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

net/publication/336252196

MONITORING OF SUSPICIOUS DISCUSSIONS ON ONLINE FORUMS USING

DATA MINING

Research · January 2018

DOI: 10.13140/RG.2.2.11235.91680

CITATIONS READS

0 2,554

3 authors, including:

Shailesh Dudala
University of Chicago
2 PUBLICATIONS   0 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Senior Year Research View project

All content following this page was uploaded by Shailesh Dudala on 04 October 2019.

The user has requested enhancement of the downloaded file.

International Journal of Pure and Applied Mathematics
Volume 118 No. 22 2018, 257-262
ISSN: 1314-3395 (on-line version)
url: http://acadpubl.eu/hub
Special Issue
MONITORING OF SUSPICIOUS DISCUSSIONS ON ONLINE FORUMS USING DATA MINING
1
Tanya Srivastava, 2R.Mangalagowri, 3Shailesh S.Dudala
2
Assistant Professor, 1,2,3Department of CSE,
SRM University, Chennai, India 603203
1
tanyasriva96@gmail.com, 3shailesh.dudala@hotmail.com
Abstract: With the increasing years, the internet has
changed the lives of so many people for better or worse.
As internet technology is progressing, many illegal
activities have also increased exponentially. The Internet
is an unacknowledged path for illegal activities such as
hacking, trafficking, betting, fraud and scams etc. The
cyber-crime branches are looking for provisions to detect
these forums for illegal feedbacks, comments or reviews
and download questionable postings as verification for
their investigation. Our proposed system will monitor for
suspicious postings, collect it from few discussion
forums, implement techniques of data mining and extract
meaningful data. In this concern, we focus on Data
Mining and Sentimental Analysis to enhance the
techniques and to extract the features of the text to
represent them.
Keywords: Illegal Activities, Discussion forums,
Sentimental Analysis.
1. Introduction
Accelerating crimes on digital mediums alert the law
implementation bodies to continuously monitor online
activities. To achieve the above we need to build a
system which detects suspicious postings on online
forums. A lot of surveys and facts have proved that it is
difficult to manage information which constantly keeps
changing on internet thus data mining is the optimal
choice to analyse and gather data. Using various data
mining techniques, raw data is extracted from a large text
corpus and this raw /unstructured data is transformed into
structured data in pre-processing. This paper highlights
the datamining techniques and sentimental algorithm
which is prototyped and implemented using python
which is functional in natural language using Natural
Language Toolkit (NLTK) library.
A. Existing systems
The digital technology has been impacting human
behaviour for a very long time. The existing system
257
ijpam.eu
analyse text sources from social media and classify the
text into different groups. The system distinguishes
between legal and illegal data using Stop-Words
Selection, Stemmer algorithm and Levenshtein
algorithm. In Stop-word Selection, the commonly used
words in English language, such as, “we”, “he”, “they”
are removed. More such words can be removed using this
algorithm.
The Porter Stemmer algorithm removes the suffixes from
English words and transforms into its root word, for
example:
The words “Stemmed”, “Stemmer”, “Stemming” have
suffixes “ED”, “ER” and “ING” will be removed during
the information retrieval and will leave one word
“STEM” as the root word.
In Levenshtein algorithm, a large set of words are
compared with each other. For example, the Levenshtein
distance between "fitten" and "bitting" is 3, since the
following three edits change one into the other, and there
is no way to do it with fewer than three edits:
fitten → bitten (substitution of "b" for "f")
bitten → bittin (substitution of "i" for "e")
bittin → bitting (insertion of "g" at the end).
Levenshtein distance is a measure of similarity between
two words.
B. Some Limitations of current systems
Although the existing system works fine, but there is still
scope for improvement. The performances of data
retrieval analysis real from online forums remains
debatable due to lack of tools. The system is to
monitor suspicious discussions automatically on an
online forum but the system does not take large amount
of data as input. The system is difficult to Moderate. It
only limits to Spam reviews and feedbacks. Security
vulnerability is also one of the major disadvantage.
International Journal of Pure and Applied Mathematics
2. Literature Survey
A framework has been developed by [4], which is used to
detect emotion on online media. EmoTxt finds the
emotions and categorize based on the input data provided
in a comma separated value (CSV) file format. The
output is in the form of CSV file. The file contains text id
and predicted label for each input data set. The model
classifies the emotions as, joy, sad, and anger etc.
According to researchers [4], the model follows a
tree structured hierarchical classification of emotions,
where latter layers provides an understanding of
emotions of the previous layers. The model includes six
basic emotions, namely love, happy, anger, sad, fear, and
surprise. The data is tested and trained on gold standard
dataset using linear Support Vector Machine (SVM).
A research paper published [1], suggests various
techniques and algorithms which can be employed. The
paper elaborates about Stop-word Selection, Stemming
algorithm, Brute-force algorithm, Learning Based
algorithm and Matching algorithm. Matching algorithms
use two constraints Stemmer Strength and Index
Compression. Using these two constraints, stem words in
database are compared and their value is calculated.
Learning based algorithms include machine learning
theories like SVM and conditional random field.
Another paper [2], describes the system will analyse
data from few discussion forums and will classify the
data into different groups i.e. legal and illegal data using
Levenshtein algorithm. Levenshtein is used to measure
similarity between two words.
In a paper [5], research is being carried out using web
mining. Using Web mining, the data set is collected by
crawling large number of web pages. It requires a user
interactive query interface intended for predicting crime
hotspots from various web pages. The main techniques
used are classification, sequential pattern mining,
association analysis, outlier analysis and cluster analysis.
Clustering and classification techniques, identify the
similar items and group them in classes. The association
rule mining and sequential pattern mining techniques are
similar. They both identify frequently occurring sets and
extract a pattern. Using all these techniques in web
mining make it more complex. Along with the
techniques, a conceptual network i.e. a dynamic structure
of nodes connecting in a functional way is required for
better visualisation of criminal networks and to reveal the
vulnerabilities inside the network. The biggest challenge
faced by the researchers was collecting the data from the
web pages which consists of hyperlinks, navigation links,
advertisements, privacy policies etc. Theses noises
should be remove d from the data before processing.
Another challenge was that on web the information is
never constant. The model intends to concentrate on
258
Special Issue
efficiency by using multiple processes, threads and
asynchronous resources.
In another research done on sentimental analysis and
opinion mining, [8] Rudy Prabowo1and Mike Thelwall
has presented a method which uses different
classification methods and hybrid classification on
multiple classifiers in their research. They showed that
hybrid classification on multiple classifiers can not only
improve the performance but can also increase the
effectiveness. The paper presents results by comparative
study using various automatic classification methods,
machine learning classification with hybrid classification.
3. Proposed Method
Data mining can be used to monitor social media as well
as discussion forums for suspicious feedbacks or
comments. Discussion forums can be used to spread any
message to a large population almost instantly. Millions
of people share their views and ideas on politics, religion
and there are also people who intentionally hurt religious
or racial sentiments through malicious posts. Hence it
becomes important to monitor the posts on these forums.
This application collects the postings and comments from
the discussion sites and analyses those comments using
data mining techniques and algorithm.
The collected data will be analysed for provoking posts
using a set of keywords in the algorithm. Further the set
of sensitive keywords are divided into 6 categories: -
hacking, sexuality, religious, piracy, gambling, fraud. If
the comments in text corpus comes across any of
keywords related to any of the 6 categories, then it is
classified into that particular category to which the
keyword belongs. Our goal is to achieve sentiment
analysis for data provided from discussion forums for
which we will build Classifiers which consists of
different machine learning classifiers.
International Journal of Pure and Applied Mathematics
Figure 3.1 Architecture Diagram
4. Conclusion
In this paper, there is a detailed explanation of various
existing types of methods to detect suspicious activities
of users in online forums. Basically, we found that the
techniques developed in this domain, focuses on data
mining in order to reduce the suspicious behaviour
possessed by the user on web. Study revealed that we
need to analyse the user’s behaviour based on feedbacks,
comments, reviews shared by them. Also, suspicious
behaviours can be categorized under groups such as
terrorist activity, financial laundering, hacking, sexual or
racial harassment etc. Using this categorization, the
corpus of probable suspicious words can be build which
will further assist in developing more refined and reliable
techniques for detecting these activities.
References
[1] Murugesan, M. Sururthi, R. Pavitha Devi, S.
Deepthi, V. Shri Lavanya, and Annie Princy. Automated
Monitoring Suspicious Discussions on Online Formus
Using Data Mining Statistical Corpus Based Approach.
259
Special Issue
Imperial Journal of Interdisiplinary Research (IJIR) Vol-
2, Issue-5, 2016
[2] Harika Upgaganlawar, Nilesh Sambhe.
Surveillance of Suspicious Discussions on Online
Forums Using Text Mining. International Journal of
Advances in Electronics and Computer Science, Volume-
4, Issue-4, April-2017
[3] Suhas Pandhe and Sahil Pawar. Algorithm to
Monitor Suspicious on Social Networking Sites Using
Data Mining Techniques. International Journal of
Computer Applications. Volume 116 – No. 12, April
2015
[4] Javad Hosseinkhani, Mohammad Koochakazei,
Solmaaz Keikhaee and Yahaya Hamedi Amin. Detecting
Suspicion Information on Web Crime Using Crime Data
Mining Techniques. International Journal of Advanced
Computer Science and Information Technology
(IJACSIT) Vol.-3, No. 1, 2014, Page 32-41
[5] G.Vinodhini, R.M Chandrasekran. Sentiment
Analysis and Opinion Mining: A Survey. International
Journal of Advanced Research in Computer Science and
Software Engineering. Volume 2, Issue 6, June-2012
[6] Fabio Calefato, Filippo Lanubile, Nicole Novielli.
EmoTxt: A Toolkit for Emotion Recognition from Text,
University of Bari Aldo Moro
[7] M.F Portar. An Algorithm for Suffix Stripping
Program. Vol. 14 Issue: 3, pp. 130-137
[8] B. Connor, R. Balasubramanyan, B.R Routledge,
and N.A Smith. From tweets to polls: Linking text
sentiment to public opinion time series. School of
Computer Science Carnegie Mellon University
[9] Rudy Prabowo1and Mike Thel wall. Sentiment
Analysis: A Combined Approach.
[10] S.V.Manikanthan and K.Baskaran “Low Cost
VLSI Design Implementation of Sorting Network for
ACSFD in Wireless Sensor Network”, CiiT International
Journal of Programmable Device Circuits and
Systems,Print: ISSN 0974 – 973X & Online: ISSN 0974
– 9624, Issue : November 2011, PDCS112011008.
[11] T.Padmapriya, Ms. N. Dhivya,Ms U.
Udhayamathi, “Minimizing Communication Cost In
Wireless Sensor Networks To Avoid Packet
Retransmission”, International Innovative Research
Journal of Engineering and Technology, Vol. 2, Special
Issue, pp. 38-42.
International Journal of Pure and Applied Mathematics Special Issue

Table 2.1 Literature Survey

Work Done Algorithms Used Result Obtained Conclusion

The purpose of this This system is to develop

[1] Automated ▪ Stop word selection ▪ system is an
Monitoring of monito automatic system for
Suspicious Discussions ▪ Stemming Algorithm to r suspicious detecting
discussions automatically
Using Data Mining ▪ Brute-Force Algorithms on illegal discussions on the
Suffix-Stripping online forums, through
Statistical Corpus-Based ▪ Algorithms online forum. which
Approach ▪ ▪ To detect suspicious post
Matching Algorithms text we can discover illegal
▪ Learning-based Methods analysis is used. activities of all users
This system also focuses
▪ Emotional Algorithms ▪ to
Keyword Spotting reduce execution time,
▪ Technique easier
▪ Learning-based Methods classification to identify
more relevant discussions.

[2] Surveillance of Text mining is used to

Suspicious ▪ Stop word selection ▪ detect This model is presented to
detect suspicious activity
Forums Using Text ▪ suspicious posts in online on
Mining Stemming algorithm
forums. smaller data sets using text
▪ Levenshtein algorithm Fails to detect the
▪ suspicious mining.
activity in larger data sets.

Using ID3. an accuracy

[3] Algorithm to Monitor Data Analysis Technique ▪ of • ID3 decision tree
is achieved which
▪ 86% is algorithm forms a
Suspicious Activity on Cleaning and Integration: comparativel
Selection and y higher than decision tree which will
Social Networking Sites ▪ transformation classificatio
other n monitor the derogatory
using Data Mining ▪ Data Mining algorithms and neural comments on social
▪ networks. networking sites.
Techniques Pattern Evaluation and 14% error
▪ The is due to• The accuracy is 86%
Presentation
unstructured data which using ID3 algorithm.
consists of grammar
mistakes
• ID3 DECISION TREE and abusive comments.

260
International Journal of Pure and Applied Mathematics Special Issue

Using the traditional data Although, the model uses

[4] Detecting Suspicion  Clustering
Information on The Web  Association rule mining
Using Crime Data Mining  Deviation detection
Techniques  Classification
 String comparator
techniques
 Sequential Pattern Mining
mining the
techniques on web mining
can traditional data mining
solve the problem of web techniques, the efficiency
crime. can
increase by using multiple
processes and optimal
utilization of asynchronous
resources.

[5] Sentiment Analysis We can use a Sentiment The hybrid classification

and  Rule Based Classification Analysis on
Opinion Mining: A General Inquirer
Survey  Based Tool (SAT), that can apply a multiple classifiers can not
semi-automatic, only improve the
Classifier (GIBC) complementary performance
approach, with in each
 Rule-Based Classifier classifier but can also increase the
(RBC) that contributes to other effectiveness.
 Statistics Based classifiers
In achieving effectiveness.
Classifier (SBC)
 Induction Rule Based
Classifier (IRBC)
 Support Vector Machines
 Hybrid Classification

261
 262

View publication stats