Exploring the Impact of Word Embedding Dimensions on Depression Data Classification Using BiLSTM Model

Depression is a mental health issue that impacts over 300 million individuals worldwide. Daily anxiety impacts the relationships of depressed individuals with their family and friends, affects their health, and, in the worst-case scenario, leads to suicide. Due to the an increase of social networks, a vast majority of people now express their thoughts, feelings, and emotions via social media. The researchers mainly focused on features and building predictions, making it difficult for them to detect depression via social media. Word embedding offers a potential solution for processing and transforming unstructured data into meaningful representations. Embedded words represent textual data as numeric numerals with equivalent meaning. Using the Bidirectional Long Short-Term Memory (Bi-LSTM) paradigm, this investigation attempts to experiment with various weighting dimensions using the Word2Vec and GloVe methods. The data used is Depression: Reddit Dataset labelled: depressed and non-depressed. The experimental results of word weight dimensions using Word2Vec with a dimension of 500 are better than Glove. The results of accuracy, precision, recall and f1-score for the Word2Vec method were 96.22%, 97.02%, 95.30% and 96.15%, while the Glove method was 95.91%, 96.40%, 95.30% and 95.85% with a dimension of 200. Overall, Word Embedding Dimensions influences the results of this study. Based on these findings, it is concluded that the proposed Word2Vec method is a significant approach.