International Journal of Computational Engineering Research(IJCER)

International Journal of Computational Engineering Research||Vol, 03||Issue, 7||
||Issn 2250-3005 || ||July||2013|| Page 65
An Integrated Approach for Plagiarism Detection System
Shilpa1
, Mr. Manoj Challa2
1,
CMRIT, BANGALORE,
2
Assoc Professor, Dept of CSE,CMRIT, Bangalore,
I. INTRODUCTION
P lagiarism is defined as the appropriation or imitation of the methods, ideas and views of another
writer and representation of them as one's original work[1]. Plagiarism in scientific publications has increased
and one day you will see your published work is used in another publication yet without proper references and
citation. Scientific publications in the same field normally share the same general information. Besides, each
publication should express a particular message that contributes to that field. Diverse contributions can be made
in diverse areas; for example, solving new problems, giving suggestions to existing problems, experimenting
new methods, comparing current methods, giving enhancement to results. Such these contributions of others are
considered their ideas and should be acknowledged when reported in a further research. The big challenge is to
provide plagiarism detecting with proper method in order to improve the time required to check the results and
percentage of finding results. Several plagiarism detection tools use character or string matching method to
detect the plagiarized content. Most of the current software’s and techniques are less effective in detecting a
plagiarized text because these tools tend to compare the suspected text with original text using characters
matching, some with chunks while others by words example MOSS[3],jplag[4]. This leads to rigorous search
which takes a long time in the matching process. The matching algorithms are working depending on the text
lexical structure rather than semantic structure. Therefore it becomes difficult to detect the text paraphrased
semantically. One of the goals of this study is to propose new semantic techniques for plagiarism detection
based on optimized Semantic Role Labeler. The proposed method does not analyze the content of a text
document as text syntax only, but also captures the underlying semantic meaning in terms of the relationships
among its terms.
In this paper we propose a new plagiarism detection technique based on analytical procedures and
Optimized Semantic Role Labeler. In analytical procedures we highlight the similar sections by giving the
statistical weights. Optimized Semantic Role Labeler can help in detecting sophisticated obfuscation such as
copy paste, anonyms, renaming, paraphrasing, splitting, merging, and changing the sentence from active to
passive voice and vice versa in the highlighted document. The absolute hash function is used to map large data
sets of variable length to smaller data of fixed length by generating hash key.
ABSTRACT:
Nowadays as Internet is becoming the primary media for information access and nearly
every information is available in the Internet. Therefore, it becomes easier to use another author's
contents from the Internet without giving proper reference. Others ideas, solutions or expressions are
representing as one's own original work is known as plagiarism. In this paper, we propose a
framework which works by integrating several analytical procedures. Scholarly documents under
investigation are segmented into logical tree-structured representation using a procedure called
DSEGMENT. Statistical methods are utilized to assign numerical weights to structural components
under a technique called C-WEIGHT. The top weighted components from the structural components
are fed into plagiarism detection technique called Optimized Semantic Role Labeler (OSRL).This
technique analyses and compares text based on the semantic allocation for each term inside the
sentence. The absolute hash function method is used to map large data sets of variable length to
smaller data of fixed length by generating hash key .In terms of Recall and Precision, this method
outperforms the results with the existing plagiarism detection methods.
KEYWORDS: Plagiarism detection, Scientific Publication, Semantic Argument, Semantic Role,
Semantic Similarity, Natural language processing, hash map.

An Integrated Approach For Plagiarism…
||Issn 2250-3005 || ||July||2013|| Page 66
By using the absolute hash function we can reduce the retrieving time in the list. The remaining section
of the paper is structured as follows; Section II discusses the related literature survey on plagiarism detection.
Section III discusses the framework for plagiarism detection using OSRL. Section IV explores the experimental
analysis. In section V, we discuss the results and the potential market of OSRL, and we draw a conclusion and
future of this work in the last section.
II. RELATED WORK
The above section gives a overview of the existing plagiarism methods. Many tools works on finger
print based method, String matching method. Fingerprint method [5] [9] works by creating fingerprint for each
document in the collection. It gives the detailed information about the number of terms per line, number of
keywords, and number of unique terms. Mostly rely on the use of K- grams (Manuel et al. 2006) because the
process of fingerprinting divides the document into grams of certain length k.Different plagiarism tools have
been surveyed in multiple works. Many online solutions are established today; for example,
CrossCheck,Turnitin,SafeAssign,EVE2, WCopyFind ,Viber, Scriptum, PlagiarismDetect, SCAM ,CHECK ,
PPChecker , SNITCH , Ferret and others. Several studies have reported that using plagiarism checker tools in
academia is effective in reducing the problem, and discouraging the students to commit plagiarism . Besides,
anti-plagiarism tools help to educate students and authors in different disciplines about plagiarism .
JPlag [4], Sherlock are the examples of String Matching- based algorithm. Many String Matching
algorithm works based on tokenization, where each file is replaced by predefined tokens. One of the problem of
JPlag[4] is that files must be parse to be included in the comparison for plagiarism because of this some similar
files that are not parsed to be missed. Unlike JPlag, in Sherlock files do not have to parse to be included in the
comparison and also displays quick visualization. The time complexity and space complexity is more in the
above method and also not checking at sentence level. DANIEL R WHITE and MIKE S JOY[6 ] proposed a
method by comparing suspected documents at sentence level there by detects paraphrasing, reordering, merging
,splitting but fails to analyze semantically and also speed is less compared to other software like CopyCatch.
Figure 1: Existing plagiarism detection framework Iplag
Iplag[7] gives a accurate judgment about type of plagiarism. It works by combining several analytical
procedures. Documents under investigation are segmented into D segment [7].Then weights are given to D-
segment called as C-weight. Relevance Ranking(R-Ranking) and plagiarism Screening approach [7] (P-screen)
is used. The architecture for Iplag which is useful for existing plagiarism detection systems is shown in

||Issn 2250-3005 || ||July||2013|| Page 67
figure1.This Iplag acts as a framework and can be applied to any of the existing plagiarism detection
technique.Plagiarism Detection Scheme Based on Semantic Role Labeling [2] analyses and compares text based
on the semantic allocation for each term inside the sentence. It identifies and label arguments in a text. The idea
behind SRL is that the sentence level semantic analysis of text determines the object and subject of a text. For
calculating the semantic labeling, to retrieve semantic arguments and labels in large document will take large
amount of time and space. The architecture as shown below
Figure 2: Structure of Semantic Role Labeler
III. FRAMEWORK FOR PLAGIARISM DETECTION USING OSRL
This paper proposes a framework that is very accurate and reliable in reporting plagiarism in scientific
publications. Checking plagiarism manually requires an Expert. An Expert must have idea about the
contributions in different parts of paper like discussions, conclusions and any suspected methods, techniques,
ideas that might be taken from somewhere else. The experts may not pay attention to every statement but only
sentences that convey original ideas and ignores important parts. The OSRL is used for sentence level semantic
analysis of the text that determines the object and subject of a text. It relays on the characterization of events
such as determining “when”, “where “, how: who” did,”what” to “whom”. The predicate of a clause (normally a
verb) establishes “what “took place and other parts of the sentence express the other arguments of the sentence.
The main goal of OSRL is to identify what semantic relation holds among a predicate and its associate
participants with relations drawn from a predefined list of possible semantic roles for that predicate.
3.1.Document Segmentation [D-segment]
The framework begins by splitting the scientific publications into several meaningful components [7] i.e.
• document->sections->paragraphs
• document->paragraphs->sentences
• document->topics->paragraphs
In this framework, the logical structural organizations of scientific publications were used in which it
includes most of the following categories: Title, Owner, Abstract, Introduction, Literature survey (related
previous works, etc.), Evaluation, Acknowledgements, and References. These categories are called generic
classes. Further, generic classes involve different structural components such as head titles, paragraphs, tables,
tablecaptions, equations, figure captions; etc, Structural components are further segmented into sentences.
3.2.Component Weighting: C-WEIGHT
By assigning numerical weights to structural component .we can show how important that component
is to the article. Depth and Spread functions are statistical measures and are useful for detecting results. Spread
of a term t defines the number of structural components that contain the term. Depth of a term t refers to the
frequency of the term in a class (unlike normal term frequency which considers the number of the term’s
occurrences in the whole document).The weight w of a structural component c in a generic class Gc can be
obtained by clubbing Spread and Depth ,that can be expressed as follows:

||Issn 2250-3005 || ||July||2013|| Page 68
(1)
Where spread (t) is the number of components that has t, is the frequency of t in the generic class Gc
which has t, and is the maximum frequency occurs in that class. Select the top weighted components
that are having more weight.
Figure 3 .Presents a Plagiarism Detection Framework procedure
3.3.OSRL for plagiarism detection
In this method, we take top weighted components of the suspected documents for pre-processing steps
.i.e. stop words removal and stemming. Then, OSRL was used to transform the sentences into arguments based
on location for each term in the sentences. The verbs play an important role in every sentence for processing,
and the analysis of the sentences. All the arguments extracted from the text were grouped in the nodes according
to the argument type. Each group was named by the argument name. This step is called Argument Label Group
(ALG). Then, we extracted all the concepts for each term in the argument groups using WordNet thesaurus. This
step is called Semantic Term Annotation (STA).
3.4.Similarity Detection
Create a hash values for the labels created from the OSRL. Similarity Detection between original and
suspected document is done by first extracting the hash values of the suspected document and then check that
with hash values of original documents. If two is matching then start finding similarity of labels with respect to
hash values which are matched.
Plagiarist can be shown through the following example:James gave Rohan the parcel (original sentence).
The parcel was given to Rohan by James (suspected sentence)
By using OSRL the produced arguments are:
James gave Rohan the parcel
Output:
OSRL Charnaik

||Issn 2250-3005 || ||July||2013|| Page 69
James giver
[A0]
(S1 | S |NP (NNP James)|
gave V:give |VP (VBD gave)
Roha
n
entity
given to
[A2]
(NP (NNP Rohan | |
the thing
given[A1
]
(NP (DT the |
parcel (NN parcel | | |
. ( . . ) ) )
Figure 4. Analysis for original sentence using OSRL
The parcel was given to Rohan by James
Output:
OSRL Charnaik
The
thing
given
[A1]
(S1 | S (NP (DT The |
Parcel (NN parcel ) )
Was (VP (AUX was )
Given V:giv
e
(VP ( VBN given )
To
entity
given
to
[A2]
(PP (T0 to)
Rohan
(NP ( NNP Rohan
) ) )
By giver[
A0]
(PP (IN by )
James (NP (NNP James)
) ) ))
. (. . ) ) )
Figure 5. Analysis for suspected sentence using OSRL
Fig 4 and 5 explains how the suspected sentence analyzed using OSRL. The structure of two sentences may
different if synonyms and antonyms or active and passive are used. But they might be same through
semantically. The OSRL captures the arguments for a sentence in spite of changing the places for the labels
inside the sentences. This method of analyzing the sentence supports our proposed method in detecting
plagiarism if comparison is applied based on the arguments of the sentence using OSRL.
3.5.Experimental Design and Dataset
The experiments were performed on 100 suspected documents, each plagiarized from one or more
original documents. In this point, Original and tokenized suspected documents were analyzed by sentence-based
similarity. Sentences in suspected documents were compared with each sentence in the candidate documents
according to the arguments of the sentences. We not only detect the arrangement similarity between sentences,
but also possible semantic similarity between two sentences. Similarity detection was conducted by comparing
the original Topic Signature terms and suspected Topic Signature terms. If the two terms were found to be
identical, we went directly to the argument label groups that contained these terms, and then determined the
label group where they belonged, thus determining the possible sentences that may be plagiarized. This step
compared the arguments of possible sentences that had been plagiarized with the corresponding arguments in
original sentences. The argument label group gave way to the main arguments and each argument inside the
group quickly take to the possible plagiarized sentence. Some parameter play a crucial role in the similarity
calculation, such as the number of matched arguments and number of arguments which exist in the sentences.
The first variable determines the similar arguments between the suspected document and original document

||Issn 2250-3005 || ||July||2013|| Page 70
while the second variable determines the argument that does not exist in the sentences. The similarity between
the arguments of the suspected document and original document was calculated according to Jaccard coefficient
measure [8] which is well-known as famous similarity measure between two sets. Jaccard coefficient [8] defined
as a following equation:
= (2)
Where,
C(Arg )=concepts of the argument sentence in the suspected document; ( Arg ) =concepts of the
argument sentence in the original document;
We then calculated the similarity between the suspected document and original document based on the
following equation:
TABLE 1.
EVALUATION MEASURE OF THE PROPOSED METHOD
Number of
documents Recall Precision F-measure
1000
0.828237 0.655606 0.761109
Total Similarity (Doc1, Doc2) = (3)
Where,
Sim (Arg , Arg ) is similarity between Arguments sentence j in suspected document containing concept
i Arguments sentence k in original document containing concept i , l = no. of concepts, m= no. of
Arguments sentence in suspected document, n = no. of Arguments sentence in original document.
IV. RESULTS AND DISCUSSION
Our technique was tested according to the group of 100 documents .Those suspected documents were
plagiarized with different ways of plagiarism such as simple copy and paste, changing some terms with their
corresponding synonyms, and modifying the structure of the sentences (paraphrasing). We provided three
general testing parameters that are commonly used in plagiarism detection as following
Recall= (4)
Precision = (5)
F-Measure = (6)
The proposed method is evaluated and compared with Latent Semantic Analysis technique. The results are
shown below

||Issn 2250-3005 || ||July||2013|| Page 71
Figure 6. Comparison of two techniques in terms of accuracy.
Figure 7. Comparison of two techniques in terms of time required for plagiarism detection.
Figure 8. Time taken in each modules present in OSRL technique.
Fig. 6 and 7 demonstrates the comparison between OSRL and Latent Semantic Analysis in terms of accuracy
and time required for Plagiarism Detection respectively. Fig 8 demonstrates the time required in each module
which are present in OSRL technique. We found that all the scores that were obtained by our proposed method
have good results than other method. The results from the comparison show that the proposed method achieved
better results in terms of accuracy and time required for Detection.

||Issn 2250-3005 || ||July||2013|| Page 72
V. CONCLUSIONS AND FUTURE WORK
Optimized Semantic role labeling can be used for plagiarism detection by extracting argument of
sentences and comparing the arguments. Tests were carried out against 100 dataset for plagiarism detection. The
proposed methods were found to achieve better performance compared to Latent Semantic Analysis [9]. Our
future work is to improve its efficiency and time complexity by introducing Hadoop Map Reduce Technique. As
of now our detection technique works for mono lingual documents. We can extend it for multi lingual
documents.
REFERENCES
[1] www.academia.edu/689297/Source_Code_Plagiarism_-a_Student_Perspective
[2] Naomie Salim, Ahmed Hamza Osman, “Plagiarism Detection Scheme Based on Semantic Role Labeling”, International
conference march 2012.
[3] Http://theory.stanford.edu/aiken/moss/, 2005.
[4] Lutz Prechelt , Guido Malpohl ,”JPlag: Finding plagiarisms among a set of programs” March , 2000 ,
http://wwwipd.ira.uka.de/EIR/
[5] http://www.dcs.warwick.ac.uk/report/pdfs/cs-rr-440.pdf
[6] DANIEL R. WHITE and MIKE S. JOY,” Sentence-Based Natural Language Plagiarism Detection”,IEEE ACM-
TRANSACTION August 23, 2005
[7] Salha Alzahrani , Naomie Salim , Ajith Abraham,” iPlag: Intelligent Plagiarism Reasoner in Scientific
Publications”,International Conference,2011.
[8] P. Jaccard, "Etude comparative de la distribution florale dans une portion des Alpes et des Jura," Bulletin de la Society Vaudoise
des Sciences Naturelles, vol. 37, pp. 547–579, 1901.
[9] Cosma, Georgina and Joy, Mike. (2012) An approach to source-code plagiarism detection and investigation using latent semantic
analysis. IEEE Transactions on Computers, Vol.61 (No.3). pp. 379-394. ISSN 0018-9340
Mrs. Shilpa is presently doing Master of Technology in Computer science and Engineering at
CMR Institute of Technology, Bangalore, Karnataka. She received her Bachelor of Engineering
degree in Information Science and Engineering from Kalpatharu Institute of Technology, Tiptur,
Karnataka in the year 2007.
Mr. Manoj Challa is pursuing Ph.D(CSE) in S.V.University, Tirupati, India. He completed his
M.E(CSE) from Hindustan College of Engineering, Tamil Nadu in 2003. He is presently working
as Associate Professor, CMR Institute of Technology, Bangalore. He presented nearly 18 papers
in national and international conferences. His research areas include Artificial intelligence and
computer networks.

International Journal of Computational Engineering Research(IJCER)

More Related Content

What's hot (16)

Viewers also liked (16)

Similar to International Journal of Computational Engineering Research(IJCER) (20)

Recently uploaded (20)

International Journal of Computational Engineering Research(IJCER)