Visual Sentiment Analysis from Disaster Images in Social Media

Hassan, Syed Zohaib; Ahmad, Kashif; Hicks, Steven A.; Halvorsen, Pål; Al‐Fuqaha, Ala; Conci, Nicola; Riegler, Michael

doi:10.3390/s22103628

Cited by 26 publications

(5 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For disaster-related image classification, there have been studies where transfer-learning-based models have been used either as feature extractors or for fine-tuning the model. Such studies include flood detection from social media multimodal content [43], disaster-related tasks in a multitask learning [44], real-time system for disaster image classification during hurricane [45], sentiment analysis from disaster images [46], aerial image classification for disaster response [47], and deep features with multimodal training [48].…”

Section: Transfer Learning For Image Classificationmentioning

confidence: 99%

Robust Training of Social Media Image Classification Models

Alam

Ofli

et al. 2024

IEEE Trans. Comput. Soc. Syst.

View full text Add to dashboard Cite

Images shared on social media help crisis managers gain situational awareness and assess incurred damages, among other response tasks. As the volume and velocity of such content are typically high, real-time image classification has become an urgent need for faster disaster response. Recent advances in computer vision and deep neural networks have enabled the development of models for image classification for a number of tasks, including detecting crisis incidents, filtering irrelevant images, classifying images into specific humanitarian categories, and assessing the severity of the damage. To develop robust models, it is necessary to understand the capability of the publicly available pretrained models for these tasks, which remains to be underexplored in the crisis informatics literature. In this study, we address such limitations by investigating ten different network architectures for four different tasks using the largest publicly available datasets for these tasks. We also explore various data augmentation strategies, semisupervised techniques, and a multitask learning setup. In our extensive experiments, we achieve promising results.

show abstract

Section: Transfer Learning For Image Classificationmentioning

confidence: 99%

Robust Training of Social Media Image Classification Models

Alam

Ofli

et al. 2024

IEEE Trans. Comput. Soc. Syst.

View full text Add to dashboard Cite

show abstract

“…Hassan et al [25] presented a deep visual sentiment analyzer for disaster-related images, employing CNN and transfer learning techniques. While their work could potentially aid responders in analyzing the emotions of affected individuals based on image content, a significant limitation arises from the lack of consideration for image authenticity.…”

Section: 1sm Disaster Image Analysismentioning

confidence: 99%

Credibility assessment of social media images shared during disasters

2024

IJATEE

View full text Add to dashboard Cite

As per the Global Assessment Report (GAR 2022) published by the "United Nations Office for Disaster Risk Reduction" [1], the frequency of disasters has increased significantly over the years and is projected to surge by 40 percent by the year 2030. This increased prevalence of disasters has also led to a rise in both human fatalities and economic ramifications across the globe. Addressing such ecological disruptions necessitates the acquisition of humancentric information for disaster response organizations [2]. In recent years, social media (SM) has emerged as a pertinent alternative information conduit alongside traditional media during disaster events such as floods, earthquakes, hurricanes, etc. [3].Throughout the past decade, various SM platforms, particularly Twitter, have played a significant role in humanitarian response tasks due to their widespread utilization in disseminating information and gathering valuable insights [4]. *Author for correspondence Recent studies have highlighted the significant role of SM images in the aftermath of disasters. They serve to convey accurate details regarding the extent of devastation [5], aid humanitarian groups in assessing infrastructural harm [6], and expedite the identification and rescue of missing or injured persons [7].However, despite the potential benefits of integrating SM as an additional resource within established disaster response frameworks, humanitarian authorities raise concerns about the veracity of the data disseminated on such platforms. The credibility of SM-driven information remains a significant issue for both the general public and disaster response agencies [4]. SM platforms have become significant contributors to the propagation of fake news [8,9] and numerous instances have occurred where forged images have gone viral on SM platforms, especially Twitter, during disasters, creating panic among the population. For instance, during Hurricane Sandy in 2012, Twitter played a crucial role in keeping people

show abstract

“…They focus on content analysis of images from Instagram (Y. Kim, Song, & Lee, 2020) and Tik-Tok videos (Zeng & Abidin, 2021), thematic framing analysis of Instagram posts (E. Lee & Weder, 2021), critical discourse analysis of memes from Twitter and Instagram (Boling, 2020), sentiment and theme analysis of the videos from TikTok (Rutherford et al, 2022), sentiment analysis of the images from Flickr and Twitter (Hassan et al, 2022), and visual cross-platform analysis to investigate platform vernaculars of multiple social media platforms (Pearce et al, 2020). Each social media is built with different algorithms: the use of hashtags may differ according to the platform as shown by Bossetta (2018).…”

Section: Hashtag-basedmentioning

confidence: 99%

Strategies and challenges for constructing and collecting visual corpora from image-based social media platforms

Samofalova

2024

SComS

View full text Add to dashboard Cite

Visual elements play an important role within the multimodal nature of social media (Pearce et al., 2020). A growing body of research has focused on the analysis of still and moving images from different social media platforms from various perspectives of communication and media studies (Hautea, Parks, Takahashi, & Zeng, 2021; Li & Xie, 2020; Veum & Undrum, 2018). Although the aforementioned studies describe visual data collection, their principal focus does not rely on this collection, but on data analysis. Little attention has been paid to the challenges of collecting visual datasets (Highfield & Leaver, 2016). In this paper, I propose a methodological overview of several strategies for collecting large corpora of visual data from image-based social media platforms. Provided with exemplary publications, I review five strategies for collecting visual corpora: hashtag-based, account-based, metadata-based, random sampling, and mixed approach. Lastly, I present a case study with my own mixed approach to the collection of visual data from Instagram. Considering the usage, advantages and limitations of each strategy, the article will contribute to the developing science of social media research. I believe that a literature analysis of visual data collection strategies and a provided case study can help researchers optimize visual data collection from image-based social media.

show abstract

Visual Sentiment Analysis from Disaster Images in Social Media

Cited by 26 publications

References 44 publications

Robust Training of Social Media Image Classification Models

Robust Training of Social Media Image Classification Models

Credibility assessment of social media images shared during disasters

Strategies and challenges for constructing and collecting visual corpora from image-based social media platforms

Contact Info

Product

Resources

About