Dongsong Zhang’s research while affiliated with University of North Carolina at Charlotte and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (36)


KETCH: A Knowledge-Enhanced Transformer-Based Approach to Suicidal Ideation Detection from Social Media Content
  • Article

May 2024

·

62 Reads

·

5 Citations

Information Systems Research

Dongsong Zhang

·

Lina Zhou

·

·

[...]

·

Guodong (Gordon) Gao

Suicide is a major cause of death among 15- to 29-year-olds globally, claiming more than 50,000 lives in the United States in 2023 alone. Despite governmental efforts to provide support, many individuals experiencing suicidal thoughts do not seek help but are increasingly turning to social media to express their feelings. This trend offers a critical opportunity for timely detection and intervention of suicidal ideation. We develop an innovative transformer-based model for suicidal ideation detection (SID) that combines domain knowledge with dynamic embedding and lexicon-based enhancements. Our model, which is tested on social media data in two languages from different platforms, outperforms existing state-of-the-art models for SID. We have also explored its applicability to detecting depression and its practical implementation in real-world scenarios. Our research contributes significantly to the field, offering new methods for timely and proactive intervention in suicidal ideation, with potential wide-reaching effects on public health, economics, and society. Methodologically, our approach advances the integration of human expertise into AI models to enhance their effectiveness.



Collaborative group embedding and decision aggregation based on attentive influence of individual members: A group recommendation perspective

October 2022

·

19 Reads

·

17 Citations

Decision Support Systems

A key group decision making task is to aggregate individual preferences. Conventional group decision methods adopt pre-defined and fixed strategies to aggregate individuals' preferences, which can be ineffective due to the varying importance and influence of individual group members. Recent studies have proposed to assign different weights to individual members automatically based on the level of consistency of their ratings with group assessment outcomes. However, they ignored the high-order influence relationship among individual group members on group decision making. In this study, from a group recommendation perspective, we propose a novel collaborative Group Embedding and Decision Aggregation (GEDA) approach by leveraging the graph neural network technique to address those limitations. Specifically, GEDA first deploys a graph convolution operation on user-item interaction and group-item interaction graphs to generate embedding representations of members, groups, and items. A novel multi-attention (MA) module then learns each member's decision weight by simultaneously considering the relationships among members for aggregating individual preferences into group preferences. The empirical evaluation using two real-world datasets demonstrates the advantage of the proposed GEDA model over the state-of-the-art group recommendation models.


Transformer-based topic modeling (TM²)
Elbow method plot for determining an optimal number of topics
KLD scores of topic pairs
Correlation coefficients between topic loadings and emotional polarity
Log(KLD) distribution plot
Does Fake News in Different Languages Tell the Same Story? An Analysis of Multi-level Thematic and Emotional Characteristics of News about COVID-19
  • Article
  • Publisher preview available

September 2022

·

116 Reads

·

16 Citations

Fake news is being generated in different languages, yet existing studies are dominated by English news. The analysis of fake news content has focused on lexical and stylometric features, giving little attention to semantic features. A few studies involving semantic features have either used them as the inputs to classifiers with no interpretations, or treated them in isolation. This research aims to investigate both thematic and emotional characteristics of fake news at different levels and compare them between different languages for the first time. It extends a state-of-the-art topic modeling technique to extract news topics and introduces a divergence measure to assess the importance of thematic characteristics for identifying fake news. We further examine associations of the thematic and emotional characteristics of fake news. The empirical findings have implications for developing both general and language-specific countermeasures for fake news.

View access options


Do Fake News Between Different Languages Talk Alike? A Case Study of COVID-19 Related Fake News

January 2022

·

39 Reads

·

1 Citation

Communications in Computer and Information Science

Social media fuels fake news’ spread across the world. English news has dominated existing fake news research, and how fake news in different languages compares remains severely under studied. To address this scarcity of literature, this research examines the content and linguistic behaviors of fake news in relation to COVID-19. The comparisons reveal both differences and similarities between English and Spanish fake news. The findings have implications for global collaboration in combating fake news.KeywordsFake newsLanguageTopics modelingContent-based behavior linguistic behavior


DNCP: An Attention-based Deep Learning Approach Enhanced with Attractiveness and Timeliness of News for Online News Click Prediction

January 2021

·

57 Reads

·

20 Citations

Information & Management

Predicting news clicks or popularity is of great importance to news providers and recommender systems. Attractiveness and timeliness of news are two prominent drivers of news clicks. Attractiveness represents the appealingness or interestingness of news to an individual, while timeliness indicates the recency of news. Existing research on news click prediction models has ignored those two important variables. There is a lack of exploration and understanding of how to represent them in a predictive model and how effective and valuable they are to the prediction performance of a model. To fill these gaps, in this research, we propose a deep news click prediction (DNCP) model that integrates attractiveness and timeliness of news in an attention-based deep neural network for predicting news clicks. We also propose new measures for those two variables. Empirical evaluation using two real-world datasets shows that the DNCP model outperforms a variety of baseline models. The findings of this research provide several novel research contributions and practical implications for improving news click prediction.


Vision Screening and Self-Testing by Mobile-Based Automated Visual Acuity Assessments Apps: A Systematic Review with Meta-Analysis (Preprint)

December 2020

·

25 Reads

BACKGROUND Vision impairments (VI) and blindness are one of the core global public health issues. Visual acuity (VA) is one of the most crucial standard psychophysical test of visual function, and used widely in a broad range of healthcare domain, especially in many clinical settings. OBJECTIVE To assess the accuracy and application of using Mobile devices-based visual acuity measurement apps. METHODS We searched Pubmed, Embase, Cochrane Library, and Google Scholar for relevant articles published between January 1, 2008 and July 1 2020. Two reviewers independently selected studies that assessed the mobile-based VA measurement apps. We included all studies that assessed a tablet and/or smartphone VA measurement apps. RESULTS Most of the enrolled 22 studies considered as high quality studies, evaluating by QUADAS-2. In meta-analysis, six studies involving 24284 participants were included. In 3~5 years old group, the pooled sensitivity was 0.87 (95% CI 0.79, 0.93); the pooled specificity was 0.78 (95% CI 0.70, 0.85); In 6~22 years old group, the pooled sensitivity was 0.86 (95% CI 0.84, 0.87); the pooled specificity was 0.91 (95% CI 0.90, 0.91). In ≥55 years old group, the pooled sensitivity was 0.85 (95% CI 0.55, 0.98); the pooled specificity was 0.98 (95% CI 0.95, 0.99). CONCLUSIONS In this study, we conducted a comprehensive review of the state-of-the-art research to investigate the diagnostic value and limitations of existing mobile-based VA test applications. Evidence from this study shows that mobile-based app VA measurements may be useful tools for VI detection.


The Use of Mobile Apps for Visual Acuity Assessment: A Systematic Review with A Meta-Analysis (Preprint)

December 2020

·

93 Reads

·

17 Citations

JMIR mhealth and uhealth

Background Vision impairments (VIs) and blindness are major global public health issues. A visual acuity (VA) test is one of the most crucial standard psychophysical tests of visual function and has been widely used in a broad range of health care domains, especially in many clinical settings. In recent years, there has been increasing research on mobile app–based VA assessment designed to allow people to test their VA at any time and any location. Objective The goal of the review was to assess the accuracy and reliability of using mobile VA measurement apps. Methods We searched PubMed, Embase, Cochrane Library, and Google Scholar for relevant articles on mobile apps for VA assessment published between January 1, 2008, and July 1, 2020. Two researchers independently inspected and selected relevant studies. Eventually, we included 22 studies that assessed tablet or smartphone apps for VA measurement. We then analyzed sensitivity, specificity, and accuracy in the 6 papers we found through a meta-analysis. ResultsMost of the 22 selected studies can be considered of high quality based on the Quality Assessment of Diagnostic Accuracy Studies–2. In a meta-analysis of 6 studies involving 24,284 participants, we categorized the studies based on the age groups of the study participants (ie, aged 3-5 years, aged 6-22 years, and aged 55 years and older), examiner (ie, professional and nonprofessional examiners), and the type of mobile devices (ie, smartphone, iPad). In the group aged 3 to 5 years, the pooled sensitivity for VA app tests versus clinical VA tests was 0.87 (95% CI 0.79-0.93; P=.39), and the pooled specificity was 0.78 (95% CI 0.70-0.85; P=.37). In the group aged 6 to 22 years, the pooled sensitivity for VA app tests versus clinical VA tests was 0.86 (95% CI 0.84-0.87; P


From Networking to Mitigation: The Role of Social Media and Analytics in Combating the COVID-19 Pandemic

October 2020

·

61 Reads

·

27 Citations

Information Systems Management

Due to quarantine and social distancing, people are attached to and rely on social media more than ever before. The main goal of this paper is to highlight several important areas of research on studying Covid-19 through the lens of social media for information system researchers and call for more future research. The paper will not only discuss their significance and urgency, but also shed light on existing work and potential future research issues and challenges.


Citations (28)


... This approach highlights the significance of collaborative innovation ecosystems in managing complex decision-making processes, complementing energy storage decision models like those in Kim and Powell (2011). Extending this line of research, Wang et al. (2019) examine the implications of cultural differences and user-driven product improvements in e-commerce platforms, bridging the gap between technical and behavioral optimization in product-service systems. ...

Reference:

Optimal Commitment Strategy of the Wind-battery-hydrogen Hybrid System in the Spot Market
DISCOVERING CULTURAL DIFFERENCES IN ONLINE CONSUMER PRODUCT REVIEWS
  • Citing Article
  • January 2019

Journal of Electronic Commerce Research

... In social networks, users who share similar tastes and preferences tend to form groups [17]. Alves et al. [24] noted that the correlation between users' personalities and preferences can be leveraged by group recommender systems to match users with similar interests. This approach helps to minimize group heterogeneity and preference conflicts, thereby alleviating the cold start problem. ...

Collaborative group embedding and decision aggregation based on attentive influence of individual members: A group recommendation perspective
  • Citing Article
  • October 2022

Decision Support Systems

... In response to these challenges, recent Fake News Detection (FND) models have started incorporating semantic features alongside traditional lexical and stylistic ones [19,36]. Notably, the overall emotion, negativity, and expressions of anger are higher in fake news compared to real news [36]. ...

Does Fake News in Different Languages Tell the Same Story? An Analysis of Multi-level Thematic and Emotional Characteristics of News about COVID-19

... A 2022 meta-analysis [14] of the performance of mobile apps for VA assessment reported that when such apps were used by non-professionals, the accuracy was better than for professionals, a finding that was attributed to adults such as parents or schoolteachers having a better understanding of children's responses, behaviour and moods than eye care professionals who are not known to the children being examined. The age of participants may also impact on the results obtained, with the sensitivity and diagnostic odds ratios of mobile VA apps being significantly better when adults are examined rather than young children. ...

Use of Mobile Apps for Visual Acuity Assessment: Systematic Review and Meta-analysis

... Given that 83.72% of the global population owns a smartphone [7], this represents a significant opportunity to use the smartphone as a platform for improving access to medical care by allowing earlier detection and treatment of diseases. However, few ophthalmology applications have shown significant success in detecting ophthalmological conditions [8]. This limits their utility to ophthalmologists as a screening tool to identify ophthalmological diseases and monitor the progression of these diseases. ...

The Use of Mobile Apps for Visual Acuity Assessment: A Systematic Review with A Meta-Analysis (Preprint)

JMIR mhealth and uhealth

... By harnessing the learning prowess inherent in neural networks, these models estimate future behaviors grounded in users' historical activities and interactional patterns within social media domains [49], [63], [78]. This includes predictions encompassing potential purchasing inclinations [13], [62], click preferences [82], or trends pertinent to information dissemination [2], [91]. Based on social cognitive theory, in low power distance environments, spiritual leadership has an amplifying effect on the positive impact of proactive customer service performance. ...

DNCP: An Attention-based Deep Learning Approach Enhanced with Attractiveness and Timeliness of News for Online News Click Prediction
  • Citing Article
  • January 2021

Information & Management

... We, therefore, decided to collect data from Twitter due to three important reasons. First, at a generic level, social media analytics have proved to be instrumental during the global pandemic in varying applications such as government crisis management (Chon & Kim, 2022), mitigating the negative effects of COVID-19 (Zhang et al., 2020), understanding the emotions of stakeholders impacted by the pandemic (Tinguely et al., 2020). Second, Twitter has shown its ability to highlight supply chain issues (Schmidt et al., 2020). ...

From Networking to Mitigation: The Role of Social Media and Analytics in Combating the COVID-19 Pandemic
  • Citing Article
  • October 2020

Information Systems Management

... The literature indicates that using such systems reduces the administrative burden on healthcare professionals, minimizes errors, and ensures all team members can access accurate and up-to-date patient information. This seamless flow of information is crucial for coordinated care and effective communication [93]. ...

What makes clinical documents helpful and engaging? An empirical investigation of experience sharing in an online medical community
  • Citing Article
  • September 2020

International Journal of Medical Informatics

... Support and encouragement can make community members feel more connected to the group, and make them more likely to remain devoted over time [22]. When it comes to OIC, knowledge and relation are two of the most common exchangeable products, whereas relation has become a rare commodity [41]. In an OIC, users can be followed by others. ...

A deeper investigation of different types of core users and their contributions for sustainable innovation in a company-hosted online co-creation community

Journal of Cleaner Production