See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/364494310

Suicidal Analysis on Social Networks Using Machine Learning

Chapter · September 2022

DOI: 10.4018/978-1-6684-3533-5.ch012

CITATIONS READS
9 206

2 authors, including:

Kanojia Sindhuben Babulal

Central University of Jharkhand
43 PUBLICATIONS 162 CITATIONS

SEE PROFILE

All content following this page was uploaded by Kanojia Sindhuben Babulal on 23 November 2022.

The user has requested enhancement of the downloaded file.

230

Chapter 12
Suicidal Analysis on
Social Networks Using
Machine Learning
Kanojia Sindhuben Babulal
https://orcid.org/0000-0003-0442-8795
Central University of Jharkhand, India

Bashu Kumar Nayak

Central University of Jharkhand, India

ABSTRACT
Suicides are the most critical issues in the present time. Early detection and prevention can assure the
safety for the people’s lives. As the technology increases rapidly, we are moving towards online chan-
nels to express our suicidal thoughts. In the chapter, the authors deal with suicidal ideation through the
user generated post on different platforms like Twitter, Facebook, Reddit, Suicide Watch, etc. Analyzing
the text, they enrich the knowledge and that can be used as an indicator for suicidal thoughts. To detect
suicidal thoughts, they use text processing using NLP, and some features are generated that can be
classified using different classifiers like random forest, SVM, naïve bayes, etc., and some neural net-
work models like CNN, LSTM, BERT, etc. are also used for final prediction of suicidal or non-suicidal
thoughts. In this chapter, the authors use Distill Bert model for predicting the results and also improve
the accuracy by changing the hyperparameters. Here, they summarize the existing work’s limitations
and discuss future research directions.

INTRODUCTION

In the Present time social networking sites are exploding by the users. People are more drawn to the
virtual life since the introduction of Facebook, Instagram, Twitter, Snapchat, Reddit, and other social
networking websites. As these websites are quite user friendly so the no of users is increasing day by
day. There are almost 900 social networking sites present on the internet at this time, and those sites have

DOI: 10.4018/978-1-6684-3533-5.ch012

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Suicidal Analysis on Social Networks Using Machine Learning

transformed people’s perceptions and points of view (Liang & Dai, 2013). We also discovered that there
is a distinction between real life and virtual life among the individuals around us. Even we acknowledge
that people are more drawn to virtual life than actual life. Here we are only talking about twitter which
is one of the most famous social sites.
In 2017, almost 500 million tweets were sent every day, with a daily active user base of 100 million
(Aslam, 2018). Here we can post about the social issues, our own thoughts etc. But Some people also
used this platform to put suicidal thoughts in their tweets. As we come to know, Suicide is the 2nd larg-
est cause of mortality among 15–29-year-olds all over the world which is given by WHO (World Health
Organization) (Bilsen, 2018). Every year, over 80000 individuals die by suicide, or one human being every
40 seconds. Suicide has become a social disease in recent years, and we should think about it. Suicide is
motivated by a lot of factors. Suicide is more likely in those with depression, but people without having
depression might also experience suicidal thinking. Suicide variables are divided into three categories
by AFSP: health factors, historical factor and environmental factors (Ferrari et al., 2014). We investigate
the challenge of recognising suicide ideation via social networking websites in this study, with attention
on comprehending and identifying suicidal ideas in online content. To comprehend suicidal thinking
from a data mining standpoint, we do rigorous analytics of the content, language preferences, and topic
description. Suicidal ideation was detected in the data using different methods of Machine Learning.
By using feature engineering and classification algorithms, it is the most effective method for detecting
suicide ideation in internet content.

LITERATURE REVIEW

Description of Previous Work

Some of the last few research papers which we have gone through and found most research gener-
ally based on Content Analysis, Feature Engineering, Deep learning etc. (De Choudhury et al., 2013)
investigated the ability to identify and predict major episodes of depression in Twitter users through
social media. The researcher used crowd-sourcing approaches to create a group of Twitter users who
scored high on the CES-D (Center for Epidemiologic Studies Depression Scale) scale for depression
and others who scored low. Furthermore, online linguistic patterns mirror previous findings about sad
people’s language use (Rude, Gortner, & Pennebaker, 2004). Researchers (De Choudhury et al., 2016)
have demonstrated that language traits can help identify those who are shifting from mental discourse
on social media to suicidal thoughts.
Authors Coppersmith et al. (2016) looked at data posted by users from Twitters before a suicide at-
tempt and conducted an empirical study of the language and emotions stated. One of the study’s most
surprising findings is the increase in the percentage of tweets expressing anguish in the weeks preced-
ing up to a suicide attempt, followed by a substantial jump in anger and despair within the week after a
suicide attempt. In a similar vein, O’dea et al. (2015) confirmed that people use Twitter to communicate
suicidality and shown that, using both human coders and an artificial machine classifier, it is possible to
discern the level of worry among suicide-related tweets. Braithwaite et al. (2016) found that algorithms
of machine learning are effective in distinguishing those at risk of suicide from those who are not.
In a paper in 2018 Vioules et al. (2018) took the method of automatically detecting unexpected
changes in user behavior online. They combine techniques of natural language processing with textual

231
 
Suicidal Analysis on Social Networks Using Machine Learning

and behavioral features and run these features using a framework popularly known as martingale, which
is commonly used to identify rapid changes in data streams, to detect the change. Syntactic, statistical,
word embedding, linguistic and topic features are extracted and compared with 6 classifiers, including 4
supervised classifiers and 2 neural network models by Ji et al. (2018). Researchers Colombo et al. (2016)
employ a social networking graph to look into the connectedness and communication characteristics of
users from twitters who publish information that is consistently categorized by a human annotator as
possibly containing suicide intent known as suicidal ideation.
Authors (Coppersmith et al., 2016) look at data from twitter users who tried suicide and do an pre-
liminary analysis of linguistic patterns and emotions on the social media platform. They discovered
quantified suicide attempt signs in the language of social media data and used a basic machine learning
classifier to measure its performance. Masuda, Kurahashi, & Onari, (2013) used logistic regression to
uncover user characteristics that lead to suicide thoughts, both connected and unrelated to social networks.
Suicide ideation is defined as a user’s participation in at least one active user-identified suicide commu-
nity. To analyse suicide factors, Chattopadhyay (2007) used the PSIS (Pierce suicidal intent scale) and
did regression analysis. Abboute et al. (2014) offer a comprehensive method for automatically collect-
ing questionable tweets based on the lexicon of subjects that suicidal people commonly discuss. Based
on a basic classification algorithm, the automatically recorded tweets suggest suicidal risk behaviour.
Psychiatrists will be able to consult questionable tweets and profiles related with these tweets using a
new interface. Authors Okhapkina, Okhapkin, & Kazarin (2017) compiled a glossary of terminology
related to suicidal behaviour. For messages, they use TF-IDF matrices, and for matrices, they use SVD.
Mulholland & Quinn (2013) used a classifier to predict the chance of linguistic suicide by extracting
lexical as well as syntactic features. In research paper Huang et al. (2014) used HOWNET to create a
psychological lexical dictionary and used Support Vector Machine to detect Cyber Suicide in Chinese
blogs. They extracted different useful features such as syntactic, syntactical, LIWC, word embedding,
and topic features, and use them in a classifier. They compared four logistic regression classification
algorithms, including Random Forest, Decision Tree, Gradient Boosting, and XGBoost. Pestian et al.
(2010) examined the performance of several multivariate approaches using word count, pos concepts,
and readability ratings as criteria. Machine learning techniques assists successfully identification of high
suicide risk. Nobles et al. (2018) used a multilayer perceptron to feed psycholinguistic characteristics and
word occurrences (MLP). To encode user’s post, (Shing et al., 2018) used user level CNN with a filter
size of 3,4,5. Textual sequences are encoded using an LSTM network, a common variation of RNN,
which is then processed for classification using a fully connected layer. For updating neural networks,
model was presented for aggregation method. For example, in private chatting rooms suicidal ideation
can be detected using LSTM and CNN targeting. Benton, Mitchell, & Hovy (2017) used neural network
models to predict mental health and suicide attempts through multitask learning. They used a bidirectional
LSTM for sequence encoding, and a self-attention method to collect the subsequence that is instructive
in a model of deep learning. For detecting intention related to suicide Sawhney et al. (2021) employed
RNN, CNN and LSTM. Authors Hevia, Menéndez, & Gayo-Avello (2019) used various models, includ-
ing a GRU-based RNN, to assess the effect of pretraining.

232

Suicidal Analysis on Social Networks Using Machine Learning

Application Domains for Suicidal Ideation

Most of the techniques of Machine Learning are used for suicidal ideation detection. This chapter can
also be viewed according to different data sources. The Application domain we cover including Live
Surveys, Questionnaires, suicide notes and online user content etc

• Live Surveys: For analysing the data we have collected some information related to suicide oc-
cured in the last 3 months from some school or college going students. This data is collected in csv
format where we can apply different machine learning algorithms or Neural Network techniques
for fulfilling the purpose of suicidal ideation detection. Apart from that we can also extract some
information like problem faced by the victim, reason for performing suicide, techniques used by
the victim etc.
• Questionnaires: The questionnaires are created based on a set of criteria and examination metrics
for self-assessment.
• Suicide Notes: The person who are going to commit suicide write suicide notes. Suicide notes
are generally written on online blogs, letters and recorded in audio or video format. These notes
are helpful for analysing the cause of death. These notes are served as material for NLP research.
• User-generated content on the Internet: People can openly communicate their feelings, emo-
tions, and life events thanks to the increased usage of mobile internet and social networking ser-
vices in recent years. As we found an anonymous space for online discussion, an increase in the
number of people who are suffering from mental problem, anxiety, depression seek for help.
Potential Suicide victims post their suicidal thoughts on social networking websites like Facebook,
Twitter, Reddit, Snapchat etc. As the data are generated from the social networking websites each
second so the researchers get an opportunity to analyse the data and perform some operation for
getting the intent from the post.

DATASET AND METHODOLOGY

Twitter: Twitter is a widely used social media platform, and many users use it to discuss suicidal
thoughts. In terms of length of the post, anonymity, and communication and interaction, Twitter differs
significantly from Reddit. For performing the analysis part, we have collected the data from GITHUB
Repository(https://raw.githubusercontent.com/laxmimerit/twitter-suicidal-intention-dataset/master/
twitter-suicidal_data.csv). The repository contains both suicidal and non-suicidal intent tweet data. The
following is the format of the data set:

Table 1.­Format of Twitter Data

TWEET INTENTION
I just want to die today 1
Today is a bad day for me I just want to take my life 1
I am quite happy today 0

233
 
Suicidal Analysis on Social Networks Using Machine Learning

In this case, the text data is Tweet, and the goal label is intention. Observing the format, we can see
that the post containing suicidal thoughts have intention 1 and the post containing non suicidal thoughts
have intention 0
Before going deep into the model let us first understand some important terminologies which are
required in predicting the sentiments of the text given by the user to the model.

1. Text Pre-Processing: It has long been regarded as a crucial stage in natural language processing.
It converts a text into a more consumable format to enhance the performance of machine learning
algorithms. Generally, Text Pre-Processing deals with different steps such as Tokenization, remov-
ing whitespaces, removing punctuation, stop words, lemmatization and noise removal.
2. Feature Extraction: Eminent feature extraction is an important step for better results (Priyanka &
Babulal, 2022). We retrieved many features after pre-processing and cleaning the data, including
statistics, word-based features, TF-IDF, LIWC, semantics, and syntactics.
3. Statistical features: The length of the posts created by users varies, and statistical information can
be gleaned from the texts. Short and simple sentences are used in certain posts, while complicated
and extensive ones are used in others. Following segmentation and tokenization, the following
statistical features were captured:
a. No sign of tokens, characters and words, or sentences in the title
4. Syntactic Features (POS): In the work of natural language processing, syntactic features are useful
information. To capture the same grammatical aspects in the users’ posts, we use parts of speech
as features for our suicidal ideation detection model.
5. LIWC: The phrases “emotions,” “anxiety,” “loneliness,” “depression,” and “harassment” appear
in the user’s posts. For extracting these characteristics, lexicons are commonly utilized. We used
LIWC (Longest Inquiry Word Count) to analyses the linguistic and emotional qualities.
6. Word Frequency features TF-IDF: Suicide is associated with a variety of phrases. We extract
these traits using TF-IDF, which assesses the relevance of different words in both suicidal and non-
suicidal messages. The TF-IDF algorithm counts the occurrence of each word in the document and
assigns a penalty based on its occurrence of that word in the given corpus.
7. Word Embedding Features: Popular word embedding techniques like Word2Vec and Glove are
used to convert natural language text into distributed vector space. CBOW and Skip-gram are the
main architectures for word2Vec embedding. CBOW predicts the current word based on the context
whereas Skip-gram predicts the word that is closest to the current word.
8. Topic Features: Suicidal and non-suicidal posts discuss different issues that can help people un-
derstand the two categories.

Introduction to BERT Model

BERT stands for Bidirectional Representations from Transformers. It is a pre-trained model which contains
a lot of unlabelled text including the entire Book Corpus (800 M words) and Wikipedia (2500 M words).
Bert model works in both directions. i.e it can get information from both the left and the right side
of a token’s context during its training phase. The concept of bidirectional is important for understand-
ing the meaning of a language. Let us consider an example to illustrate it. We take two sentences in this
example and both of them involve the word “bank”

234

Suicidal Analysis on Social Networks Using Machine Learning

1. We went to the river bank.

2. I need to go to bank to make a deposit.

In both the sentences the bank has different context according to the sentences so rather choosing
only left context or right context we cannot able to understand the meaning so we choose both left or
right context for better understanding of the word.
Bert uses two pre-training techniques:

1. Masked LM(MLM): In Masked LM technique the word sequences are used as an input to the
BERT. After that 15% of the word in each word sequence are replaced with the token termed as
[MASK], then the model tries to predict the initial value of the masked words, based on the context
provided by other non-masked words within the sequence. Steps to predict the output Words:

a. Add a classification layer on the top of Encoder Output

b. The embedding matrix is multiplied by the output vectors before being transformed into the
vocabulary dimension.
c. Using SoftMax, assess the probability of each word in the lexicon.

The Bert loss function only considers masked value predictions and ignores non-masked word pre-
dictions. As a result, it takes longer for the model to converge than directed models.

Figure 1. Masked LM (MLM)

235
 
Suicidal Analysis on Social Networks Using Machine Learning

2. Next Sentence Prediction: - In this technique we provide two sentences as an input such as
Sentence A and Sentence B then it is used to learn and predict whether the second sentence is the
next sentence in the original document.

To distinguish the sentences in training the input is processed in the following ways:

• A token termed as [CLS] is placed at the beginning of the first sentence, and a token [SEP] is
placed at the end of each sentence.
• A sentence embedding denotes the addition of Sentences A or Sentence B to each token.
• Each token has a numeric assigned that indicates its place in the sequence.

Steps to predict the second sentence is connected to first sentence are given below:

• The transformer model is used to process the entire input sequence.

• A basic classification layer is used to turn the [CLS] token’s output into a 2X1 shaped vector.
• Softmax is used to calculate the probability of the Next Sequence.

Bert’s Architecture

BERT’s architecture is built around transformers. Currently, there are two options available BERT Base
in which there are 12 layers, 12 attention heads, and 110 million parameters in the BERT Base and BERT
large where there are 24 layers, 16 attention heads, and 340 million parameters in this model. For better
understanding we have to know about the Transformer Model Architecture

236

Suicidal Analysis on Social Networks Using Machine Learning

Figure 2. Transformer Model Architecture

The Encoder block has a Multi-Head Attention layer, which is followed by a Feed Forward Neural
Network layer. A Masked Multi-Head Attention has been added to the Decoder. The encoder and decoder
stacks both have the same number of units. A hyperparameter is the number of encoder and decoder units.
Now we’ll look at how the encoder and decoder stacks work:

237
 
Suicidal Analysis on Social Networks Using Machine Learning

• The input sequence’s Word embedding is sent to the first encoder.

• After that, it was converted and passed on to the next encoder.
• After that, all decoders receive the output from the last encoder.

Figure 3. Encoder Decoder Stack

Self-attention allows you to look at the other words in the input sequence to help you grasp a par-
ticular word. Decoder attention assists the decoder in focusing on the correct bits of the input sequence.

238

Suicidal Analysis on Social Networks Using Machine Learning

Text Pre-Processing in BERT Model

Figure 4. Text Pre-processing

By Vladimir Ilievski

239
 
Suicidal Analysis on Social Networks Using Machine Learning

The Bert Model describes a set of rules that will be used to represent the input text. The input embedding
is made up of three different embeddings:

1. Position Embeddings: It learns and employs positional embeddings to represent a word’s position
in a sentence.
2. Segment Embeddings: This assignment can accept sentence pairs as input (Question-Answering).
3. Token Embeddings: In this Embedding, the representation of a particular token is built by adding
the relevant token, segment, and position embeddings.

ARCHITECTURE FOR SUICIDAL IDEATION DETECTION

In this work our focus is on detecting the suicidal and non-suicidal text with maximum accuracy. For
detecting those words in the dataset, we use a pre-trained model DISTIL BERT for better classification.
As we can observe that our first task is to collect data from different social networking websites such
as Facebook, Twitter, Snapchat, Instagram etc. As the data collected from different resources contains
lot of information including emoticons, symbols etc which are not good for the Pre-processing task.
So, the idea of deleting the emoticons, symbols etc known as cleaning or pre-processing the text data
from the dataset. After cleaning or Pre-processing, the data is transferred to our model which is going
to predict the sentiment and classified the text as suicidal or non-suicidal text with a maximum accuracy

IMPLEMENTATION OF DISTIL BERT MODEL

IDE Required for Implementation Google Collab

Collab is a cloud based Jupyter notebook environment that is free to use. It doesn’t require any setup,
and the notebooks we can make can be changed at the same time. Many prominent machine learning
libraries are supported, and they may be readily imported into the notebook.
(https://www.tutorialspoint.com/google_colab/what_is_google_colab.html.)

Libraries Required for Implementation

• NumPy: Numerical Python is referred to as NumPy. It is generally used to work with arrays. It
consists a lot of functions for working in the field of linear algebra, Fourier transform and matri-
ces. It aims to provide an array object that is upto 50x faster than traditional python lists https://
www.w3schools.com/python/numpy/numpy_intro.asp
• Pandas: It is generally used to analyse the data . One of the important function of panda is read_
csv() which is used to load the data set in the editor for further execution. https://www.w3schools.
com/python/pandas/default.asp
• TensorFlow: It is a Google-created and distributed python library for fast numerical process-
ing. It is used to generate models of deep learning directly or by using wrapper libraries cre-
ated on top of TensorFlow to make the process easier. https://machinelearningmastery.com/
introduction-python-deep-learning-tensorflow/

240

Suicidal Analysis on Social Networks Using Machine Learning

• Ktrain: It is a library that is used in the deep learning software framework to construct, train,
debug, and deploy neural networks. It is inspired by fastai library. With few lines of code, we can
estimate an optimal learning rate for our model on given data using a learning rate finder.
• Sklearn: In Python, it is the widely used and stable library. It’s used to choose the best statistical
modelling methods.

CODING

#Install all libraries

import pandas as pds
import tensorflow as tsf
import ktrain
from ktrain import text
from sklearn.model_selection
import train_test_split
#Loading the dataset
dframe=pds.read_csv(“https://raw.githubusercontent.com/laxmimerit/twitter-sui-
cidal-intention-dataset/master/twitter-suicidal_data.csv”)
dframe.head()
#Creating train set and test set
target=[‘intention’]
data=[‘tweet’]
Xd=dframe[data]
yt=dframe[target]
trx,tstx,try,tsty=train_test_split(Xd,yt,test_size=0.3,random_state=0)
#Common parameters
max_length=25
batchsz=6
learning_rate=1e-4
epochs=1

#Defining the model

model_=’distilbert-base-uncased’
t_mod=text.Transformer(model_,maxlen=max_length,classes=[0,1])

241
 
Suicidal Analysis on Social Networks Using Machine Learning

#Converting Split Data to List

TRX=trx[‘tweet’].tolist()
TRY=try[‘intention’].tolist()
TSX=tstx[‘tweet’].tolist()
TSY=tsty[‘intention’].tolist()
#preprocessing training and test data
train=t_mod.preprocess_train(TRX,TRY)
test=t_mod.preprocess_train(TSX,TSY)

#Classifier

model=t_mod.get_classifier()
lnr=ktrain.get_learner(model, train_data=train,val_data=test,batch_
size=batchsz)
#Plotting Learning Rate
lnr.lr_plot():- it is used to plot the learning rate.

242

Suicidal Analysis on Social Networks Using Machine Learning

Figure 5. Learning Rate

#Training Model

lnr.fit_onecycle(learning_rate,epochs)
#Evaluating the Model and printing the classification report
x=lnr.validate(class_names=t_mod.get_classes()

RESULT AND DISCUSSION

Here we found the accuracy of 92% by using only one epoch which is quite better as compared to previ-
ous research work. Macro average is averaging the unweighted mean per label while weighted average is
averaging the support-weighted mean per label. The accuracy of the model can be upgraded to a better
extent by changing the epoch value. Accuracy 92% is obtained with Precision, recall f1-score and sup-
port parameters. Macro Average and Weighted Average is also 92%.

243
 
Suicidal Analysis on Social Networks Using Machine Learning

Table 2. Accuracy with difference parameters

Precision Recall f1-score Support

0 0.93 0.93 0.93 1560
1 0.91 0.91 0.91 1176
Accuracy -- -- 0.92 2736
Macro Avg 0.92 0.92 0.92 2736
Weighted Avg 0.92 0.92 0.92 2736

LIMITATIONS

1. Shortage of Data: The shortage of data is the most important issue in the current research. Most
of the current solutions depend on supervised learning technique, which needs manual annotation.
As far as we know, there is a scarcity of annotated data to help future research.
2. Annotation Bias: There isn’t much evidence to back up the suicide attempt to get ground truth.
As a result, current data is gathered through hand labelling using established annotation criteria.
Labels may be skewed as a result of crowdsourcing-based annotation. In terms of demographic
statistics, the data related to suicide are quite complex, because estimation of mortality is based on
general death rather than suicide. Some suicide related incidents are also misclassified as accidents
or deaths.
3. Unbalanced Data: In the Massive social post there are less proportion of suicidal intent.so we
can see that there is imbalance of data i.e., the post contain more non-suicidal text as compared to
suicidal text.
4. Lack of Understanding of Suicidal Intention: Suicidal intention was not well understood by
the existing statistical learning techniques. Suicide attempts are complicated psychologically. To
improve predicting performance, the major technique is to focus on picking characteristics or using
complicated neural architectures.

CONCLUSION

Suicidal Prevention is the important task in this new technology based on modern society. Early iden-
tification of suicidal ideation detection is the best way to prevent suicide. We learn about numerous
approaches for detecting suicidal ideation by reading this research, such as lexicon-based analysis, word
cloud visualisation, and feature engineering that includes tabular, textual, and emotional data. We also
found some deep learning model such as RNN, LSTM are used for detecting the suicidal or non-suicidal
text on the given dataset. We may conclude that, while we have other domains to analyse, online user
material will be the primary avenue for detecting suicidal ideation in the future. As a result, developing
new approaches and using new models to detect suicide intent in online content with optimum accuracy
is critical. Apart from suicidal ideation detection we should also focus on preventing the suicide by de-
veloping some API or Software’s. In the current work we are limited to only one social media platform
i.e. Twitter but for further research work different platforms can be used such as Weibo, Instagram,
Snapchat etc. Lots of pretrained model are already used in this paper but XLNET can be a better model

244

Suicidal Analysis on Social Networks Using Machine Learning

for getting maximum accuracy. Apart from all these models we can also take reference of Unsupervised
learning and Reinforcement learning for getting better results.

REFERENCES

Abboute, A., Boudjeriou, Y., Entringer, G., Azé, J., Bringay, S., & Poncelet, P. (2014). Mining twitter for
suicide prevention. Paper presented at the International Conference on Applications of Natural Language
to Data Bases/Information Systems.
Aslam, S. (2018). Twitter by the numbers: Stats, demographics & fun facts. Omnicoreagency.com.
Benton, A., Mitchell, M., & Hovy, D. (2017). Multi-task learning for mental health using social media
text. arXiv preprint arXiv:1712.03538.
Bilsen, J. (2018). Suicide and youth: Risk factors. Frontiers in Psychiatry, 9, 540. doi:10.3389/
fpsyt.2018.00540 PMID:30425663
Braithwaite, S. R., Giraud-Carrier, C., West, J., Barnes, M. D., & Hanson, C. L. (2016). Validating
machine learning algorithms for Twitter data against established measures of suicidality. JMIR Mental
Health, 3(2), e4822. doi:10.2196/mental.4822 PMID:27185366
Chattopadhyay, S. (2007). A study on suicidal risk analysis. Paper presented at the 2007 9th International
Conference on e-Health Networking, Application and Services. 10.1109/HEALTH.2007.381606
Colombo, G. B., Burnap, P., Hodorog, A., & Scourfield, J. (2016). Analysing the connectivity and
communication of suicidal users on twitter. Computer Communications, 73, 291–300. doi:10.1016/j.
comcom.2015.07.018 PMID:26973360
Coppersmith, G., Ngo, K., Leary, R., & Wood, A. (2016). Exploratory analysis of social media prior to
a suicide attempt. Proceedings of the third workshop on computational linguistics and clinical psychol-
ogy. 10.18653/v1/W16-0311
De Choudhury, M., Gamon, M., Counts, S., & Horvitz, E. (2013). Predicting depression via social
media. Paper presented at the Seventh international AAAI conference on weblogs and social media.
De Choudhury, M., Kiciman, E., Dredze, M., Coppersmith, G., & Kumar, M. (2016). Discovering shifts
to suicidal ideation from mental health content in social media. Proceedings of the 2016 CHI conference
on human factors in computing systems. 10.1145/2858036.2858207
Ferrari, A. J., Norman, R. E., Freedman, G., Baxter, A. J., Pirkis, J. E., Harris, M. G., ... Vos, T. (2014).
The burden attributable to mental and substance use disorders as risk factors for suicide: Findings from
the Global Burden of Disease Study 2010. PLoS One, 9(4), e91936. doi:10.1371/journal.pone.0091936
PMID:24694747
Hevia, A. G., Menéndez, R. C., & Gayo-Avello, D. (2019). Analyzing the use of existing systems for
the clpsych 2019 shared task. Proceedings of the Sixth Workshop on Computational Linguistics and
Clinical Psychology.

245
 
Suicidal Analysis on Social Networks Using Machine Learning

Huang, X., Zhang, L., Chiu, D., Liu, T., Li, X., & Zhu, T. (2014). Detecting suicidal ideation in Chinese
microblogs with psychological lexicons. Paper presented at the 2014 IEEE 11th Intl Conf on Ubiquitous
Intelligence and Computing and 2014 IEEE 11th Intl Conf on Autonomic and Trusted Computing and
2014 IEEE 14th Intl Conf on Scalable Computing and Communications and Its Associated Workshops.
10.1109/UIC-ATC-ScalCom.2014.48
Ji, S., Yu, C. P., Fung, S., Pan, S., & Long, G. (2018). Supervised learning for suicidal ideation detection
in online user content. Complexity. doi:10.1155/2018/6157249
Liang, P.-W., & Dai, B.-R. (2013). Opinion mining on social media data. Paper presented at the 2013
IEEE 14th international conference on mobile data management. 10.1109/MDM.2013.73
Masuda, N., Kurahashi, I., & Onari, H. (2013). Suicide ideation of individuals in online social networks.
PLoS One, 8(4), e62262. doi:10.1371/journal.pone.0062262 PMID:23638019
Mulholland, M., & Quinn, J. (2013). Suicidal tendencies: The automatic classification of suicidal and
non-suicidal lyricists using nlp. Proceedings of the sixth international joint conference on natural lan-
guage processing.
Nobles, A. L., Glenn, J. J., Kowsari, K., Teachman, B. A., & Barnes, L. E. (2018). Identification of im-
minent suicide risk among young adults using text messages. Proceedings of the 2018 CHI Conference
on Human Factors in Computing Systems. 10.1145/3173574.3173987
O’dea, B., Wan, S., Batterham, P. J., Calear, A. L., Paris, C., & Christensen, H. (2015). Detecting sui-
cidality on Twitter. Internet Interventions: The Application of Information Technology in Mental and
Behavioural Health, 2(2), 183–188. doi:10.1016/j.invent.2015.03.005
Okhapkina, E., Okhapkin, V., & Kazarin, O. (2017). Adaptation of information retrieval methods for
identifying of destructive informational influence in social networks. Paper presented at the 2017 31st
International Conference on Advanced Information Networking and Applications Workshops (WAINA).
10.1109/WAINA.2017.116
Pestian, J., Nasrallah, H., Matykiewicz, P., Bennett, A., & Leenaars, A. (2010). Suicide note classifica-
tion using natural language processing: A content analysis. Biomedical Informatics Insights, 3, S4706.
Priyanka, & Babulal, K. S. (2022, August 12). Hematological image analysis for segmentation and
characterization of erythrocytes using FC-TriSDR. Multimedia Tools and Applications. Advance online
publication. doi:10.100711042-022-13613-5
Rude, S., Gortner, E.-M., & Pennebaker, J. (2004). Language use of depressed and depression-vulnerable
college students. Cognition and Emotion, 18(8), 1121–1133. doi:10.1080/02699930441000030
Sawhney, R., Joshi, H., Gandhi, S., Jin, D., & Shah, R. R. (2021). Robust suicide risk assessment on
social media via deep adversarial learning. Journal of the American Medical Informatics Association:
JAMIA, 28(7), 1497–1506. doi:10.1093/jamia/ocab031 PMID:33779728
Shing, H.-C., Nair, S., Zirikly, A., Friedenberg, M., Daumé, H., III, & Resnik, P. (2018). Expert, crowd-
sourced, and machine assessment of suicide risk via online postings. Proceedings of the fifth workshop
on computational linguistics and clinical psychology: from keyboard to clinic. 10.18653/v1/W18-0603

246
 
Suicidal Analysis on Social Networks Using Machine Learning

Vioules, M. J., Moulahi, B., Azé, J., & Bringay, S. (2018). Detection of suicide-related posts in Twitter
data streams. IBM Journal of Research and Development, 62(1), 1-7.

247

View publication stats