Search

A visual presentation has attracted researchers' interests as an alternative for textual description in presenting choice profiles in discrete choice experiments. The purpose of this study is to compare two different visual presentation formats, namely, the one showing an integrated landscape picture that includes several attributes in it and the other presenting several images each representing a single attribute separately in a tabular format, in addition to textual information. We employed a hypothetical urban park landscapes in Tokyo as a case study, adopted a split sample approach and utilized a generalized multinomial logit model for parameter estimations. Consequently, we found that the visual presentation encouraged more systematic choices compared to the textual information and the landscape image enabled more homogeneous choice randomness within the group in relation to the separated attribute images. The landscape picture also elicited more confident choices of respondents and generated narrower confidence intervals of willingness-to-pay estimates for the park attributes. These results suggest that the landscape image was superior to the other two formats from the researcher's perspectives. [ABSTRACT FROM AUTHOR]

Knowledge graphs are essential tools for representing real-world facts and finding wide applications in various domains. However, the process of constructing knowledge graphs often introduces noises and errors, which can negatively impact the performance of downstream applications. Current methods for knowledge graph error detection primarily focus on graph structure and overlook the importance of textual information in error detection. Therefore, this paper proposes a novel error detection framework that combines both structural and textual information. The framework utilizes a confidence module for error detection while generating knowledge embeddings. The performance of this approach outperforms baseline methods in error detection and link prediction experiments, particularly achieving state-of-the-art performance in the error detection task.

Knowledge graph completion (KGC) utilizes known knowledge graph triples to infer and predict missing knowledge, making it one of the research hotspots in the field of knowledge graphs. There are still limitations in generating high-quality entity embeddings and fully understanding the contextual information of entities and relationships. To overcome these challenges, this paper introduces a novel pre-trained language model-based method for knowledge graph completion that significantly enhances the quality of entity embeddings by integrating entity categorical information with textual descriptions. Additionally, this method employs an innovative multi-layer residual attention network in combination with PLMs, deepening the understanding of the joint contextual information of entities and relationships. Experimental results on the FB15k-237 and WN18RR datasets demonstrate that our proposed model significantly outperforms existing baseline models in link prediction tasks. [ABSTRACT FROM AUTHOR]

This paper introduces a structural classification scheme for folk music using a multimodal fusion of musical features and textual information. A data corpus of Thiruvathirakali music (a popular folk dance in India) is compiled for this purpose. The musical features are learned using a gated recurrent unit (GRU) with an attention mechanism from musical texture features. A term frequency-inverse document frequency (TF-IDF) model is employed to derive text-based features from transcription data. Early integration of the GRU-derived features and TF-IDF features is employed for multimodal fusion. The performance is evaluated using Thiruvathirakali music corpus. The performance of the proposed system is also compared with that of a support vector machine-based classifier. The proposed system reports an overall precision, recall and F1 measure of 0.81 each for the GRU-attention-based model for multi-modal fusion. The scope of the proposed work for structural segmentation is also demonstrated with the help of an example. [ABSTRACT FROM AUTHOR]

Knowledge graph completion (KGC) utilizes known knowledge graph triples to infer and predict missing knowledge, making it one of the research hotspots in the field of knowledge graphs. There are still limitations in generating high-quality entity embeddings and fully understanding the contextual information of entities and relationships. To overcome these challenges, this paper introduces a novel pre-trained language model-based method for knowledge graph completion that significantly enhances the quality of entity embeddings by integrating entity categorical information with textual descriptions. Additionally, this method employs an innovative multi-layer residual attention network in combination with PLMs, deepening the understanding of the joint contextual information of entities and relationships. Experimental results on the FB15k-237 and WN18RR datasets demonstrate that our proposed model significantly outperforms existing baseline models in link prediction tasks.

As location information of numerous Internet of Thing (IoT) devices can be recognized through IoT sensor technology, the need for technology to efficiently analyze spatial data is increasing. One of the famous algorithms for classifying dense data into one cluster is Density-Based Spatial Clustering of Applications with Noise (DBSCAN). Existing DBSCAN research focuses on efficiently finding clusters in numeric data or categorical data. In this paper, we propose the novel problem of discovering a set of adjacent clusters among the cluster results derived for each keyword in the keyword-based DBSCAN algorithm. The existing DBSCAN algorithm has a problem in that it is necessary to calculate the number of all cases in order to find adjacent clusters among clusters derived as a result of the algorithm. To solve this problem, we developed the Genetic algorithm-based Keyword Matching DBSCAN (GKM-DBSCAN) algorithm to which the genetic algorithm was applied to discover the set of adjacent clusters among the cluster results derived for each keyword. In order to improve the performance of GKM-DBSCAN, we improved the general genetic algorithm by performing a genetic operation in groups. We conducted extensive experiments on both real and synthetic datasets to show the effectiveness of GKM-DBSCAN than the brute-force method. The experimental results show that GKM-DBSCAN outperforms the brute-force method by up to 21 times. GKM-DBSCAN with the index number binarization (INB) is 1.8 times faster than GKM-DBSCAN with the cluster number binarization (CNB). [ABSTRACT FROM AUTHOR]

Entity alignment is an effective means of matching entities from various knowledge graphs (KGs) that represent the equivalent real-world object. With the development of representation learning, recent entity alignment methods learn entity structure representation by embedding KGs into a low-dimensional vector space, and then entity alignment relies on the distance between entity vectors. In addition to the graph structures, relations and attributes are also critical to entity alignment. However, most existing approaches ignore the helpful features included in relations and attributes. Therefore, this paper presents a new solution RAEA (Relation Awareness and Attribute Involvement for Entity Alignment), which includes relation and attribute features. Relation representation is incorporated into entity representation by Dual-Primal Graph CNN (DPGCNN), which alternates convolution-like operations on the original graph and its dual graph. Structure representation and attribute representation are learned by graph convolutional networks (GCNs). To further enrich the entity embedding, we integrate the textual information of the entity into the entity graph embedding. Moreover, we fine-tune the entity similarity matrix by integrating fine-grained features. Experimental results on three benchmark datasets from real-world KGs show that our approach has superior performance to other representative entity alignment approaches in most cases. [ABSTRACT FROM AUTHOR]

We clarify the information value of changes in the textual information of analyst reports. The analyses reveal that stock prices react considerably to changes in the linguistic tone of reports. In addition, prices react asymmetrically to the levels and changes of report tone, overreact to the levels, and underreact to changes. Our evidence suggests that a change in the textual tone provides incremental information that investors overlook, supporting the informational value of changes in textual information. [ABSTRACT FROM AUTHOR]

Because the nature of numerical information is intuitive and comprehensible, it has been widely used to form a basis for decision making, yet numerical information based on historical principle does not reflect messages about future corporate performance. To confront this issue, one may consider textual information that can transmit future corporate potential without any hysteresis. The key point is how to digest an extensive amount of textual information and identify those topics most likely to precede changes in operation status. Topic modeling can categorize these textual disclosures based on their underlying content and help examine which topics have a strong relevance to corporate operations. To extract decisive words from textual information, we set up a statistical-based approach with objectivity as opposed to frequently used heuristics (i.e., dictionary-based approaches with human involvement). Joint utilization of topic modelling and a statistical-based approach can compress an excessive amount of textual information into a manageable size in a timely manner and further realize a discrepancy among various topics in terms of relevance and influence on corporate operations. Our results benefit managers and current and future investors in how to structure regulatory filings and how word choices are decisive to them in their decision judgments. [ABSTRACT FROM AUTHOR]

A new recommendation model is proposed to solve the problems of data sparsity and deep semantic information learning when using score information as auxiliary recommendation. Feature learning of scoring information is achieved through depth auto-encoder to encode and decode implicit feedback scoring matrix. And the features of type text information are learned through the Flatten Layer and Full-connection Layer operate, with user-film type matrix being the input to the Embedded Layer. Meanwhile, the model uses BERT+BiLSTM structure to realize features learning and features splicing of contextual information on film title text. After the fusion of the three features, the prediction score is obtained by auto-encoder processing. Movielens 1 M and Movielens 100 k are used as datasets, mean absolute error and mean square error are used as evaluation indexes, and SVD, PMF, PMMMF, SCC, RMbDn and Hern are used as comparison models. The results show that the MAE values of the model are reduced to 0.045 8 and 0.046 0 respectively, and the MSE values of the model are reduced to 0.027 3 and 0.039 0 respectively, which are better than the results of the comparison algorithms. The new recommendation model achieves performance improvement.

Since the volume of corporate information disclosures is constantly increasing, this study seeks to distinguish information with a "text" nature from information with a "numbers" nature to examine the existence and the direction of the volume of these two types of information on the efficiency of the stock price formation in companies. This study also deals with the effect that the volume of these two types of disclosed information have on each other in the process of price formation. However, it does not concentrate on the quality of the content of accounting information and its usefulness. The research sample consists of 169 companies listed on the Tehran Stock Exchange in 1397 (Iranian calendar). Research hypotheses were tested cross-sectionally using ordinary least squares regression. The results revealed that the amount of disclosed information, whether in the form of "text" or "numbers," in general, does not significantly affect the efficiency of the process of the companies' stock price formation in the Iranian capital market. Also, the volume of these two types of information cannot strengthen the influence of each other on the efficiency of the stock price formation process. In addition to expanding the accounting literature in the areas of "disclosure" and "market efficiency," the results of this study contain significant achievements for policymakers, analysts, shareholders, and company managers regarding the volume of information disclosure. The users' ability and the disclosure quality are also some other factors that should be considered when deciding whether to increase the volume of disclosure.

We develop insights into how analysts' social ties affect their objectivity and the information content of reports, based on quantitative measures of social ties and textual information content. Using a sample ranges from January 2006 to December 2016 in China, we show that with wide social ties to fund managers, analysts' ability to burnish their reputation with informative reports and to achieve career success with optimistic stock ratings is enhanced. Further evidence indicates that the trade-off between reputation and career success is dynamic. Specifically, after being elected as New Fortune stars, analysts attach more importance on textual information and are less subjective on stock ratings. These findings are robust across various robustness tests. Overall, our evidence highlights the role of social ties in the information-dissemination process in China A-share stock markets from the perspective of analysts and fund managers. [ABSTRACT FROM AUTHOR]

This paper examines the impact of the tone used in the annual reports of listed companies on the probability of regulatory inquiries from 2014 to 2019. The empirical results verify that the more positive tone of the annual report, the more it can exert a psychological framing effect, which makes the regulator less likely to issue an annual report inquiry letter. Further analysis shows that this impact that the tone of annual reports has on the probability of a regulatory inquiry is significantly weakened when the company has irregularities or has been issued a modified audit opinion by the auditor. Overall, the results provide the basis for the government to regulate the information disclosure of the annual reports. [ABSTRACT FROM AUTHOR]

Purpose: Information and information sources can be divided into three broad categories according to their nature or type: textual information (book, journal article, conference paper, dissertation, newspaper, etc.), visual information (infographic, photo, Cartoons, films, etc.) and audiovisual information. The purpose of this study is to determine the effect of reading textual information in comparison with visual information (infographics) on the health information perception of high school girl students in Tehran. Method: The current study was carried out by applying a semi-experimental method in which a pretest/posttest design was used on two groups. The research population included 30 high-school girl students of Boshra High School in Tehran. Two groups of 15 students were randomly selected and invited to participate in groups. In fact, 15 participants in one group read textual health information and 15 participants in another group read infographic health information. An acquaintance pre-test was performed on both groups and two post-tests of learning and retention were used to evaluate whether the participants learnt and retained textual and visual health information, which was presented to them during the period under study using a 27-item self-report questionnaire based on textual and visual information. Finally, the results of the two groups were analyzed by descriptive statistics and dependent and independent t-test. Findings: The findings of the study showed that the mean scores of the two groups of participants were the same as in the pre-test, the difference between the observed mean scores for the pre-test scores was not statistically significant. Furthermore, the mean scores of the two groups of participants for both textual information and visual information (infographics) were equal in the post-test of learning, and the difference in mean scores for post-test scores was not statistically significant. This means that both textual information and visual information (infographics) have increased the participants' learning to the same degree and do not differ significantly in terms of learning. Additionally, the mean scores of both groups for textual information and visual information (infographics) in the post-test were also equal and the difference in observed mean for post-test scores was not statistically significant. The latter finding means that textual information and visual (infographic) information were remembered equally and were not significantly different. Originality/value: The findings of the present study indicates that textual information and visual information (infographics) have increased the learning and retention of participants to the same extent and are not different in terms of learning and retention. It can also be concluded that despite the appearance of the formats and the media, the text still has an important place in the learning and retention of students.

Abstract The analytics on graph-structured data in cyber spaces has advanced many human-centered computing technologies. However, if only utilizing the structural properties, we might be prohibited from unraveling unknown social relations of nodes especially in the structureless networked systems. Up-to-date ways to unfold latent relationships from graph-structured data are network representation learning (NRL) techniques, but it is difficult for most existing ones to deal with the network-structureless situations due to the fact that they largely depend on the observed connections. With the ever-broader spectrum of human-centered networked systems, large quantities of textual information have been generated and collected from social and physical spaces, which may provide the clues of hidden social relations. In order to discover latent social relations from the accompanied text resources, this paper attempts to bridge the gap between text data and graph-structured data so that the textual information can be encoded to substitute for those incomplete structural information. Generative adversarial networks (GANs) are employed in the cross-modal framework to make the transformed data indistinguishable in graph-domain space and also capable of depicting structure-aware relationships with network homophily. Experiments conducted on three text-based network benchmarks demonstrate that our approach can reveal more realistic social relations from text-domain information compared against the state-of-the-art baselines.

A new recommendation model is proposed to solve the problems of data sparsity and deep semantic information learning when using score information as auxiliary recommendation. Feature learning of scoring information is achieved through depth auto-encoder to encode and decode implicit feedback scoring matrix. And the features of type text information are learned through the Flatten Layer and Full-connection Layer operate, with user-film type matrix being the input to the Embedded Layer. Meanwhile, the model uses BERT+BiLSTM structure to realize features learning and features splicing of contextual information on film title text. After the fusion of the three features, the prediction score is obtained by auto-encoder processing. Movielens 1 M and Movielens 100 k are used as datasets, mean absolute error and mean square error are used as evaluation indexes, and SVD, PMF, PMMMF, SCC, RMbDn and Hern are used as comparison models. The results show that the MAE values of the model are reduced to 0.045 8 and 0.046 0 respectively, and the MSE values of the model are reduced to 0.027 3 and 0.039 0 respectively, which are better than the results of the comparison algorithms. The new recommendation model achieves performance improvement. [ABSTRACT FROM AUTHOR]

Techniques that map the entities and relations of the knowledge graph (KG) into a low-dimensional continuous space are called KG embedding or knowledge representation learning. However, most existing techniques learn the embeddings based on the facts in KG alone, suffering from the issues of imperfection and spareness of KG. Recently, the research on textual information in KG embedding has attracted much attention due to the rich semantic information supplied by the texts. Thus, in this paper, a survey of techniques for textual information based KG embedding is proposed. Firstly, we introduce the techniques for encoding the textual information to represent the entities and relations from perspectives of encoding models and scoring functions, respectively. Secondly, methods for incorporating the textual information in the existing embedding techniques are summarized. Thirdly, we discuss the training procedure of textual information based KG embedding techniques. Finally, applications of KG embedding with textual information in the specific tasks such as KG completion in zero-shot scenario, multilingual entity alignment, relation extraction and recommender system are explored. We hope that this survey will give insights to researchers into textual information based KG embedding.

We deal with a heterogeneous pharmaceutical knowledge-graph containing textual information built from several databases. The knowledge graph is a heterogeneous graph that includes a wide variety of concepts and attributes, some of which are provided in the form of textual pieces of information which have not been targeted in the conventional graph completion tasks. To investigate the utility of textual information for knowledge graph completion, we generate embeddings from textual descriptions given to heterogeneous items, such as drugs and proteins, while learning knowledge graph embeddings. We evaluate the obtained graph embeddings on the link prediction task for knowledge graph completion, which can be used for drug discovery and repurposing. We also compare the results with existing methods and discuss the utility of the textual information.

Collaborative filtering (CF) is a common method used by many recommender systems. Traditional CF algorithms exploit users' ratings as the sole information source to learn user preferences. However, ratings usually sparse cause a serious impact on the recommendation results. Most existing CF algorithms use ratings and textual information to alleviate the sparsity of data and then utilize matrix factorization to achieve the latent feature interactions for rating prediction. Nevertheless, the following shortcomings remain in these studies: (1) The word orders and surrounding words of the textual information are ignored. (2) The nonlinearity of feature interactions is seldom exploited. Therefore, we propose a novel hybrid neural network to combine textual information and rating (NCTR) information for item recommendation. The proposed NCTR model is built upon a hybrid neural network framework with fine-grained modeling of latent representation and nonlinearity feature interactions for rating prediction. Specifically, convolution neural network is applied to extract effectively contextual features from textual information. Meanwhile, a fusion layer is exploited to combine features, and the multilayer perceptions are used to model the nonlinear interactions between the merged item latent features and user latent features. Experimental results over five real-world datasets show that NCTR significantly outperforms several state-of-the-art recommendation methods. Source codes are available in https://github.com/luojia527/NCTR_master. [ABSTRACT FROM AUTHOR]

Much of medical knowledge is stored in the biomedical literature, collected in archives like PubMed Central that continue to grow rapidly. A significant part of this knowledge is contained in images with limited metadata available which makes it difficult to explore the visual knowledge in the biomedical literature. Thus, extraction of metadata from visual content is important. One important piece of metadata is the type of the image, which could be one of the various medical imaging modalities such as X‐ray, computed tomography or magnetic resonance images and also of general graphs that are frequent in the literature. This study explores a late, score‐based fusion of several deep convolutional neural networks with a traditional hand‐crafted bag of visual words classifier to classify images from the biomedical literature into image types or modalities. It achieved a classification accuracy of 85.51% on the ImageCLEF 2013 modality classification task, which is better than the best visual methods in the challenge that the data were produced for, and similar compared to mixed methods that make use of both visual and textual information. It achieved similarly good results of 84.23 and 87.04% classification accuracy before and after augmentation, respectively, on the related ImageCLEF 2016 subfigure classification task. [ABSTRACT FROM AUTHOR]