing. The first one performs three operations: concentration, formatting and distribution of semantic cores on the Internet. The second managesthis process, forming a virtual promotion map. This map is a graph of Internet nodes. It is required to define such a tree of Internet nodes so that virtualpromotion has maximum efficiency. The paper analyzes modern metrics related to the processes of search engine optimization on the Internet.Unfortunately, these metrics evaluate only statistically after the fact of visiting a web resource or the budget of the Internet site in which theadvertising message about the product was placed. Therefore, based on the conversion metric, a criterion for the effectiveness of virtual promotion wasproposed in the work, which takes into account both the attractiveness of the semantic core and the attractiveness of the Internet site where thesemantic core will be located. The criterion reflects the income that we receive depending on the attractiveness of the semantic kernel and the Internetsite.

CLUDS: COMBINING LABELED AND UNLABELED DATA WITH LOGISTIC REGRESSION FOR SOCIAL MEDIA ANALYSIS

Automatic text classification and sentiment polarity detection are two important research problems of social media analysis. The meanings of the words are so important that they need to be captured by a document classification algorithm to reach an accurate classification performance. Another important issue with the text classification is the scarcity of labeled data. In this study, Combining Labeled and Unlabeled Data with Semantic Values of Terms (CLUDS) is presented. CLUDS has the following steps: preprocessing, instance labeling, combining labeled and unlabeled data, and prediction. In preprocessing step Latent Dirichlet Allocation (LDA) algorithm is used. In instance labeling step Logistic Regression is applied. In CLUDS, relevance values computation has been applied as a supervised term weighting methodology in the text classification field. Still, according to the literature, CLUDS is the first attempt that uses both relevance and weighting calculation in a semi-supervised semantic kernel for Support Vector Machines (SVM). In this study, Sprinkled-CLUDS and Adaptive-Sprinkled-CLUDS have also been implemented. Evaluated experimental results show that CLUDS, Sprinkled-CLUDS and Adaptive-Sprinkled-CLUDS generate a valuable performance gain over the baseline algorithms on test sets.

MODELLING SEMANTIC KERNEL OF WEB RESOURCE

The article presents an attempt to describe mathematically the effect of the semantic kernel of a web resource on the Internet. In accordance with the theory of marketing, the product that we want to sell on the network is characterized by the following basic properties: price, time and place. In other words, a potential buyer wants to receive a given product in the right place at a given time. To satisfy this need, it is necessary to use the classic component of marketing, product promotion. However, this component is now becoming a fully virtual instrument. This tool functions in a hypertext, video and image environment. Therefore, the user analyzes the meaning of these elements in order to get the desired product. The results of web projects carried out in this area indicate the emergence of a new phenomenon, which reflects the main meaning of virtual promotion – this is the semantic core. The core is a short annotation of the main properties of the product, its location and time of appearance. Therefore, the purpose of this article is both a presentation of a new object of research and a mathematical description. It is assumed that the semantic core is formed on the basis of natural language terms. In other words, the semantic core is a set of keywords that are grouped by meaning. We propose to use data mining approaches for clustering to group terms. The classic clustering method at the moment is k-means. The article presents a model of the semantic core based on this method. This method and its distance function are considered as the second stage of web content processing. At the first stage, web content is converted into a semantic web. However, the k-means technique has significant drawbacks when modeling the semantic core. Therefore, in the development of this idea, the work shows an alternative way to modeling the kernel. As an alternative approach, the construction of clusters based on the concept of maximum flow is considered. This approach has the significant advantage that the type of links in the semantic network overlaps with the type of distance function in this method. As a result, on a real web project, the effect of the connection between the semantic core model and the level of new users of the web resource was demonstrated over the past five years.&#x0D; Keywords: semantic kernel, keyword, k-means, max flow.

Bi-directional Long Short-Term Memory Model with Semantic Positional Attention for the Question Answering System

The intelligent question answering system aims to provide quick and concise feedback on the questions of users. Although the performance of phrase-level and numerous attention models have been improved, the sentence components and position information are not emphasized enough. This article combines Ci-Lin and word2vec to divide all of the words in the question-answer pairs into groups according to the semantics and select one kernel word in each group. The remaining words are common words and realize the semantic mapping mechanism between kernel words and common words. With this Chinese semantic mapping mechanism, the common words in all questions and answers are replaced by the semantic kernel words to realize the normalization of the semantic representation. Meanwhile, based on the bi-directional LSTM model, this article introduces a method of the combination of semantic role labeling and positional context, dividing the sentence into multiple semantic segments according to semantic logic. The weight is given to the neighboring words in the same semantic segment and propose semantic role labeling position attention based on the bi-directional LSTM model (BLSTM-SRLP). The good performance of the BLSTM-SRLP model has been demonstrated in comparative experiments on the food safety field dataset (FS-QA).

Ranking influencers of social networks by semantic kernels and sentiment information

Inspired by the importance of social media, a Social Network Opinion Leaders (SNOL) system has been proposed in this paper. The purpose of this system is to identify topic-based opinion leaders of social media. In order to accomplish this goal, several steps have been taken, such as data collection, data processing, data analysis, data classification, ranking of topic-based opinion leaders, and evaluation. The SNOL system has two main parts. In the first part, collected tweets are classified by semantic kernels for topic-based analysis. In the second part, leadership scores are given to each user in the network according to topic modeling and user modeling results. Leadership scores are then calculated with the formula generated and opinion leaders are determined for each category. Experiments are performed on data gathered from Twitter including 17,234,924 tweets from 38,727 users. The evaluation of opinion leader detection is a difficult job since there is no standard method for identifying opinion leaders. Therefore, the evaluation of the results of this study has been done using two different methods, retweet count and spread score, to prove that the suggested methodology outperforms the PageRank algorithm. The results have also been evaluated considering the user-topic sentiment correlation of the retrieved lists. Furthermore, SNOL has been compared against some opinion leader detection methods previously presented in the literature. The experimental results show that SNOL generates remarkably higher performance than the PageRank algorithm and other existing algorithms in the literature for nearly all topics and all selected top N opinion leaders.

Identifying Topic-based Opinion Leaders in Social Networks by Content and User Information

Social media is like a revolution since it has changed many things in people’s lifestyles by bringing new trends in communication, shopping, working…etc. In microblogging sites of the social media, more and more users meet every day, a wide range of ideas are quickly emerging, spreading, and creating interactive environments. In this context, social media can be seen as one of the most important sources of information affecting public opinion. Being inspired by the importance of social media, we propose Opinion Leader Detection (OLED) system in this paper. OLED has two main parts. A network was created by collecting users' information and their tweets shared within a specific time. In the first part, the tweets were labeled with their categories by semantic kernels for topic-based analysis. After the category information is obtained, the second part is attempt to detect whether the users are opinion leaders in their category. Then, the leadership scores were calculated with the formula generated and opinion leaders were determined in each category. We performed our experiments on a data collection gathered from Twitter that includes 17,234,924 tweets and 38,727 users. According to topic modeling and user modeling results, we give leadership scores to each user in the network. Users with highest scores are stated as opinion leaders. In order to evaluate OLED’s performance we also run PageRank algorithm on the same dataset. The experimental results show that our framework OLED generates remarkable performance in compare to PageRank algorithm nearly in all topics and all selected topN opinion leaders.

VIRTUAL PROMOTION KNOWLEDGE MANAGEMENT TECHNOLOGY

The article presents a new concept of “Management of knowledge about virtual promotion” on the Internet. Usually a real produ ct or service is being divided into four components (product, price, promotion and place) in accordance with the theory of marketing. One of the components is a product promotion. But now this element is becoming a fully virtual tool. It is necessary to consider product promotion as an image or a copy of a real product in a virtual space that lives in parallel on the network. Therefore, the objective of the paper is the presentation of a new object of research based on the experience of more than thirty real projects performed in Ukraine, USA, Europe and Canada. We regard the promotion as a software product, which works according to principles of knowledge management and machine learning. It is proposed that virtual promotion is characterized by four views: customer or user, data, technology and marketing. Thus, the structure of virtual promotion business process was presented. It includes four steps: selection of hypertext sources, knowledge representation and extraction, semantic kernel building and quality criterion evaluation to stop the process. Based on the process structure the research tasks were identified. The central task is semantic kernel forming. Then the software architecture was developed. IT solution contains CRM system as accounting tool and Web site as an image of virtual promotion. CRM plays main role as a commander center. Here we form semantic kernel and then send it via marketing channels such as Web site, telegram or viber accounts. Another part of IT solution is Web service such as Bing API or Google API. They help us to build the kernel. Also the paper demonstrates the list of future tasks that should be solved and the example of real project of proposed approach.

An improved random forest-based computational model for predicting novel miRNA-disease associations

BackgroundA large body of evidence shows that miRNA regulates the expression of its target genes at post-transcriptional level and the dysregulation of miRNA is related to many complex human diseases. Accurately discovering disease-related miRNAs is conductive to the exploring of the pathogenesis and treatment of diseases. However, because of the limitation of time-consuming and expensive experimental methods, predicting miRNA-disease associations by computational models has become a more economical and effective mean.ResultsInspired by the work of predecessors, we proposed an improved computational model based on random forest (RF) for identifying miRNA-disease associations (IRFMDA). First, the integrated similarity of diseases and the integrated similarity of miRNAs were calculated by combining the semantic similarity and Gaussian interaction profile kernel (GIPK) similarity of diseases, the functional similarity and GIPK similarity of miRNAs, respectively. Then, the integrated similarity of diseases and the integrated similarity of miRNAs were combined to represent each miRNA-disease relationship pair. Next, the miRNA-disease relationship pairs contained in the HMDD (v2.0) database were considered positive samples, and the randomly constructed miRNA-disease relationship pairs not included in HMDD (v2.0) were considered negative samples. Next, the feature selection based on the variable importance score of RF was performed to choose more useful features to represent samples to optimize the model’s ability of inferring miRNA-disease associations. Finally, a RF regression model was trained on reduced sample space to score the unknown miRNA-disease associations. The AUCs of IRFMDA under local leave-one-out cross-validation (LOOCV), global LOOCV and 5-fold cross-validation achieved 0.8728, 0.9398 and 0.9363, which were better than several excellent models for predicting miRNA-disease associations. Moreover, case studies on oesophageal cancer, lymphoma and lung cancer showed that 94 (oesophageal cancer), 98 (lymphoma) and 100 (lung cancer) of the top 100 disease-associated miRNAs predicted by IRFMDA were supported by the experimental data in the dbDEMC (v2.0) database.ConclusionsCross-validation and case studies demonstrated that IRFMDA is an excellent miRNA-disease association prediction model, and can provide guidance and help for experimental studies on the regulatory mechanism of miRNAs in complex human diseases in the future.

Analysis of Problem of Perceiving and Understanding Literary Texts by Philology Students

Introduction. The article discloses the importance of conducting psycholinguistic analysis of perceiving and understanding voluminous literary texts by philology students; the issue is relevant as there is a lack of research that concerns revealing the essential aspects of these processes, as well as the influence of conducting psycholinguistic analysis on effective performance in philology students’ professional activities is highlighted. The goal of the article is to conduct a theoretical analysis of the problem of perceiving and understanding literary texts by philology students and identify the level of development of future teachers of language and literature individual practical skills on decoding the content of a large prose fictional text.&#x0D; Methods. To achieve the goal, the authors used theoretical and empirical research methods. Empirical methods included methodology for determining semantic kernels of a text and its probabilistic connection, and methodology for determining the subtext, motives of the author and characters.&#x0D; Results. Theoretical analysis of the problem led the authors to conclusion that the processes of perceiving and understanding large literary text by philology students are based on two main components: cognitive skills and personal emotional feelings. The process of a literary text perceiving includes such structural elements as the reader’s orientation and the type of student’s work with a literary text, depends on students’ personal professional neoformations (reading competence, speech-cognitive competence), and is also influenced with philology student’s psychological type. The results of empirical study conducted by the authors indicate that future teachers of language have following developed reproductive skills with voluminous literary texts: they at a high level reproduce factual information of a text, can accurately convey an emotional load of a literary work, determine its theme, main idea, etc.&#x0D; Conclusions. The productive types of voluminous literary texts understanding associated with the definition of subtext, motives of the author and characters are mastered by philological students in a rather limited degree and this fact necessitates following studying underlying mechanisms of these processes functioning and developing tools for its targeted progress.

Word sense disambiguation using semantic kernels with class-based term values

Word sense disambiguation using semantic kernels with class-based term values

Neural embeddings: accurate and readable inferences based on semantic kernels

AbstractSentence embeddings are the suitable input vectors for the neural learning of a number of inferences about content and meaning. Similarity estimation, classification, emotional characterization of sentences as well as pragmatic tasks, such as question answering or dialogue, have largely demonstrated the effectiveness of vector embeddings to model semantics. Unfortunately, most of the above decisions are epistemologically opaque as for the limited interpretability of the acquired neural models based on the involved embeddings. We think that any effective approach to meaning representation should be at least epistemologically coherent. In this paper, we concentrate on thereadabilityof neural models, as a core property of any embedding technique consistent and effective in representing sentence meaning. In this perspective, this paper discusses a novel embedding technique (the Nyström methodology) that corresponds to the reconstruction of a sentence in a kernel space, inspired by rich semantic similarity metrics (a semantic kernel) rather than by a language model. In addition to being based on a kernel that captures grammatical and lexical semantic information, the proposed embedding can be used as the input vector of an effective neural learning architecture, calledKernel-based deep architectures(KDA). Finally, it also characterizesby designthe KDA explanatory capability, as the proposed embedding is derived from examples that are both human readable and labeled. This property is obtained by the integration of KDAs with an explanation methodology, calledlayer-wise relevance propagation (LRP), already proposed in image processing. The Nyström embeddings support here the automatic compilation of argumentations in favor or against a KDA inference, in form of an explanation: each decision can in fact be linked through LRP back to the real examples, that is, the landmarks linguistically related to the input instance. The KDA network output is explained via the analogy with the activated landmarks. Quantitative evaluation of the explanations shows that richer explanations based on semantic and syntagmatic structures characterize convincing arguments, as they effectively help the user in assessing whether or not to trust the machine decisions in different tasks, for example, Question Classification or Semantic Role Labeling. This confirms the epistemological benefit that Nyström embeddings may bring, as linguistically rich and meaningful representations for a variety of inference tasks.