A data mining tool for the detection of suicide in social networks

PEOPLE’S DEMOCRATIC REPUBLIC OFALGERIA
MINISTERY OF HIGHER EDUCATION AND SCIENTIFIC RESEARCH
UNIVERSITY OF MOHAMED BOUDIAF - M’SILA
DOMAIN: Mathematics and Computer
Science
FIELD: Computer Science
SUB-FIELD: Information Communication
Technologies
FACULTY: Mathematics and
Computer Science
DEPARTMENT of Computer
Science
A Dissertation in Fulfillment
For the Requirements of the Degree of Master
SUBJECT
Academic year: 2016 /2017
Mr: Y. ARIOUAT University of M’sila Chairman
Mr: TAHAR MEHENNI University of M’sila Supervisor
Mr: M. KAMEL University of M’sila Examiner
By: Yassine Bensaoucha
A data mining tool for the detection of
suicide in social networks
1

Introduction
Suicide
Social Network
Data mining
2
Naive Bayes Classifiers
Proposed classification
Implementation Of The System
Conclusion
plan

Introduction
At present, the suicide phenomenon is raising, having a relevant impact on our
society. each year millions of people die as a result of suicidal, behavior becoming
an economic social and human problem. on the other hand, the use of SN as a means
of communication is becoming extremely popular, and people find that writing about
their feeling and sharing it to the world through this SNs is much easier then talking
about it in real life. the increasing number of publications in SN has made the field
of data mining evolving.
Based on these considerations, we have developed a program that can
detect the suicidal tendencies of people via Twitter.
3

4
Suicide
Suicide is when people direct violence at themselves with the intent to end their
lives, and they die as a result of their actions. Suicide is a leading cause of death
in the United States.
According to the analysis carried out by the World Health Organization
(WHO) in 2012 every 40 seconds a person died by suicide, The WHO estimates
that suicide is the 13th leading cause of death in the world and the third one
between youth aged 15-44

5
The Social Network: is a network structure composed of social
entities (might be people, organisations or groups) as nodes. And
connections in the form of edges, that represent the connection or
relationship amongst these entities. Such connections might be
friendships, or business connection etc.
Social Network
Figure 1: Social Network

Data mining is the process of sorting through large datasets to identify patterns
and establish relationships to solve problems through data analysis.
Data mining
Text Mining
the process at which text is transferred into data that can be analysed. this
incurs the procedures of creating an index for the individual terms, based on the
location of the term within the original text, or based on other techniques or
protocols. The words and indexes can then be used for a variety of analysis
methods.
Sentiment Analysis
the process of determining the emotional tone behind a series of words, used to
gain an understanding of the the attitudes, opinions and emotions expressed
within an online mention.
6

Figure 2 : Machine learning techniques
7

Naive Bayes Classifiers
The Naive Bayes algorithm is a widely used algorithm for document
classification. Given a feature vector table, the algorithm computes the posterior
probability that the document belongs to different classes and assigns it to the
class with the highest posterior probability. There are two commonly used models
( multinomial model and multi-variate Bernoulli model) for using Naive Bayes
approach for text categorization.
8

Figure 3: trainining multinomial naive bayes
Figure 4: test multinomial naive bayes
9

Example multinomial naive Bayes classifier:
Figure 5 Example multinomial naive Bayes classifier
Decide:
whether document d5 belonging to class c=China?
10

Figure 6: resulut of the example multinomial naive Bayes classifier
11

Advantages of Naïve Bayes
Easy to implement
Requires a small amount of training data to estimate the parameters
Good results obtained in most of the cases
12

Data Pre-processing:
The goal behind preprocessing is to
represent each document as a
feature vector, to separate the text
into individual words.
Preprocessing text is called
tokenization or text
normalization.
Figure 7: Data Pre-processing
13

Proposed classification:
Each sentence is first preprocessed and then passed into three categories of
classifiers, each deciding whether the sentence belongs to the
corresponding category or not.
This category contains positive
and regular sentences
Contains negative sentences that do not contain suicidal
phrases like
"the taste of pizza today very bad".
contains suicidal phrases that are classifed as
emergency phrases in the previous search
“Each day, each hour, each minute is just torture. I want it to
end.”
“Waking up every day wishing I hadn’t “
“I feel like the only way to no longer carry this pain is to die.”
“I am weird and slow. Every social interaction is painfully
awkward. “
“I want to finish my life.”
“No one understands me in this life, I'm leaving.” 14
positive negative suicidal

Proposal Solution for Zero problem(document):
This happens in case the probability of each category is 0, that is when
we pass a new text (tweet) entirely on our classification and we do not
have any word from this text in the dataset
first solution: is switching each word from the new tweet (sentence )
with synonyms, then we re-categorize the new tweet (sentence with the
same meaning ) .
second solution:
the solution depends on the extraction of sentiment from the sentence
using the textblob library.
Sentiment(polarity,subjectivity).
The polarity score is a float within the
range [-1.0, 1.0].
The subjectivity is a float within the
range [0.0, 1.0]
if the number of “polarity” is greater than 0 Then positive else negative 15

Implementation
Of The System
download
tweets
pre-process
the tweets
Classification
Positive negative suicidal
16
Analysis of the results of
Classification
The user has
suicidal tendencies
The user has no
suicidal
tendencies

Username of the person
that we want to apply the
calassification on it
Username
Twitter
API
17
download
tweets

18
Analysis of the results of
Classification
Calculate the percentage of positiveTweets = positive
Calculate the percentage of negativeTweets = negative
Calculate the percentage of suicidalTweets = suicidal
elif suicidal > 20%
and negative > 60%
If suicidal > 70% The user has
suicidal
tendencies
else
The user has
no suicidal
tendencies

20
This work aims to create a program, which will be capable of detecting suicide
in social networks, so the results obtained from this study or this program were
acceptable.
While we have identified some interesting and promising results, future work
will be compare the results of this work with other classification. and also
merging our classification with decision tree classification.
CONCLUSION

A data mining tool for the detection of suicide in social networks

Recommended

More Related Content

What's hot (20)

Similar to A data mining tool for the detection of suicide in social networks (20)

Recently uploaded (20)

A data mining tool for the detection of suicide in social networks

Editor's Notes