Search

Whether it is an NLP (natural language processing) task or an NLU (natural language understanding) task, many methods are model oriented, ignoring the importance of data features. Such models did not perform well for many tasks based on feature loose, unbalanced tricky data including text classification tasks. In this regard, this paper proposes a classification method called LSTM-SN (long-short term memory RNN fusion social network) based on extremely complex datasets. The approach condenses the characteristics of the dataset. LSTM combines with social network methods derived from specific datasets to complete the classification task, and then use complex network structure evolution methods to discover dynamic social attributes. The experimental results show that this method can overcome the shortcomings of traditional methods and achieve better classification results. Finally, a method to calculate the accuracy of fusion model is proposed. The research ideas of this paper have far-reaching significance in the domain of social data analysis and relation extraction. [ABSTRACT FROM AUTHOR]

Negative social media usage during the COVID-19 pandemic has highlighted the importance of understanding the spread of misinformation and toxicity in public online discussions. In this paper, we propose a novel unsupervised method to discover the structure of online COVID-19-related conversations. Our method trains a nine-state Hidden Markov Model (HMM) initialized from a biclustering of 23 features extracted from online messages. We apply our method to 16,000 conversations (1.5 million messages) that took place on the Facebook pages of 15 Canadian newspapers following COVID-19 news items, and show that it can effectively extract the conversation structure and discover the main themes of the messages. Furthermore, we demonstrate how the PageRank algorithm and the conversation graph discovered can be used to simulate the impact of five different moderation strategies, which makes it possible to easily develop and test new strategies to limit the spread of harmful messages. Although our work in this paper focuses on the COVID-19 pandemic, the methodology is general enough to be applied to handle communications during future pandemics and other crises, or to develop better practices for online community moderation in general. [ABSTRACT FROM AUTHOR]

Computing directed Minimum Spanning Tree ( D M S T ) is a fundamental problem in graph theory. It is applied in a wide spectrum of fields from computer network and communication protocol design to revenue maximization in social networks and syntactic parsing in Natural Language Processing. State-of-the-art solutions are online algorithms that compute  D M S T  for a given graph and a root. For multi-query requirements, the online algorithm is inefficient. To overcome the drawbacks, in this paper, we propose an indexed approach that reuses the computation result to facilitate single and batch queries. We store all the potential edges of  D M S T  in a hierarchical tree in  O (n)  space complexity. Furthermore, we answer the  D M S T  query of any root in  O (n)  time complexity. Experimental results demonstrate that our approach can achieve a speedup of 2–3 orders of magnitude in query processing compared to the state-of-the-art while consuming O(n) index size. [ABSTRACT FROM AUTHOR]

Nowadays, group decision making is an everyday occurrence in different scenarios, such as marketing or social networks. These social networks have facilitated communication between experts because they do not need to meet in person. However, communication between experts through the internet generates three problems: the management of large amounts of information, the fact that experts often provide their information using natural language, and the lack of analysis of experts' intentions. In this paper, we propose a novel large scale group decision making method to manage the information generated by a large number of experts, using the natural language processing approach, specifically sentiment analysis. This approach makes it possible to detect the degree of positivity and aggressiveness of each expert and thus proceed to a classification. Once the behaviours are detected, the experts are grouped according to them and, for each group, a weight and a unique preference relation is obtained. In addition, we propose an optimised consensus analysis process, in which it is not necessary to compare all experts with each other, but only groups of experts. • We created a classifier to detect aggressiveness and positivity. • Experts are clustered according to their aggressiveness and positivity. • The expert groups are used to create an LGDM system. • The LSGDM system determines a ranking of alternatives agreed upon by the experts. [ABSTRACT FROM AUTHOR]

In recent years, the use of nontraditional data sources in statistical production has been increasing, given the additional need for more timely and disaggregated data. In the scope of nontraditional sources, citizen science represents an innovative approach to filling data gaps and including citizens as part of the recent innovation processes of national statistical offices (NSOs) in both the production of statistics and its role as data stewards of the national statistical systems (NSSs). The National Statistical Office of Colombia (DANE, acronym in Spanish) has structured a project within the framework of the Data4Now initiative for Sustainable Development Goal (SDG) indicators from SDG 16, using social networks as an alternative source of data generated by citizens, and natural language processing (NLP) to extract actionable intelligence from this data. In this paper, we describe the proposed strategy followed by DANE to estimate SDG indicators 16.b.1 and 16.7.2 as a pilot exercise in the framework of experimental statistics. Preliminary results illustrate potential use cases of unconventional information streams to analyze social phenomena through virtual environments such as online social media, and raise compelling challenges regarding representativity and quality assurance based on the statistical standards used by NSOs. [ABSTRACT FROM AUTHOR]

In today's world, mental health diseases have become highly prevalent, and depression is one of the mental health problems that has become widespread. According to WHO reports, depression is the second-leading cause of the global burden of diseases. In the proliferation of such issues, social media has proven to be a great platform for people to express themselves. Thus, a user's social media can speak a great deal about his/her emotional state and mental health. Considering the high pervasiveness of the disease, this paper presents a novel framework for depression detection from textual data, employing Natural Language Processing and deep learning techniques. For this purpose, a dataset consisting of tweets was created, which were then manually annotated by the domain experts to capture the implicit and explicit depression context. Two variations of the dataset were created, on having binary and one ternary labels, respectively. Ultimately, a deep-learning-based hybrid Sequence, Semantic, Context Learning (SSCL) classification framework with a self-attention mechanism is proposed that utilizes GloVe (pre-trained word embeddings) for feature extraction; LSTM and CNN were used to capture the sequence and semantics of tweets; finally, the GRUs and self-attention mechanism were used, which focus on contextual and implicit information in the tweets. The framework outperformed the existing techniques in detecting the explicit and implicit context, with an accuracy of 97.4 for binary labeled data and 82.9 for ternary labeled data. We further tested our proposed SSCL framework on unseen data (random tweets), for which an F1-score of 94.4 was achieved. Furthermore, in order to showcase the strengths of the proposed framework, we validated it on the "News Headline Data set" for sarcasm detection, considering a dataset from a different domain. It also outmatched the performance of existing techniques in cross-domain validation. [ABSTRACT FROM AUTHOR]

Sentiment analysis is a natural language processing task that has received increased attention in the last decade due to the vast amount of opinionated data on social media platforms such as Twitter. Although the methodologies employed have grown in number and sophistication, analysing irony and sarcasm still poses a severe problem. From the linguistic perspective, sarcasm has been studied in discourse analysis from several perspectives, but little attention has been given to specific metrics that measure its relevance. In this paper we describe the creation of a manually-annotated dataset where detailed text markers are included. This dataset is a sample from a larger corpus of tweets (n= 76,764) on two highly controversial films: Cats and Star Wars: The Rise of Skywalker. We took two different samples for each film, one before and one after their release, to compare reception and presence of sarcasm. We then used a sentiment analysis tool to measure the impact of sarcasm in polarity detection and then manually classified the mechanisms of sarcasm generation. The resulting corpus will be useful for machine learning approaches to sarcasm detection as well as discourse analysis studies on irony and sarcasm. [ABSTRACT FROM AUTHOR]

Influencer marketing through social networks is becoming an important alternative to traditional ways of advertising. Various solutions have been proposed that often take advantage of graph-based approaches to discover influencers in social networks. This paper designs a new method for the discovery of influential users in Instagram, by focusing on user-generated posts as an alternative source of information, to potentially augment the existing solutions based on network topology or connections. The text associated with each Instagram post potentially consists of a set of hashtags and a descriptive caption. Various word embedding methods such as Co-occurrence and fastText are examined to represent captions and hashtags. These representations are combined within a support vector machines framework to distinguish influential posts from non-influential ones. Extensive experiments show that the text data can play a significant role in identifying influential posts, and further demonstrate the strength of the proposed method for discovering influencers on Instagram. [ABSTRACT FROM AUTHOR]

With the rapid development of the internet, social media has become an essential tool for getting information, and attracted a large number of people join the social media platforms because of its low cost, accessibility and amazing content. It greatly enriches our life. However, its rapid development and widespread also have provided an excellent convenience for the range of fake news, people are constantly exposed to fake news and suffer from it all the time. Fake news usually uses hyperbole to catch people’s eyes with dishonest intention. More importantly, it often misleads the reader and causes people to have wrong perceptions of society. It has the potential for negative impacts on society and individuals. Therefore, it is significative research on detecting fake news. In the paper, we built a model named SMHA-CNN (Self Multi-Head Attention-based Convolutional Neural Networks) that can judge the authenticity of news with high accuracy based only on content by using convolutional neural networks and self multi-head attention mechanism. In order to prove its validity, we conducted experiments on a public dataset and achieved a precision rate of 95.5% with a recall rate of 95.6% under the 5-fold cross-validation. Our experimental result indicates that the model is more effective at detecting fake news. [ABSTRACT FROM AUTHOR]

Big web data from sources including online news and Twitter are good resources for investigating deep learning. However, collected news articles and tweets almost certainly contain data unnecessary for learning, and this disturbs accurate learning. This paper explores the performance of word2vec Convolutional Neural Networks (CNNs) to classify news articles and tweets into related and unrelated ones. Using two word embedding algorithms of word2vec, Continuous Bag-of-Word (CBOW) and Skip-gram, we constructed CNN with the CBOW model and CNN with the Skip-gram model. We measured the classification accuracy of CNN with CBOW, CNN with Skip-gram, and CNN without word2vec models for real news articles and tweets. The experimental results indicated that word2vec significantly improved the accuracy of the classification model. The accuracy of the CBOW model was higher and more stable when compared to that of the Skip-gram model. The CBOW model exhibited better performance on news articles, and the Skip-gram model exhibited better performance on tweets. Specifically, CNN with word2vec models was more effective on news articles when compared to that on tweets because news articles are typically more uniform when compared to tweets. [ABSTRACT FROM AUTHOR]

With the rapid growth of Internet penetration, more and more people choose the Internet to express their views on topics of interest. In recent years, named entity recognition (NER) is becoming a popular task for the public to obtain structured information from public opinion text. At present, NER models with good results, such as deep learning model, need a lot of labeled data for training. However, this will give rise to a problem: labeling a large amount of data requires a lot of human resources, which is thankless in some areas. Therefore, in this paper, we propose a NER model combining active learning and deep learning methods. Firstly, the active learning method can solve the above problem. The strategy combines uncertainty-based sampling and diversity-based sampling to estimate the information of data. We use highly informative data as the initial training dataset. Secondly, this paper uses a deep learning model combining bidirectional encoder representations from Transformers, bidirectional long–short-term memory and conditional random field (BERT-BiLSTM-CRF). BERT extracts the semantic features of data, and BiLSTM predicts the probability distribution of entity labels. We use the CRF for decoding the probability distribution into corresponding entity labels. Finally, we use the initial training dataset for training BERT-BiLSTM-CRF. This model predicts the entity labels of the unlabeled data. Then, we judge if the machine-labeled data is highly reliable and expand the highly reliable data to the initial training dataset. The updated dataset retrains the NER model, so that the trained model has higher precision than the previous model. The results show that our model performs well without a large number of labeled datasets. The model achieves a precision value of 70.31%, recall rate of 74.93% and F1 score of 72.55% in the named entity recognition task, which proves the effectiveness of our model. Besides, the F1 score of BERT-BiLSTM-CRF with uncertainty-based sampling and diversity-based sampling (UD_BBC) is higher than the BiLSTM-CRF based on maximum normalized log-probability (MNLP_BiLSTM-CRF) by 9.00%, when recognizing overall entity categories. It provides a solution to the problem of named entity recognition in educational public opinion. [ABSTRACT FROM AUTHOR]

The Moral Foundations Dictionary (MFD) is a useful tool for applying the conceptual framework developed in Moral Foundations Theory and quantifying the moral meanings implicated in the linguistic information people convey. However, the applicability of the MFD is limited because it is available only in English. Translated versions of the MFD are therefore needed to study morality across various cultures, including non-Western cultures. The contribution of this paper is two-fold. We developed the first Japanese version of the MFD (referred to as the J-MFD) using a semi-automated method—this serves as a reference when translating the MFD into other languages. We next tested the validity of the J-MFD by analyzing open-ended written texts about the situations that Japanese participants thought followed and violated the five moral foundations. We found that the J-MFD correctly categorized the Japanese participants’ descriptions into the corresponding moral foundations, and that the Moral Foundations Questionnaire (MFQ) scores correlated with the frequency of situations, of total words, and of J-MFD words in the participants’ descriptions for the Harm and Fairness foundations. The J-MFD can be used to study morality unique to the Japanese and also multicultural comparisons in moral behavior. [ABSTRACT FROM AUTHOR]

Geographical location words (GLWs) are words associated with geographical locations. GLWs are the significant foundation for text data processing and social network location inference. In this paper, we propose a framework for constructing high-precision general GLWs datasets, and a Chinese GLWs dataset (named GeoCN) is constructed. To some extent, GeoCN solves the problem of lacking a Chinese GLWs lexicon with diverse categories, high accuracy, and robust versatility. GeoCN consists of three parts: a) points of interest (POI) data collected based on the electronic map API, b) administrative division data constructed based on the national information platform, and c) GLWs data expanded and filtered by automated procedures and manual processing. We establish a GLWs glossary for each administrative region and map each GLW to its location. GeoCN covers 34 provincial-level administrative regions, 392 prefecture-level administrative regions, and 3,160 county-level administrative regions in China. The number of GLWs in GeoCN reaches 1,763,476, and the compressed file size is 117 MB. • A framework for constructing high-precision general GLWs datasets is proposed. • A Chinese geographical location word dataset (called GeoCN) is constructed. • GeoCN has more categories of location words, higher precision, and stronger universality. • GeoCN can support natural language processing and social network location inference. [ABSTRACT FROM AUTHOR]

On September 2013 an attack on the Westgate mall in Kenya led to a four day siege, resulting in 67 fatalities and 175 wounded. During the crisis, Twitter became a crucial channel of communication between the government, emergency responders and the public, facilitating the emergency management of the event. The objectives of this paper are to present the main activities, use patterns and lessons learned from the use of the social media in the crisis. Using TwitterMate, a system developed to collect, store and analyze tweets, the main hashtags generated by the crowd and specific Twitter accounts of individuals, emergency responders and NGOs, were followed throughout the four day siege. A total of 67,849 tweets were collected and analyzed. Four main categories of hashtags were identified: geographical locations, terror attack, social support and organizations. The abundance of Twitter accounts providing official information made it difficult to synchronize and follow the flow of information. Many organizations posted simultaneously, by their manager and by the organization itself. Creating situational awareness was facilitated by information tweeted by the public. Threat assessment was updated through the information posted on social media. Security breaches led to the relay of sensitive data. At times, misinformation was only corrected after two days. Social media offer an accessible, widely available means for a bi-directional flow of information between the public and the authorities. In the crisis, all emergency responders used and leveraged social media networks for communicating both with the public and among themselves. A standard operating procedure should be developed to enable multiple responders to monitor, synchronize and integrate their social media feeds during emergencies. This will lead to better utilization and optimization of social media resources during crises, providing clear guidelines for communications and a hierarchy for dispersing information to the public and among responding organizations. [ABSTRACT FROM AUTHOR]

Traffic accident detection is an important strategy governments can use to implement policies intended to reduce accidents. They usually use techniques such as image processing, RFID devices, among others. Social network mining has emerged as a low-cost alternative. However, social networks come with several challenges such as informal language and misspellings. This paper proposes a method to extract traffic accident data from Twitter in Spanish. The method consists of four phases. The first phase establishes the data collection mechanisms. The second consists of vectorially representing the messages and classifying them as accidents or non-accidents. The third phase uses named entity recognition techniques to detect the location. In the fourth phase, locations pass through a geocoder that returns their geographic coordinates. This method was applied to Bogota city and the data on Twitter were compared with the official traffic information source; comparisons showed some influence of Twitter on the commercial and industrial area of the city. The results reveal how effective the information on accidents reported on Twitter can be. It should therefore be considered as a source of information that may complement existing detection methods. [ABSTRACT FROM AUTHOR]

The epidemic spread of fake news is a side effect of the expansion of social networks to circulate news, in contrast to traditional mass media such as newspapers, magazines, radio, and television. Human inefficiency to distinguish between true and false facts exposes fake news as a threat to logical truth, democracy, journalism, and credibility in government institutions. In this paper, we survey methods for preprocessing data in natural language, vectorization, dimensionality reduction, machine learning, and quality assessment of information retrieval. We also contextualize the identification of fake news, and we discuss research initiatives and opportunities. [ABSTRACT FROM AUTHOR]