Data Analytics, Data Mining, and Machine Learning for Social Media

Permanent URI for this collectionhttps://hdl.handle.net/10125/107454

Browse

Now showing 1 - 13 of 13

Systematic Contextual-based Affinity Analytics Research on Association of Manager Response and Customer Reviews
(2024-01-03) Babu, Xavier; Zhang, Juheng
We study the similarity between managers' responses and customer reviews and explore its influence on review convergence, customer ratings, and prices. While previous research has explored the influence of product reviews on price and reputation, little attention has been given to the effectiveness of managers' responses and their impact on product price and rating. This study fills this gap by examining managers' responses and their relationship with product review convergence/divergence. Additionally, we investigate whether managers exhibit similar responses to their peers and whether their responses are tailored to specific product review issues or broadly resemble their past responses. We develop a deep learning framework to understand semantic textual information in managers' responses and analyze the semantic affinity score with reviews. We investigate the dynamic relationships among managers' responses, product reviews, review convergence, product reputation and price with the Panel Vector Autoregression model with a travel dataset from TripAdvisor.com.
Effectively Delivering Author’s Point to Reader: Pointer-Generator Network Approach
(2024-01-03) Kim, Hyeonjo; Hong, Sukhwa
We propose a communication-based text summarization approach utilizing a pointer-generator network. The proposed summarization algorithm combines elements of abstractive summarization with keyword extraction. These keywords are associated with threats, causality, evidence, and solutions, all aimed at influencing readers' behavior. This paper is grounded in the theoretical frameworks of Language-Action Perspective (LAP) and Speech Act Theory (SAT), which are applied within the context of social and behavior change communication (SBCC). Consequently, our ultimate model considers the target readers to enhance the effectiveness of communication for promoting social and behavioral change. The SBCC-based summary incorporates more persuasive and emotive language while prioritizing SBCC-related content from the original article. This study demonstrates the potential effectiveness of a summary that encapsulates the author's intentions in shaping readers' thoughts and behaviors.
All is Fair in Love and War: Moral Foundations in English-Language Tweets during the First 36 Weeks of Conflict Between Ukraine and Russia
(2024-01-03) Amjadi, Eimon; John, Richard
We apply Moral Foundations Theory (MFT) to explore how the public conceptualizes the current conflict between Ukraine and the Russian Federation (Russia). Our analysis includes over 1.1 million English Tweets over the first 36 weeks related to the conflict. We used a LIWC (Luke) moral foundations dictionary to identify the moral components (care, fairness, loyalty, authority, and sanctity) of tweets from the U.S., pre- and post-Cold War NATO countries, Ukraine, and Russia. Following an initial spike at the beginning of the conflict, tweet volume declined and stabilized by week ten. The level of moral content varied significantly across the five regions and the five moral components. Tweets from different regions included significantly different moral foundations to conceptualize the conflict. Across all regions, tweets were dominated by loyalty content, while fairness content was infrequent. Moral content over time was relatively stable, and variations were linked to reported conflict events.
Inferences from Social Media Conversations about the Adoption of Chatbots
(2024-01-03) Pawlik, V. Phoebe; Pan, Yan
Chatbots, such as ChatGPT, are a rapidly emerging technology because of their potential applications in many fields. Existing research has addressed chatbot user adoption, mostly through experimental studies. Despite the increasing relevance of applying big data analytics to social media data to ascertain user impressions, research from this perspective on chatbot adoption is scarce. Therefore, this exploratory research investigates the topics of 44,310 conversations from the platform Reddit by applying deep learning topic modeling, sentiment analysis, and question-answering retrieval modeling combined with qualitative content analysis. This study (1) examines the topics associated with chatbots regarding the adoption process over the last seven years, (2) draws on the Unified Theory of Acceptance and Use of Technology 2 to refine the key chatbot user adoption factors, and (3) is an early contribution of applying deep learning textual analysis in this context.
Fake news detection by Machine Learning in Latin America: A Systematic Review
(2024-01-03) Nguema Ngomo, Jean Gabriel; Torres De Paiva, Raquel; Garcia, Ana Cristina
The growing spread of fake news on social media is a major scourge in society. To combat this problem, many studies have focused on different aspects to automatically detect fake news on social networks using artificial intelligence, especially Machine Learning. In the present work, to understand the current state of existing proposals, we conducted a systematic literature review. We propose to organize this literature in the light of a taxonomy of approaches.
From Virality to Veracity: Examining False Information on Telegram vs. Twitter
(2024-01-03) Rieskamp, Jonas; Mirbabaie, Milad; Langer, Marie; Kocur, Alexander
The COVID-19 pandemic gave rise to various false information including that Ivermectin is effective against COVID-19 disease, which spread on social media. Because Telegram's structure poses a high risk for radicalization, it is imperative to understand the underlying spreading processes. Therefore, we gathered a network of German-speaking channels that spread false information about Ivermectin to analyze the network structure and the spread of false information. By comparing results from Telegram to Twitter network, important insights are gained for research and practice. Results revealed that opinion leaders play a significant role in the spreading process of false information. This is evident because false information on Telegram can reach more users and requires fewer distributors compared to Twitter. The study outlines avenues for future research regarding false information on Telegram.
Visual Uniqueness: An Unsupervised Contrast Learning Approach
(2024-01-03) Feng, Xiaohang; Li, Charis; Zhang, Shunyuan
This paper develops an unsupervised machine learning model that scores a product image on its visual uniqueness. Based on large-scale images of Airbnb properties in New York City, our model used contrastive loss and random data augmentation to compute the visual uniqueness of a property image automatically. The model achieves 88.10% accuracy on a hold-out set. We identified key image features that make a room unique. Leveraging the advanced explainable AI techniques to generate interpretable uniqueness heatmaps, we found certain decorations (e.g., pillows, paintings) may help enhance room uniqueness. Next, we validated the model against human perceptions via two lab studies and an eye-tracking controlled experiment: both the model-predicted uniqueness and key image features are consistent with human judgment. We discussed discriminative validity between uniqueness and aesthetics. This research offers important managerial implications for individual hosts to optimize the visual presentation to stand out in the crowded market.
Enhancing User Behavior Modeling via Machine Learning with Combined Text and Image Data
(2024-01-03) Crowe, Chad; Ricks, Brian; Hall, Margeret
Existing works typically operate on either image or text data from social media, but rarely work with both content types simultaneously. We propose and validate a technique for combining image and text data for predicting user engagement metrics based on social media data. We collected image and text data from 366,415 Facebook posts and a respective 1,305,375 million comments. The combined model achieves a 3.5x improvement in mean squared error when predicting share count and a 14% improvement for comment sentiment over single data type models. Finally, the study demonstrates the ability to pick more performant advertisement out of 16.7 billion pairs; the resulting machine learning models successfully predicts for a greater comment sentiment, comment count, and share count 93%, 65%, and 63% of the time.
What if Social Bots Be My Friends? Estimating Causal Effect of Social Bots Using Counterfactual Graph Learning
(2024-01-03) Wu, Ziyue; Zhang, Yiqun; Chen, Xi
Social bots wield significant impact within social networks. Despite the widely recognized variations in individual responses to humans and bots, existing research has not thoroughly investigated the impact differences between human and social bots on individuals’ opinions. However, such differences are challenging to be estimated due to the presence of confounders introduced by homophily and the absence of counterfactual outcomes in observational network data. This study designs a counterfactual graph learning approach to accurately estimate causal effects, which exhibits superior performance in our simulations. The subsequent empirical results demonstrate that social bots yield a weaker influence than humans, and we further uncover diverse influential patterns of different types of opinions expressed by influence sources. Nevertheless, the impact difference is overestimated without applying our approach to control the confounders. Our research provides a practical approach and offers insights for stakeholders to scrutinize bots' impact from network perspectives.
A Framework for Perception Analysis of Social Media Data During Disease Outbreaks: Uncovering Patterns of Resentment Towards Bats
(2024-01-03) Okpala, Izunna; Romera Rodriguez, Guillermo; Han, Chaeeun; Meierhofer, Melissa; Mammola, Stefano; Halse, Shane; Kropczynski, Jess; Johnson, Joseph
Despite the growing number of natural language processing (NLP) tools developed for decision-makers to leverage social media for public perception evaluation during crises, a more robust framework is needed. This study explores a domain-specific machine learning framework for perception analysis using tweets about bats during disease outbreaks as a case study. Zoonotic disease outbreaks such as COVID-19 and Ebola are often attributed to bats and have resulted in unnecessary culling of wildlife; therefore, this is a case where perception is meaningful to a species. Analysis of 15,968 tweets showed a pattern in which tweets with anti-bat perceptions were most common during the early phases of an outbreak but declined over time while remaining negative, with 87.6% reliability of the framework according to manual coding of 300 randomly selected tweets. The framework can help stakeholders understand trends in public perception in near real-time and guide responses to spreading misinformation.
What Symptoms and How Long? An Interpretable AI Approach for Depression Detection in Social Media
(2024-01-03) Kuang, Junwei; Xie, Jiaheng; Yan, Zhijun
Depression is the most prevalent and serious mental illness, which induces grave financial and societal ramifications. Depression detection is key for early intervention to mitigate those consequences. Such a high-stake decision inherently necessitates interpretability. Although a few depression detection studies attempt to explain the decision, these explanations misalign with the clinical depression diagnosis criterion that is based on depressive symptoms. To fill this gap, we develop a novel Multi-Scale Temporal Prototype Network (MSTPNet). MSTPNet innovatively detects and interprets depressive symptoms as well as how long they last. Extensive empirical analyses show that MSTPNet outperforms state-of-the-art depression detection methods. This result also reveals new symptoms that are unnoted in the survey approach. We further conduct a user study to demonstrate its superiority over the benchmarks in interpretability. This study contributes to IS literature with a novel interpretable deep learning model for depression detection in social media.
How COVID-19 Conspiracy Theories Spread on Twitter
(2024-01-03) Gruzd, Anatoliy; Ghenai, Amira; Mai, Philip
Since the onset of the COVID-19 pandemic, conspiracy theories (CTs) related to the virus have been widely circulated on social media. The uncertainty surrounding the pandemic and available treatment options likely contributed to the wide dissemination of such theories on social media platforms like Twitter. This retrospective study examines the spread of CTs surrounding Bill Gates and COVID-19 vaccines on Twitter and identifies what accounts contributed to their dissemination. Based on the social network analysis of 100,601 Bill Gates and vaccine-related tweets shared by 71,364 users between March 1 and May 31, 2020, the study found that automated and suspended accounts had a significant impact on the spread of CTs around this topic. Their tweets were more likely to be reshared by others than by chance alone. This highlights the need for social media platforms to continue to act against harmful automated accounts, particularly considering recent trends to ease content moderation policies and debunking interventions by social media giants in the post-pandemic era.
Introduction to the Minitrack on Data Analytics, Data Mining, and Machine Learning for Social Media
(2024-01-03) Mentzer, Kevin; Gerhart, Natalie; Yates, David

Browse

Recent Submissions