Early Detection of Suicidal Tendencies from Text Data using LSTM
Debadyuti Bhattacharya, Sri Hari Karthick N, A. Shahina · 2021 Innovations in Power and Advanced Computing Technologies (i-PACT) · 2021
Social networking sites provide us with an abun-dance of textual data for analysis, when combined with a neural network to consume data to any level depending on its architecture, it makes the frontier of social media suitable for textual analysis. One such analysis is the ability to detect intentions of suicide from text content posted by users on such platforms. Although conventional algorithms in machine learning serve this purpose to a certain degree, neural networks that mainly specialize in handling text content such as Long Short Term Memory networks perform this function more efficiently and accurately preserving temporal properties, while still being flexible enough to neither over-fit nor under-fit through the control of the depth of the network. This suits to the varying volume of data available for training, which is collected through the scraping of publicly available content posted by actual users, so that the trained model's performance resembles the real world, when compared against conventional classifying algorithms, prove its efficiency through several standard metrics.