Using Social Media Data to Monitor Natural Disaster: A Multi Dimension Convolutional Neural Network Approach with Word Embedding

Abstract

Social media has a significant role in natural disaster management, namely as an early warning and monitoring when natural disasters occur. Artificial intelligence can maximize the use of natural disaster social media messages for natural disaster management. The artificial intelligence system will classify social media message texts into three categories: eyewitness, non-eyewitness and don't-know. Messages with the eyewitness category are essential because they can provide the time and location of natural disasters. A common problem in text classification research is that feature extraction techniques ignore word meanings, omit word order information and produce high-dimensional data. In this study, a feature extraction technique can maintain word order information and meaning by using three-word embedding techniques, namely word2vec, fastText, and Glove. The result is data with 1D, 2D, and 3D dimensions. This study also proposes a data formation technique with new features by combining data from all word embedding techniques. The classification model is made using three Convolutional Neural Network (CNN) techniques, namely 1D CNN, 2D CNN and 3D CNN. The best accuracy results in this study were in the case of earthquakes 78.33%, forest fires 81.97%, and floods 78.33%. The calculation of the average accuracy shows that the 2D and 3D v1 data formation techniques work better than other techniques. Other results show that the proposed technique produces better average accuracy.

Identification of Social Media Posts Containing Self-reported COVID-19 Symptoms using Triple Word Embeddings and Long Short-Term Memory

“…This stage transforms the raw tweet data into more structured data. This study's stages are cleaning (Faisal et al, 2022), converting uppercase letters to lowercase (HaCohen-Kerner, Miller, & Yigal, 2020), and tokenization. Data cleaning involves removing unwanted characters and specific…”

“…Following this, the maximum length, mean, mode, and median values are derived from the statistical analysis of the word count. Consequently, a process known as word padding is implemented in order to ensure the uniformity of the word count across all messages within the given text data set (Faisal et al, 2022).…”

“…Using three-word embedding models, various data structures are formed with varying dimensions. The integration of structured data derived from the output of feature extraction techniques compensates for the limitations of a particular technique through the implementation of alternative techniques (Faisal et al, 2022). All models underwent training with the parameters as specified in the list in Table 3.…”

LSTM and Bi-LSTM Models For Identifying Natural Disasters Reports From Social Media

“…A value of 0 serves to discard or forget a feature, while a value of 1 signifies that the feature should be stored within the network. The equations for the input gate are represented by equation ( 1), the forget gate by equation ( 2), and the output gate by equation (3).…”

“…In some cases of natural disasters, the presence of social media plays a crucial role in assisting every activity within the disaster management cycle. During the pre-disaster stage, social media can serve as an early warning before a disaster occurs [3]. Therefore, numerous studies have been conducted utilizing social media data as a social network sensor, as demonstrated by [4], which can function as a primary information dissemination platform.…”