IRJET- Social Network Mental Disorders Detection Via Online Social Media Mining

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 01 | Jan 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 1895
Social Network Mental Disorders Detection via Online Social
Media Mining
Hardik kathed1, Ashish Dixit2, Shubham Bhongale3, Vishvabhushan Gaikwad4
1,2,3,4Department of Computer Engineering, SKN Sinhgad Institute of Technology and Science Lonavala,
Maharashtra, India
----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Mental disorders are becoming a threat to
people’s health now days. With the rapid pace of life, more
and more people are feeling stressed. It is veryhardtodetect
users mental disorders in an early time to protect user. An
increasing number of social network mental disorders
(SNMDs), such as Cyber-RelationshipAddiction, Information
Overload, and Net Compulsion, have been recently noted.
Symptoms of these mental disorders are usually observed
passively today, resulting in delayed clinical intervention
.With the increase in use of electronic long range
interpersonal communication, people are accustomed to
sharing their step by step exercises and interfacing with
companions by means of online systems administration
media stages, making it conceivable to utilize online
informal community information for stress discovery.
Through our system, we find that users stress state is closely
related to that of his/her friends in social media, and we
employ a large-scale dataset from real-world social
platforms to systematically study the correlation of users’
stress states and social interactions. Variousaspectsof social
network mental disorders are defined using a set of stress-
related textual, visual, and social attributes , I proposed a
system using CNN by which we can do sentiment analysis of
facebook post, after Formation of topic using Transductive
Support Vector Method(TSVM) we can classify user are in
detecting mental disorders or not. After classification user
are in mental disorders or not k-nearest neighbour’s
algorithm (KNN) is used for recommendation of hospital on
a map as well as Admin can send mail of precaution list for
user to become healthy and happy in life.
Key Words: Feature Extraction, Healthcare, Online
Social Network, Mental Disorder Detection, Social
Media, Social Interaction.
1. INTRODUCTION
A mental disorder is turning into a risk toindividual’swell-
being these days. With the fast pace of life, progressively
and more individuals are feeling stressed. With the
explosive growth in popularity of social networking and
messaging apps, online social networks (OSNs) have
become a part of many people’s daily lives. Excessive and
chronic stress can be rather harmful to people’s physical
and mental health. Users’ social interactions on social
networks contain useful cues for mental disorders
detection. Social psychological studies have made two
interesting observations. The first is mood contagions: a
bad mood can be transferred from one person to another
during social interaction. The second is Social Interaction:
people are known to social interactionofuser.Throughthe
advancement of social networks like Instagram Post
dataset, Facebook post dataset, people share their every
day events and moods, and interact with friends through
the social networks in an ever increasing number.Userare
in stress or not can be classified using support vector
method. Due to leverage both facebook post content
attributes and social interactions helps to enhance stress
detection. After getting mental disorders level, system can
recommended user hospital for further treatment, Admin
can show that hospital on map and system also
recommended to take precaution for avoid mental
disorder.
2. LITERATURE SURVEY
H. Lin et al [1] states the around a programmed pressure
recognition strategy from cross-media microblog data.
Structure of three levels for pressure location from cross-
media microblog information. By consolidating a Deep
Sparse Neural Network to fuse distinctive highlights from
cross-media microblog information, the system is very
possible and effective for push detection. This structure,
the proposedtechniquecan helptoconsequentlyrecognize
mental worry from informal organizations.H.Linintendto
examine the social relationships in mental worry to
additionally enhance the identification execution.
Liqiang Nie et al [2] proposed about bridging the
vocabulary hole between wellbeing searchers and human
services information with a worldwide learning approach.
A rapeutic phrasing task plan to connect the vocabulary
hole between wellbeing searchers and social insurance
information. The plan includes two segments,
neighborhood mining and worldwide learning .Extensive
assessments on a true dataset show that our plan can
create promising execution when contrasted with the
overarching coding techniques. Liqiang Nie will explore
how to adaptably compose the unstructured restorative
substance into client needs-mindful cosmology byutilizing
the suggested therapeutic wordings.
Chi Wang et al [3] introducing an find out around an
impact boost issue, which expects tolocatea littlesubset of
hubs (clients) in an interpersonal organization that could
expand the spread of impact. A Pairwise Factor Graph
(PFG) model to formalize the problem in probabilistic
model, and Chi Wang extend it by incorporating the time

information, which results in the Dynamic Factor Graph
(DFG) mode. The proposed approach can effectively
discover the dynamic social influences. Parallelization of
our algorithm can be done in future work to scale it up
further.
Lexing Xie and Xuming He.[4] have presented about
Picture labels and world information: taking in label
relations from visual semantic sources examines the
utilization of regular words to depict pictures. The
proposed labeling calculation sums up to concealed labels,
and is additionally enhanced joining tag-connection
highlights acquired by means of ICR. Techniques to better
fuse multi-word terms and out-of vocabulary words;
propelled NLP procedures for taking in word relations
from freestyle content; assessment of idle idea connection
recommendation, and anticipating the kind of relations.
Yuan Zhang et al [5] proposed learn a novel problem of
emotion prediction in social networks. A strategy alluded
to as Moodcast for demonstrating and foreseeing feeling
flow in the informal organization. The new approach can
enough show each customer's inclination status and the
desire execution is better than a couple of benchmark
procedures for feeling forecast. It is used to as a result of
the set number of individuals. For display learning, it
utilizes a Metropolis-Hastings calculation to get a rough
arrangement. Trial comes about on two diverse genuine
informal communities exhibit that the proposed approach
can successfully displayeveryclient’sfeelingstatusandthe
forecast execution is superior to a few standard strategies
for feeling expectation.
Michela Ferron et al [6] presented Studies about Daily
pressure acknowledgment from cell phone information,
climate conditions and individual attributes. That step by
step pressure can be constantly seen in perspective of
behavioral measurements. This is got from the customer's
mobile phone activity what's more, from additional
markers, for instance, the atmosphere conditions (data
identifying with fleeting properties of the condition) and
the character characteristics. Stress has turned into a
major issue influencing profitability in workplaces,
prompting word related issues and causing wellbeing
diseases. This framework could be broadened and utilized
for early discovery of stress-related clashes and stress
virus, and for supporting adjusted workloads.
Dan C Ciresan et al [7] introduced an new deep CNN
architecture, MaxMin-CNN, to better encode both positive
and negative filter detections in the net. Dan C Ciresan
propose to adjust the standard convolutional square of
CNN keeping in mind the end goal to exchange more data
layer after layer while keeping some invariance inside the
system. Our fundamental thought is to abuse both positive
and negative high scores got in the convolution maps. This
conduct is acquired by altering the customary enactment
work venture before pooling1.Time required for this is
more. It is time consuming process.
Jennifer Golbeck et al [8] presented an inspired by the
personality of customers. Character has been seemed, by
all accounts, to be appropriate to numerous sorts of
collaborations. Jennifer Golbeck are occupied with the
personality of customers. Character has been had all the
earmarks of being pertinent to numerous sorts of
collaborations; it has been seemed, by all accounts, to be
useful in suspecting work fulfillment, relationship
accomplishment, and even slant. Cristina Robles are
charmed in the character of customers. Characterhasbeen
gave off an impression of being pertinent to numerous
sorts of interchanges; it has been seemed, by all accounts,
to be important in predicting work satisfaction,masterand
nostalgic relationship accomplishment, and even slant for
different interfaces. Michon Edmondson can start to
answer more modern inquiries concerning how to
introduce trusted,socially-important,andtopnotchdata to
clients.
Quan Guo et al [9] introduced about an adapting intense
uniform features for cross-media social data byusingcross
auto encoders. To handle learning models to address issue
handle the cross-strategy associations in cross-media
social segments. Quan Guo propose CAE to learn uniform
strategy invariant features, and Jia propose AT and PT
stages to utilize immense crossmedia data tests and set up
the CAE. Adapting effective uniform features for cross-
media social data by using cross auto encoders take an
extra time.
Sepandar D. Kamvar [10] haveintroducedanstudiesabout
any person feel fine and searching the emotional web.Feel
fine to suggest a class of visualizations which is called as
Experiential Data Visualization. The focus is on immersive
item-level interaction with data. The implications of such
visualizations for crowdsourcing qualitative research in
the social sciences. Repeated information in relevant
answers requires the user to browse through a huge
number of answers in order to actuallyobtaininformation.
To date, most research in assessment examination has
been engaged on calculations to extricate, order, and
condense conclusion.
3. METHODOLOGY USED IN PROPOSED SYSTEM
1. Methodology
• Natural Language Processing focuses on the field
of study of interactions between human language
and computers. Natural Language Processing
contain different techniques like:
• Sentiment Analysis: process of determining
whether a piece of writing is positive, negative or
neutral, deriving the opinion or attribute of a
user.It is also known as opinion mining. This
technique is used for discovering how people feel
about particular topic. Natural Language
Processing for sentiment analysis focused on
emotions is extremely useful.

• Topic Extraction: one of the most important
tasks when working with text is Extracting topic.
In this technique, clustering about a similar topic
occur in a collection of a documents or an
information, from this we get more accurate
information. Readers get benefit from topic
keywords as they can judge morequicklywhether
the text is worth reading or not. Website creators
benefit from topic keywords because they can
group similar content by its topics.
• Part-Of-Speech Tagging: It is a piece of software
that reads text in some language andassigns parts
of speech to each word based on both its
definition and its context—i.e., its relationship
with adjacent and related words in
a phrase, sentence, or paragraph. A simplified
form of this is commonly taught to school-age
children, in the identification of words
as nouns, verbs, adjectives, adverbs, etc. Many
words, especially common ones, can serve as
multiple parts of speech. For example, "book" can
be a noun ("the book on the table") or verb ("to
book a flight"); "set" can be
a noun, verb or adjective; and "out" can be any of
at least five different parts of speech.
• Stemming: Stemming is the process of reducing
inflected words to their word stem, base
or root form generally a written word form. The
stem need not be identical to the morphological
root of the word; it is usually sufficient that
related words map to the same stem, even if this
stem is not in itself a valid root. for example,
should identify the string "cats" (and possibly
"catlike", "catty" etc.) as based on the root "cat",
and "stems", "stemmer", "stemming", "stemmed"
as based on "stem".Astemmingalgorithm reduces
the words "fishing", "fished", and "fisher" to the
root word.
2. Support Vector Machine Algorithm
In machine learning, support vector machines (SVMs,
likewise support vector machines systems) are
administered learning models with related learning
calculations that examine information utilized for order
and relapse investigation. Given an arrangement of
preparing cases, each set apart as having a place with
either of two classes, a SVM preparing calculation
fabricates a model that doles out new cases to one class or
the other, making it a non probabilistic two fold straight
classifier (in spite of the fact that strategies, for example,
Platt scaling exist to utilize SVM in a probabilistic
arrangement setting). Utilizing this calculation we can
characterized the positive or a negative post .after
characterization we predict user are in stressed or not.
Input:-User facebook post
Algorithm Steps:
Step1: SVMs augment the edge around the separating
hyperplane. Assume linear separability for now: in 2
dimensions, can separate by a line in higher dimensions,
need hyperplanes Can findseparatinghyperplane bylinear
programming (e.g. perceptron): separator can be
expressed as ax + by = c
Step2: The decision function is fully specified by a subset
of training samples, the support vectors.
Step3: Quadratic programming problem
Step4: Text classification method For example, A
combinationof these 0s and 1s in the feature vector along
with the known label will be the Training input to our SVM
classifier. It should be noted that the label in the feature
vector should be numeric only ortheSVMclassifier.Finally
we can get 0 for positive, 1 for negative and 2 for neutral
labels.
Output:-Classified user stress positive post or negative
post
3. KNN (K Nearest Neighbours) algorithm
In design acknowledgment KNN is a non-parametric
method used for classification and regression. In both
cases, the input consists of the k closest training examples
in the feature space. The output depends on whetherk-NN
is used for classification or regression. Using this KNN
algorithm we can recommendation of hospital to user on a
map also show shorted distance from a current location to
that hospital on goggle map.I also recommendation of
precaution according to level of user stress.
Step1 :-Find k most similar users (KNN).
Step2:-Identify set of items, C, Visited by the group of user
together with their frequency.
Step3:-Recommend the top N- most frequent items in C
that the active user visited or not.
4. A Convolutional Neural Network (CNN)
A Convolutional Neural Network (CNN) is contained at
least one convolutional layers (frequently with a
subsampling step) and after that took after by at least one
completely associated layers as in a standard multilayer
neural system. The engineering of a CNN is intended to
exploit the 2D structure of an information picture(orother
2D information, for example, a discourse flag). This is
accomplished with nearby associations and tied weights
took after by some type of pooling which brings about
interpretation invariant highlights. Another advantage of
CNNs is that they are less demanding to prepare and have

numerous less parameters than completely associated
systems with a similar number of concealed units.
Input: - User facebook post.
Output:-Extraction of topic.
In a proposed system architecture we can detect user are
in mental disorders or not due to interaction social
network. In a social network contain facebook,twitter.on a
facebook user are interact with other people. User can
different posts on a facebook. There are three types of
information that we can use as the initial inputs, i.e,
facebook-level attributes, user-level posting behaviour
attributes, and user-level social interaction attributes. I
first define a set of stress-related textual, visual, and social
attributes from various aspects in social network mental
disorders (SNMDs), Facebook-level attributes describe
the linguistic i.e. positive and negative words and visual
content like brightness, cool color, dull color, as well as
social attention factors (being liked, commented,) of a
single facebook post. Userlevel posting behaviorattributes
as summarized from a user’s montly facebook postings,
post time, post type; social interaction attributesextracted
from a user’s social interactions with friends.
In a proposed system architecture we can detect user are
in mental disorders or not due to interaction social
network. In a social network contain facebook,twitter.on a
facebook user are interact with other people. User can
different posts on a facebook. There are three types of
information that we can use as the initial inputs, i.e,
facebook-level attributes, user-level posting behaviour
attributes, and user-level social interaction attributes. I
first define a set of stress-related textual, visual, and social
attributes from various aspects in social network mental
disorders (SNMDs), User level posting behaviorattributes
as summarized from a user’s montly facebook postings,
post time ,post type; social interaction attributesextracted
from a user’s social interactions with friends. In particular,
the social interaction attributes canfurther be brokeninto:
(i) social interaction content attributes extracted from the
content of users’ social interactionswithfriendslikewords
and emotions; and (ii) social interaction structure
attributes extracted from the structures of users’ social
interactions with friends. On this user input post we can
fetch user level facebook post features On that input of
facebook post .Conventional neural network(CNN )* is
used for topic extraction. Using CNN we can sentiment
analysis of facebook post after Formation of topic Using
Transductive Support Vector Method(TSVM) we can
classified user are in stress or not. After classification user
are in stress or not k-nearest neighbours algorithm (KNN)
is used for recommendation hospital on a map as well as
Admin can send mail of precaution list for user for
become healthy and happy in life.
Fig.Proposed System Architecture
In proposed system experimental setup,weidentifiedthatin
proposed system number stressed user and number of non-
stressed. In a following table, 35 user are in stressed and 40
non-stressed user.
Sr.No No. Stressed
User
No. Non-
Stressed
User
1 35 45
Table1: Stressed and Non-Stressed User
In proposed system experimental setup, we identified total
75 posts from Facebook in that in proposed 25 post are
positive 15 post are negative and 35 post are neutral as
given follow table 2.
Social Network Stress Detection
System
User
Facebook
Post
Facebook-level
Attributes
User
Level
Facebook
Post
Features
User level Statistic
Attribute
User-level posting
Behaviour attributes
TSVM to
Classification of
Users are in stress
CNN is used for
Topic extraction
KNN
Recommendation
Based on location and
User’s disorders level
Users are in
Stress or not
Hospital on Map
Precaution on
mail

Sr.No No.
Positive
Post
No.
Negative
Post
No.Neutral
Post
1 25 15 35
Table2: Number of Facebook Posts
From above table, in proposed system, following graph
shows The stressed and non-stressed user in the graph; we
see 35 users are in stressed and 45 users in the non-stressed
user. In graph 1 shows number of stress user in the graph 1.
Graph 1: Stressed and Non-Stressed User
From below graph 2, in proposed system, number of post
from facebook where In proposed system above graph we
identified total 75 posts from Facebook in that in proposed
25 post are positive 15 post are negative and 35 post are
neutral
4. CONCLUSION
Mental disorders is threatening people’s health. It is non-
trivial to detect mental disorders or stress timely for
proactive care. Therefore we presented a framework for
detecting users’ psychological stress states from users’
montly social media data, leveraging facebook post’content
as well as users’ social interactions. Employing real-world
social media data as the basis, we studied the correlation
between user’ psychological stress states and their social
interaction behaviors. I recommended the user for health
consultant or doctor. I can show the hospitals for further
treatment on a graph which locate shortest path from
current location user to that hospital.
I recommended the user for health precaution send on mail
for user interaction purpose. .
5. REFERENCES
[1] H. Lin, J. Jia, Q. Guo, Y. Xue, J. Huang, L. Cai, and L. Feng.
Psychological stress detection from cross-media microblog
data using deep sparse neural network. In proceedings of
IEEE International Conference on Multimedia & Expo, 2017.
[2] Liqiang Nie, Yi-Liang Zhao, Mohammad Akbari, Jialie
Shen, and Tat-Seng Chua. Bridging the vocabulary gap
between health seekers and healthcare knowledge 2016
[3] Chi Wang, Jie Tang, Jimeng Sun, and Jiawei Han Dynamic
social influence analysis through time-dependent factor
graphsIEEE 2014
[4] Lexing Xie and Xuming He. Picture tags and world
knowledge: learning tag relations from visual semantic
sources 2015
[5] Yuan Zhang, Jie Tang, Jimeng Sun, YiranChen,and Jinghai
Rao.Moodcast: Emotion prediction via dynamic continuous
factor graph model 2016 IEEE International Conference on
Data Mining
[6]Andrey Bogomolov, Bruno Lepri, Michela Ferron, Fabio
Pianesi, and Alex Pentland. Daily stress recognition from
mobile phone data, weather conditions andindividual traits.
In ACM International Conference on Multimedia,pages477–
486, 2014.
[7] Dan C Ciresan, Ueli Meier, Jonathan Masci, Luca Maria
Gambardella,and J ¨ urgen Schmidhuber. Flexible, high
performanceconvolutional neural networks for image
classification. In Proceedings of International Joint
Conference on Artificial Intelligence, pages 1237–1242,
2011.
[8] Jennifer Golbeck, Cristina Robles, Michon Edmondson,
and Karen Turner. Predicting personality from twitter. In
Passat/socialcom 2011, Privacy, Security, Risk and Trust,
pages 149–156, 2013
[9] Quan Guo, Jia Jia, Guangyao Shen, Lei Zhang, Lianhong
Cai, and Zhang Yi. Learning robust uniform features for
cross-media social data by using cross autoencoders.
Knowledge Based System, 102:64– 75, 2016.
[10]Sepandar D. Kamvar. We feel fine and searching the
emotional web. In In Proceedings of WSDM, pages 117–16,
2013

IRJET- Social Network Mental Disorders Detection Via Online Social Media Mining

More Related Content

What's hot (19)

Similar to IRJET- Social Network Mental Disorders Detection Via Online Social Media Mining (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Social Network Mental Disorders Detection Via Online Social Media Mining