Search

The growth of e-commerce has led to fraud practices emerging as one of the major risks to the industry. Fraudulent activities substantially harm the ranking systems of e-commerce platforms and have a negative impact on users' buying experiences. Today's consumers strive to learn about all the benefits and drawbacks of a product or service before making a purchase; therefore, online polling websites are crucial to increasing product sales. This paper focuses on detecting false reviews posted by users on e-commerce platforms by utilizing deep learning technique. To classify the fake reviews, a deep learning model called Bi-LSTM CNN is proposed and developed utilizing Doc2vec with term frequency-inverse document frequency (TF-IDF) as feature extraction method. The proposed model achieves an accuracy of 97.3% on the Ott dataset as compared to other deep learning based models. [ABSTRACT FROM AUTHOR]

Recently, the air pollution has been seriously regarded in the urban environment. Particularly, the substantial relationship between the air pollution and the daily movements of citizens has not been sufficiently investigated yet. This study attempts to empirically identify the patterns of air pollution using association rule mining from Seoul, the metropolitan city in South Korea. As a result, 214 patterns on air pollution are discovered, and those are embedded into vectors based on Doc2Vec technique. Then, this paper further examines how the movement of citizens reacts to the discovered patterns of air pollution by deploying the linear regression on the floating population with emphasis on the walk-traffic. Specifically, the walk-traffic is categorized into 14 categories by gender and age group, and the effects of air pollution patterns on each subgroup walk traffic were analyzed. Findings of this paper provide the empirical evidences on the estimated air pollution sensitivity by generation and gender to researchers and practitioners. This paper has the contribution on newly proposing the methodological framework for further managing the air pollution in the urban environment. [ABSTRACT FROM AUTHOR]

As a driving force behind innovation, technological fusion has emerged as a prevailing trend in knowledge innovation. However, current research lacks the semantic analysis and identification of knowledge fusion across technological domains. To bridge this gap, we propose a strategy that combines the latent Dirichlet allocation (LDA) topic model and the Doc2vec neural network semantic model to identify fusion topics across various technology domains. Then, we fuse the semantic information of patents to measure the characteristics of fusion topics in terms of knowledge diversity, homogeneity and cohesion. Applying this method to a case study in the fields of digital medicine and machine learning, we identify six fusion topics from two technology domains, revealing two distinct trends: diffusion from the center to the periphery and clustering from the periphery to the center. The study shows that the fusion measure of topic-semantic granularity can reveal the variability of technology fusion processes at a profound level. The proposed research method will benefit scholars in conducting multi-domain technology fusion research and gaining a deeper understanding of the knowledge fusion process across technology domains from a semantic perspective. [ABSTRACT FROM AUTHOR]

People usually prepare themselves by reading online reviews before purchasing a product. Sellers sometimes try to imitate user experience as a deceptive review to increase profits. Deceptive opinion spam detection has emerged as a challenging task in the field of opinion mining. Feature reduction techniques play the most important role in data mining which finds the essential features and removes the unnecessary dimensions that only contribute to the noise. This article extracts various textual features of gold-standard deceptive hotel reviews using different representation techniques like Part of Speech tag (POS tag), Bag of Word (BoW), and Doc2Vec. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are applied to reduce the features' dimensions. Various supervised classifiers like Decision Tree (DT), Na¨ıve Bayes (NB), Logistic Regression (LR), and Support Vector Machine (SVM) are used to classify deceptive opinions and truthful opinions. The features used by these supervised classifiers cannot retain sequential information from reviews. To overcome this problem, we used the Words Attention-based Bidirectional Long Short-Term Memory (WABiLSTM) network model that trains to learn the patterns of words. The article examines machine and deep learning-based spam detection models and provides their outline and results. The metrics like accuracy, precision, recall, and F-Measure are used to analyze the performance of these classification models. The experimental results showed the model's performance improved after reducing the features. [ABSTRACT FROM AUTHOR]

Public support for net zero is an important determinant of the solution for climate change. Newspapers can be used as a data source for observing public discourse on a variety of subjects including net zero. Public discourse produced through newspapers is not value-neutral; in particular, media inevitably has political orientation. This study examines the major topics of net zero articles, and if and how articles from different political orientations differ in the tone of news report of the topic of net zero. Latent Dirichlet allocation (LDA)-based topic modeling was applied to infer the major topics, and sentiment analysis was applied to understand the sentiments and tones of the articles. We found that 16 major topics were composed of 5 groups to discuss various aspects such as environment, socio-politics, and the economy. We also found that although the majority of articles are neutral, differences in the tone of South Korean media was particularly noticeable in negatively classified articles. These findings suggest that newspapers, particularly politically oriented newspapers, can influence public acceptance, support for net zero, and therefore the implementation of government policies in certain ways. [ABSTRACT FROM AUTHOR]

Trending topics are the most discussed topics at the moment on social media platforms, particularly on Twitter and Facebook. While the access to trending topics are free and available to everyone, marketing specialists and specific software are more expensive, therefore there are companies that do not have the budget to support those costs. The main goal of this work is to search for associations between trending topics and companies on social media platforms and HotRivers prototype was developed to fill this gap. This approach was applied to Twitter and used text mining techniques to process tweets, train personalized models of companies and deliver a list of the matched trending topics of the target company. So, in this work were tested different pre-processing text techniques and a method to select tweets called Centroid Strategy used on trending topics to avoid unwanted tweets. Also, were tested three models, an embedding vectors approach with Doc2Vec model, a probabilistic model with Latent Dirichlet Allocation, and a classification task approach with a Convolutional Neural Network used on the final architecture. The approach was validated with real cases like Adidas, Nike and Portsmouth Hospitals University. In the results stand out that trending topic Nike has an association with the company Nike and #WorldPatientSafetyDay has an association with Portsmouth Hospitals University. This prototype, HotRivers, can be a new marketing tool that points the direction to the next campaign. [ABSTRACT FROM AUTHOR]

In recent years, internet connectivity and the ubiquitous use of digital devices have afforded a landscape of expanding opportunity for the proliferation of scams involving attempts to deceive individuals into giving away money or personal information. The impacts of these schemes on victims have shown to encompass social, psychological, emotional and economic harms. Consequently, there is a strong rationale to enhance our understanding of scams in order to devise ways in which they can be disrupted. One way to do so is through crime scripting, an analytical approach which seeks to characterise processes underpinning crime events. In this paper, we explore how Natural Language Processing (NLP) methods might be applied to support crime script analyses, in particular to extract insights into crime event sequences from large quantities of unstructured textual data in a scalable and efficient manner. To illustrate this, we apply NLP methods to a public dataset of victims' stories of scams perpetrated in Singapore. We first explore approaches to automatically isolate scams with similar modus operandi using two distinct similarity measures. Subsequently, we use Term Frequency-Inverse Document Frequency (TF-IDF) to extract key terms in scam stories, which are then used to identify a temporal ordering of actions in ways that seek to characterise how a particular scam operates. Finally, by means of a case study, we demonstrate how the proposed methods are capable of leveraging the collective wisdom of multiple similar reports to identify a consensus in terms of likely crime event sequences, illustrating how NLP may in the future enable crime preventers to better harness unstructured free text data to better understand crime problems. [ABSTRACT FROM AUTHOR]

Collaborations between universities and firms provide a key pathway for innovation. In the big data era, however, the interactions between these two communities are being reshaped by information of much higher complexity and knowledge exchanges with more volume and pace. With this research, we put forward a methodology for comprehensively measuring both actor collaboration and produced knowledge in shaping network dynamics of university-industry collaboration. Using dual-layer networks consisting of organizations and topics, we mapped the longitudinal correlations between partnerships and knowledge in terms of both co-applications of patents and semantics. Network structures, individual characteristics, and knowledge proximity indicators were used to depict the longitudinal networks and then model the network dynamics. Further, a stochastic actor-oriented model was used to provide insights into the factors contributing to the network's evolution. A case study on university-industry collaborations in the information and communications technology sector demonstrates the feasibility of the methodology. The result of this study can be used for future research into the mechanisms that underpin university-industry collaborations and opportunity discovery. [ABSTRACT FROM AUTHOR]

With the rapid development of knowledge payment, customers are faced with a large number of knowledge products when purchasing, leading to the need for an effective recommendation system. However, existing recommendation systems cannot accurately and adequately represent paid knowledge products with implicit but specialized features and sparse interactive histories, and thus are deemed not suitable for such products. In this paper, we propose a novel recommendation system for knowledge products, the core of which is the designed customer-oriented representation of knowledge products. Specifically, we utilize customer activity information on the free knowledge sharing platform as the knowledge document for each customer of paid knowledge products, to extract customer knowledge background and preference. Then, a deep learning-based model Doc2vec is adopted to transfer knowledge documents to customer knowledge background vectors. Such vectors of a particular paid knowledge product are further aggregated to a product-level vector for customer-oriented product representation, based on which two recommendation results are generated with product ratings and similarities of paid knowledge products, respectively. Extensive comparative experiments are conducted to demonstrate the effectiveness of the proposed system for the representation and recommendation of paid knowledge products. This paper will contribute to the literature of knowledge payment and recommendation systems, as well as provide practical implications for the information service and the operation of knowledge products on knowledge payment platforms. [ABSTRACT FROM AUTHOR]

In a Common Law system, legal practitioners need frequent access to prior case documents that discuss relevant legal issues. Case documents are generally very lengthy, containing complex sentence structures, and reading them fully is a strenuous task even for legal practitioners. Having a concise overview of these documents can relieve legal practitioners from the task of reading the complete case statements. Legal catchphrases are (multi-word) phrases that provide a concise overview of the contents of a case document, and automated generation of catchphrases is a challenging problem in legal analytics. In this paper, we propose a novel supervised neural sequence tagging model for the extraction of catchphrases from legal case documents. Specifically, we show that incorporating document-specific information along with a sequence tagging model can enhance the performance of catchphrase extraction. We perform experiments over a set of Indian Supreme Court case documents, for which the gold-standard catchphrases (annotated by legal practitioners) are obtained from a popular legal information system. The performance of our proposed method is compared with that of several existing supervised and unsupervised methods, and our proposed method is empirically shown to be superior to all baselines. [ABSTRACT FROM AUTHOR]

As the amount of documents has exploded in the Internet era, many researchers have tried to understand the relationships between documents and predict the links between similar but unconnected documents. However, existing link prediction techniques that use the predefined links of documents might provide incorrect results, because of the generic problem of citation analysis. Moreover, they may fail to reflect important contents of documents in the link prediction process. Thus, we propose a new link prediction approach that employs the Doc2vec algorithm, a document-embedding method, in order to predict potential links between documents, by reflecting the functional context of technological words. For this, first, we collected both citation information and documents of patents of interest, and generated a patent network by using the citation relationship between patents. Second, we identified unconnected links between nodes and transformed the patent document into document vectors, based on the Doc2vec algorithm. In particular, since patent documents include useful functions for solving technological problems, the proposed approach extracts subject-action-object (SAO) structures that we used to generate document vectors. Then, we calculated the similarity between patents in the unconnected links of a patent network, and could predict potential links by using the similarity. Third, we validated the results of the proposed approach by comparing them using the Adamic–Adar technique, one of the traditional link prediction techniques, and word vector-based link prediction. We applied the Doc2vec-based link prediction approach to a real case, the unmanned aerial vehicle (UAV) technology field. We found that the proposed approach makes better predictions performance than the Adamic–Adar technique and the word vector approach. Our results can help analyzers accurately forecast future relationships between nodes in a network, and give R&D managers insightful information on the future direction of technological development by using a patent network. [ABSTRACT FROM AUTHOR]

With the continuous deepening of academic research in various disciplines and the continuous increase in the number of scientific researchers, exploring the mechanism of matching scientific research results and academic journal subjects is a key topic that can assist researchers in selecting suitable journals for submission. The classification and recommendation of academic journals based on a traditional text representation model cannot take advantage of the semantic relationship between words and cannot take into account the diversity of topics received by different journals, which affects the classification and recommendation effect. To solve these problems, this paper uses doc2vec to perform distributed representation of the bibliographic text so that the semantics between the text features are fully preserved. Then, the XGBoost algorithm is used to consider the impact of the different characteristics of the title, abstract, and keywords of the bibliography on the published journal. The academic journal submission recommendation model proposed in this paper can solve the problem that traditional methods cannot make full use of the contextual semantic information and improve the efficiency of scientific research personnel's academic achievement publications. Experiments on Common SCI English journals in the computer field show that when recommending three candidate journals, the accuracy rate reached 84.24%. [ABSTRACT FROM AUTHOR]

Document vectorization with an appropriate encoding scheme is an essential component in various document processing tasks, including text document classification, retrieval, or generation. Training a dedicated document in a specific domain may require large enough data and sufficient resource. This motivates us to propose a novel document representation scheme with two main components. First, we train TD2V, a generic pre-trained document embedding for English documents from more than one million tweets in Twitter. Second, we propose a domain adaptation process with adversarial training to adapt TD2V to different domains. To classify a document, we use the rank list of its similar documents using query expansion techniques, either Average Query Expansion or Discriminative Query Expansion. Experiments on datasets from different online sources show that by using TD2V only, our method can classify documents with better accuracy than existing methods. By applying adversarial adaptation process, we can further boost and achieve the accuracy on BBC, BBCSport, Amazon4, 20NewsGroup datasets. We also evaluate our method on a specific domain of sensitivity classification and achieve the accuracy of higher than 95 % even with a short text fragment having 1024 characters on 5 datasets: Snowden, Mormon, Dyncorp, TM, and Enron. [ABSTRACT FROM AUTHOR]

The special nature, volume and broadness of biomedical literature pose barriers for automated classification methods. On the other hand, manually indexing is time-consuming, costly and error prone. We argue that current word embedding algorithms can be efficiently used to support the task of biomedical text classification even in a multilabel setting, with many distinct labels. The ontology representation of Medical Subject Headings provides machine-readable labels and specifies the dimensionality of the problem space. Both deep- and shallow network approaches are implemented. Predictions are determined by the similarity between extracted features from contextualized representations of abstracts and headings. The addition of a separate classifier for transfer learning is also proposed and evaluated. Large datasets of biomedical citations are harvested for their metadata and used for training and testing. These automated approaches are still far from entirely substituting human experts, yet they can be useful as a mechanism for validation and recommendation. Dataset balancing, distributed processing and training parallelization in GPUs, all play an important part regarding the effectiveness and performance of proposed methods. [ABSTRACT FROM AUTHOR]

In the domain of legal information retrieval, an important challenge is to compute similarity between two legal documents. Precedents (statements from prior cases) play an important role in The Common Law system, where lawyers need to frequently refer to relevant prior cases. Measuring document similarity is one of the most crucial aspects of any document retrieval system which decides the speed, scalability and accuracy of the system. Text-based and network-based methods for computing similarity among case reports have already been proposed in prior works but not without a few pitfalls. Since legal citation networks are generally highly disconnected, network based metrics are not suited for them. Till date, only a few text-based and predominant embedding based methods have been employed, for instance, TF-IDF based approaches, Word2Vec (Mikolov et al. 2013) and Doc2Vec (Le and Mikolov 2014) based approaches. We investigate the performance of 56 different methodologies for computing textual similarity across court case statements when applied on a dataset of Indian Supreme Court Cases. Among the 56 different methods, thirty are adaptations of existing methods and twenty-six are our proposed methods. The methods studied include models such as BERT (Devlin et al. 2018) and Law2Vec (Ilias 2019). It is observed that the more traditional methods (such as the TF-IDF and LDA) that rely on a bag-of-words representation performs better than the more advanced context-aware methods (like BERT and Law2Vec) for computing document-level similarity. Finally we nominate, via empirical validation, five of our best performing methods as appropriate for measuring similarity between case reports. Among these five, two are adaptations of existing methods and the other three are our proposed methods. [ABSTRACT FROM AUTHOR]

As technology convergence is attracting increasing attention, many companies pay attention to human resources planning for the convergence R&D. This study proposes a patent-based R&D employee search framework for the firms planning to develop convergence technologies. We apply this framework to an automobile company searching patent inventors in the Information and Communication Technology (ICT) field, reflecting the future convergence trends in self-driving vehicles. ICT patent inventors were evaluated based on their ICT patents scored in three steps. First, the importance of the ICT areas they cover is considered in terms of International Patent Classification code by taking into account the converge potential with the focal firm's technologies in the future. Second, the feasibility of convergence with the ICT technologies that the focal firm wants to cover is assessed. Finally, various inherent values of ICT patent are considered. The proposed framework is expected to improve the efficiency of the process of spotting candidates for convergence R&D. [ABSTRACT FROM AUTHOR]

Science mapping using document networks comes often with the implicit assumption that scientific papers are indivisible units with unique links to neighbour documents. Research on proximity in co-citation analysis and the study of lexical properties of sections and citation contexts indicate that this assumption doesn't always hold. Moreover, the meaning of words and co-words depends on the context in which they appear. This study proposes the use of a neural network architecture for word and paragraph embeddings (Doc2Vec) for the measurement of similarity among those smaller units of analysis. It is shown that paragraphs in the "Introduction" and the "Discussion" Section are more similar to the abstract, that the similarity among paragraphs is related to -but not linearly- the distance between the paragraphs. The "Methodology" Section is least similar to the other sections. Abstracts of citing-cited documents are more similar than random pairs and the context in which a reference appears is most similar to the abstract of the cited document. This novel approach with higher granularity can be used for bibliometric aided retrieval and to assist in measuring interdisciplinarity through the application of network-based centrality measures. [ABSTRACT FROM AUTHOR]

In-domain adaptation (DA), the knowledge trained in one domain, is used to test an unknown domain. Existing approaches use limited efforts on DA in sentiment classification (SC) using neural networks. The challenging task here is the dissimilarity in the semantic behavior across domains. In this paper, convolutional neural networks (CNNs) learn the knowledge of a particular domain using Doc2Vec feature representation which provides good performance for DA in SC for the target domain. Our empirical analysis with one-layer CNN exhibits significant change in the accuracy by tuning the hyperparameters involved with the CNN. This paper derives into a suitable CNN architecture accompanying hyperparameters which favor DA between different domains. Our empirical analysis with multi-domain dataset demonstrates that with suitable hyperparameters, CNN works well for DASC problems. The comparative study shows that CNN with Doc2Vec model provides a strong capability of learning large data representation semantically with other state-of-the-art methods for the DASC. [ABSTRACT FROM AUTHOR]

This paper presents an act-utterance convergence between children and their guardians (an umbrella term for the adults who serve as children’s interaction partner in this paper). Act and utterance are two different carriers of intention. Some convergence patterns exist in act-utterance interaction between young children and their guardians. Guardians tend to respond to their children by describing what the children are doing, reflecting this convergence. This research has built the A-U Assimilation Model to assimilate act and utterance and narrow the gap of expression between act and utterance while keeping their meanings unchanged. The unified expression framework in this model makes the act and utterance comparable. The research analyzed the convergence between act and utterance from children’s preverbal period to the age of 60 months by adopting a large-scale corpus. And the analysis of the convergence between act and utterance is conducted by the Doc2Vec in a neural network. The experiment results show that the degree of guardian-led act-utterance convergence increases rapidly during the period that children begin to learn to speak and decrease gradually as children become more proficient language learners and users. [ABSTRACT FROM AUTHOR]

In this paper, we develop an automatic product classifier that can become a vital part of a natural user interface for an integrated online-to-offline (O2O) service platform. We devise a novel feature extraction technique to represent product descriptions that are expressed in full natural language sentences. We specifically adapt doc2vec algorithm that implements the document embedding technique. Doc2vec is a way to predict a vector of salient contexts that are specific to a document. Our classifier is trained to classify a product description based on the doc2vec-based feature that is augmented in various ways. We trained and tested our classifier with up to 53,000 real product descriptions from Groupon, a popular social commerce site that also offers O2O commerce features such as online ordering for in-store pick-up. Compared to the baseline approaches of using bag-of-words modeling and word-level embedding, our classifier showed significant improvement in terms of classification accuracy when our adapted doc2vec-based feature was used. [ABSTRACT FROM AUTHOR]

Recently, document embeddings methods have been proposed aiming at capturing hidden properties of the texts. These methods allow to represent documents in terms of fixed-length, continuous and dense feature vectors. In this paper, we propose to learn document vectors based on n-grams and not only on words. We use the recently proposed Paragraph Vector method. These n-grams include character n-grams, word n-grams and n-grams of POS tags (in all cases with n varying from 1 to 5). We considered the task of Cross-Topic Authorship Attribution and made experiments on The Guardian corpus. Experimental results show that our method outperforms word-based embeddings and character n-gram based linear models, which are among the most effective approaches for identifying the writing style of an author. [ABSTRACT FROM AUTHOR]

Analyze performance of unsupervised embedding algorithms in sentiment analysis of knowledge-rich data sets. We apply state-of-the-art embedding algorithms Word2Vec and Doc2Vec as the learning techniques. The algorithms build word and document embeddings in an unsupervised manner. To assess the algorithms' performance, we define sentiment metrics and use a semantic lexicon SentiWordNet (SWN) to establish the benchmark measures. Our empirical results are obtained on the Obesity data set from i2b2 clinical discharge summaries and the Reuters Science dataset. We use the Welch's test to analyze the obtained sentiment evaluation. On the Obesity data, the Welch's test found significant difference between the SWN evaluation of the most positive and most negative texts. On the same data, the Word2Vec results support the SWN results, whereas the Doc2Vec results partially correspond to the Word2Vec and the SWN results. On the Reuters data, the Welch's test did not find significant difference between the SWN evaluation of the most positive and most negative texts. On the same data, Word2Vec and Doc2Vec results only in part correspond to the SWN results. In unsupervised sentiment analysis of medical and scientific texts, the Word2Vec sentiment analysis has been more consistent with the SentiWordNet sentiment assessment than the Doc2Vec sentiment analysis. The Welch's test of the SentiWordNet results has been a strong indicator of future correspondence between Word2Vec and SentiWordNet results., Competing Interests: Conflict of interestOn behalf of all authors, the corresponding author states that there is no conflict of interest., (© The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd 2021.)