University of Oulu

Understanding user behavior aspects on emergency mobile applications during emergency communications using NLP and text mining techniques

Saved in:
Author: Shaik, Fuzel Ahamed1
Organizations: 1University of Oulu, Faculty of Information Technology and Electrical Engineering, Computer Science
Format: ebook
Version: published version
Access: open
Online Access: PDF Full Text (PDF, 4.1 MB)
Pages: 65
Persistent link:
Language: English
Published: Oulu : F. Shaik, 2023
Publish Date: 2023-10-10
Thesis type: Master's thesis (tech)
Tutor: Oussalah, Mourad
Tamminen, Satu
Reviewer: Tamminen, Satu
Oussalah, Mourad


The use of mobile devices has been skyrocketing in our society. Users can access and share any type of information in a timely manner through these devices using different social media applications. This enabled users to increase their awareness of ongoing events such as election campaigns, sports updates, movie releases, disaster occurrences, and studies. The attractiveness, affordability, and two-way communication capabilities empowered these mobile devices that support various social media platforms to be central to emergency communication as well. This makes a mobile-based emergency application an attractive communication tool during emergencies. The emergence of mobile-based emergency communication has intrigued us to learn about the user behavior related to the usage of these applications. Our study was mainly conducted on emergency apps in Nordic countries such as Finland, Sweden, and Norway. To understand the user objects regarding the usage of emergency mobile applications we leveraged various Natural Language Processing and Text Mining techniques. VADER sentiment tool was used to predict and track users’ review polarity of a particular application over time. Lately, to identify factors that affect users’ sentiments, we employed topic modeling techniques such as the Latent Dirichlet Allocation (LDA) model. This model identifies various themes discussed in the user reviews and the result of each theme will be represented by the weighted sum of words in the corpus. Even though LDA succeeds in highlighting the user-related factors, it fails to identify the aspects of the user, and the topic definition from the LDA model is vague. Hence we leveraged Aspect Based Sentiment Analysis (ABSA) methods to extract the user aspects from the user reviews. To perform this task we consider fine-tuning DeBERTa (a variant of the BERT model). BERT is a Bidirectional Encoder Representation of transformer architecture which allows the model to learn the context in the text. Following this, we performed a sentence pair sentiment classification task using different variants of BERT. Later, we dwell on different sentiments to highlight the factors and the categories that impact user behavior most by leveraging the Empath categorization technique. Finally, we construct a word association by considering different Ontological vocabularies related to mobile applications and emergency response and management systems. The insights from the study can be used to identify the user aspect terms, predict the sentiment of the aspect term in the review provided, and find how the aspect term impacts the user perspective on the usage of mobile emergency applications.

see all

Copyright information: © Fuzel Ahamed Shaik, 2023. Except otherwise noted, the reuse of this document is authorised under a Creative Commons Attribution 4.0 International (CC-BY 4.0) licence ( This means that reuse is allowed provided appropriate credit is given and any changes are indicated. For any use or reproduction of elements that are not owned by the author(s), permission may need to be directly from the respective right holders.