KR20070041918A - Design and implementation of a text plagiarism detection method using omucs and sequence alignment technique - Google Patents

Abstract

The present invention relates to a text plagiarism search system for checking whether plagiarism of text and plagiarism checking method, comprising: an original document sentence classification step of classifying an input original document in sentence units; An original document word classification step of classifying sentences classified through the original document sentence classification step in word units; A comparative document sentence classification step of classifying the input comparative document into sentence units; A comparison document word classification step of classifying sentences classified by the comparison document sentence classification step in word units; Comparing the original sentences classified in word units and the comparative text sentences through the original document word classification step and the comparative original document word classification step, and identifying the same words in the original sentence and the comparison text sentence compared to each other; An OMUCS calculation step of applying the same word identified in the same word checking step to an OMUCS [original text] [comparative text] with a modified cosine similarity; And comparing the result output through the OMUCS operation step with a first threshold value, and determining whether or not the similarity of the sentence is determined according to the similarity between the original sentence and the comparative text.

Description

Design and implementation of a text plagiarism detection method using OMUCS and sequence alignment technique}

1 is a diagram and a formula of the geometric configuration of the vector model applied to the text plagiarism check method according to the present invention,

2 is a diagram illustrating an example for explaining a plagiarism determination method performed in the text plagiarism check method according to the present invention,

3 is a block diagram illustrating a text plagiarism search system according to the present invention;

4 is a flowchart sequentially showing a method for checking text plagiarism according to the present invention;

5 is a diagram illustrating a document analysis process of a document analysis module constituting a text plagiarism search system according to the present invention;

6 and 7 is a table listing the original English text and the comparative English text that is an example applied for the implementation of the text plagiarism check method according to the present invention,

8 is a graph showing the results of applying the original English text and the comparative English text described in FIGS. 6 and 7 to the conventional cosine similarity and OMUCS according to the present invention, respectively.

9 is a graph showing the results of applying the original English text and the comparative English text described in FIGS. 6 and 7 in the sequence alignment process according to the present invention.

The present invention relates to a text plagiarism search system for confirming whether plagiarism of text and plagiarism checking method thereof.

Technological advances in areas such as computer storage, processors, networks, database systems, scanning systems, and user interfaces have made it possible to be digital libraries. In addition, technological advances in the computer field and the emergence of the Internet have made it possible to easily upload and share information on the World Wide Web server through web browsers such as Internet Explorer and Netscape. This has brought about many changes in culture and culture.

However, the popularity of the Internet is not only positive but also negatively affecting society.

Information that exists online such as electronic libraries and the web exists in the form of digital documents. Increasing access to information in the form of electronic documents increases the threat of piracy and distribution, and has created a complex problem of intellectual property infringement. These problems are aggravated by the development of search sites such as google, altavista, yahoo, naver, and empas.

Also, Hangul and Word, the text editor programs, are widely used as easy editing tools for those who want to plagiarize documents.

The development of search sites and the development of editing tools can be a good tool for anyone who wants to be informed quickly and easily, but it can also be a tool for infringing intellectual property. On the other hand, document plagiarism is made easier due to the convenience of the text editor that can quickly and easily handle the font, font size, and paragraph structure of the document.

As a result, the convenience of retrieval and document editing further developed the form of plagiarism for documents and made it more difficult to determine whether they were plagiarizing documents.

There are two approaches to preventing such document plagiarism: copy prevention and copy detection. Copy protection includes storing on a medium such as a physically independent CD to prevent plagiarism of the document, authorizing access to information on the document, and encrypting the document with a program.

However, these methods require a great deal of time and distribution, do not allow access and distribution of documents, and cannot detect plagiarism for plagiarized documents.

The copy search is to check whether the document is plagiarized. However, the conventional copy confirmation method is weak in searching for similarity of documents and copying of partial syntax. In the case of a system using a hash function, a hash value collision occurs. In addition, copy searching was not possible for documents whose syntax characteristics were unknown, and plagiarism of various sizes could not be detected.

Accordingly, the present invention has been made to solve the above problems, the technical problem is to provide a text plagiarism search system and a text plagiarism check method that can increase the reliability of determining whether the document plagiarism in the copy search. .

The present invention to achieve the above technical problem,

An original document sentence classification step of classifying the input original document in sentence units;

An original document word classification step of classifying sentences classified through the original document sentence classification step in word units;

A comparative document sentence classification step of classifying the input comparative document into sentence units;

A comparison document word classification step of classifying sentences classified by the comparison document sentence classification step in word units;

Comparing the original sentences classified in word units and the comparative text sentences through the original document word classification step and the comparative original document word classification step, and identifying the same words in the original sentence and the comparison text sentence compared to each other;

An OMUCS calculation step of applying the same word identified in the same word checking step to an OMUCS [original text] [comparative text] with a modified cosine similarity; And

A similarity determination step of a sentence comparing the result output through the OMUCS operation step with a first threshold value to determine similarity according to the similarity between the original sentence and the comparison sentence;

Text plagiarism check including.

In the text plagiarism check method for achieving the above technical problem,

The method further includes a sequence sorting step of obtaining a maximum value by applying at least one selected from a global sorting method, a local sorting method, and a semi-global sorting method to the original text and the comparative text that are identified as similar through the similarity determining step. It is.

In the text plagiarism check method for achieving the above technical problem,

Comparing the sum of the maximum values calculated through the sequence alignment step with the second threshold value, and further comprising a plagiarism determination step of determining whether the plagiarism according to the similarity between the original sentence and the comparative sentence.

The present invention to achieve the above technical problem,

An original document sentence classification step of classifying the input original document in sentence units;

An original document word classification step of classifying sentences classified through the original document sentence classification step in word units;

A comparative document sentence classification step of classifying the input comparative document into sentence units;

A comparison document word classification step of classifying sentences classified by the comparison document sentence classification step in word units;

The original document classified in word units and the comparative text, respectively, through the word classification step of the original document and the word classification step of the comparative document, respectively, and applied to a COS similar to the cosine [original text] [comparative text]. Operation step;

A similarity determination step of determining whether or not the similarity is based on the similarity between the original sentence and the comparative text sentence, by comparing the result output through the COS operation step with a third threshold value; And

A sequence sorting step of obtaining a maximum value by applying one or more methods selected from a global sorting method, a local sorting method, and a semi-global sorting method to the original text and the comparative text that are identified as similar through the similarity determining step;

Text plagiarism check including.

In the text plagiarism check method for achieving the above technical problem,

Comparing the sum of the maximum values calculated by the sequence alignment step with the fourth threshold value, and further comprises a plagiarism determination step of determining whether the plagiarism according to the similarity between the original sentence and the comparative sentence.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

In order to detect the plagiarism of documents, the selection of units to compare between documents is the most important. The elements of a document may be referred to as letters, symbols, etc., but they are not a unit that can express the meaning of the document, and the word is the minimum unit of meaning.

Therefore, the comparison unit of the text plagiarism search system and the text plagiarism checking method according to the present invention is a word.

Overlap Measure Using Cosine Similarity (OMUCS) is a measure of similarity between documents using cosine similarity, which is a method of measuring similarity between documents using a vector model that is used most frequently in the field of information retrieval to evaluate the degree of copying between documents. In this method, the OMUCS of the present invention is applied without considering the order of words, so that the text plagiarism searching system and the text plagiarism checking method according to the present invention are used. Sequence sorting techniques are further applied.

As described above, the OMUCS uses cosine similarity to measure the degree of overlap between the original sentence and the sentence to be compared by expressing each sentence in the document as a vector model.

The vector model applied to the text plagiarism search system and the text plagiarism check method according to the present invention is shown in FIG. 1 (the diagram and the equation of the geometrical configuration of the vector model applied to the text plagiarism search system and the text plagiarism check method). Where j and k are terms (words), S is a sentence, and W is a weight.

The weight Wij of the term (word) and sentence pair (Sj, Sk) is a positive non-binary value, and the vector Sj of the sentence Sj (the vector cannot be displayed on the document to distinguish between the 'sentence' and 'vector' The value is darkened and underlined.) Is expressed as (W1j, W2j, ..., Wtj). Where t is the total number of index words in the system.

The sentence Sj and the sentence Sk are represented by t-dimensional vectors, which are the total number of index words. The similarity measure between sentence Sj and sentence Sk in the vector model can be calculated as the correlation between two vectors Sj and Sk , which is the cosine of the angle between the two vectors as shown in FIG. It can be quantified as in the formula.

Where | Sj | and | Sk | is the norm of two sentences. Sj | and | Sk | provides normalization of sentence space.

Since Wij and Wik have a value greater than or equal to 0, sim (Sj, Sk) is a value between 0 and 1. Therefore, the vector model may assign the degree of similarity according to the similarity value between sentences, rather than predicting only the relation between sentences. This is based on the sim (Sj, Sk) value, which is used as a basis for determining whether the sentence is plagiarized in the document plagiarism search.

The occurrence vector for the list of words in the sentence S is defined as O (S), and the frequency vector is defined as F (S). Fi (S) is the frequency with which the words j and k in the sentence S occur. sim (S1, S2) is the similarity between sentences S1 and S2 and is calculated by applying the cosine similarity measure.

For example, r is the original sentence, Sj is the new sentence to be compared with r, and O (r) = {a, b, c}, O (S1) = {a, b}, O (S2) = {a, b, c}, O (S3) = {a, b, c, d, e, f, g, h} (a, b, c, d, e, f, g, h Word), and the copy search results show that r and S1 are very similar, and that r, S2, and S3 are exact copies, but applying the conventional SCAM method, Equation 1 The result will be.

Analyzing the results, the sim (r, S1) and sim (r, S2) values are quite satisfactory, so r, S1, and S2 can be judged as plagiarism. All of them have the result of insufficient values. That is, since the original sentence r has the words a, b, and c, and the sentence S3 has a, b, c, d, e, f, g, and h, the sentence S3 contains the original sentence, There is a problem with this 0.61.

This is because the sentence S3 has the words d, e, f, g, and h, which do not matter even though they have all the words a, b, and c of the original sentence r.

2 is a view for explaining a plagiarism determination method performed in the text plagiarism search system and the text plagiarism check method according to the present invention, will be described with reference to this.

Fig. 2 (a) shows the relation between r and S1 as a set, where r∩S1 = {a, b}. Therefore, O (r) = {a, b, c} and O (r∩S1} = {a, b}, and the cosine similarity is measured by substituting this in Fig. 1 (b). Become together.

As a result, the result of r and S1 is 'significantly similar'.

Fig. 2 (b) is a diagram representing the relationship between r and S2 as a set. r∩S2 = {a, b, c}. Therefore, O (r) = {a, b, c} and O (r∩S2) = {a, b, c}. Substituting this in Figure 1 (b) to measure the cosine similarity is as shown in [Equation 3].

As a result, r and S2 result in 'very similar' according to the result value.

Fig. 2 (c) is a diagram representing the relationship between r and S3 as a set. r∩S3 = {a, b, c} Therefore, O (r) = {a, b, c} and O (r∩S2) = {a, b, c}. Substituting this in Figure 1 (b) to measure the cosine similarity is as shown in [Equation 4].

As a result, r and S3 result in 'very similar' according to the result value.

Generalizing the OMUCS according to the present invention described above is defined as follows.

For reference, if the general formula is defined to obtain the cosine similarity will be summarized as follows.

Here, 'i' means original sentence, 'j' means comparative sentence, and Si and Sj mean word based set of original sentence and comparative sentence, respectively.

Sequence Alignment is a technique used in bioinformatics to analyze the sequence information of DNA or RNA. A sequencing definition represents a correlation between a protein sequence or a nucleic acid sequence, and the relationship between sequences can be expressed in a way that indicates how closely the sequences are functionally and evolutionarily related, and which parts of the sequence have such associations. have. Analyzing the sequence after aligning it includes the same meaning when comparing two organisms to select organs of common origin, such as head-head, leg-leg, and tail-tail.

The purpose of sequence alignment is to identify sequences of interest and sequences with high homology to infer the function of sequences or to predict quantitative correlations or related functional sites among related sequences. do. In addition, sequence alignment is required for DNA probe or nucleic acid amplification reaction to be applied to a specific population. Representative methods of sequencing include global sorting, regional sorting, and semi-global sorting. In the text plagiarism search system and the text plagiarism identification method according to the present invention, the sorting methods are used to compare the similarity of strings having a sequence of sequences in searching for document plagiarism.

Global alignment assumes that the two sequences are already known and must be aligned over their entire length. A typical algorithm for sorting global sequences is the Needleman-Wunsch algorithm. The sorting method creates an optimal alignment from the high-score alignment of the partial matrix while sweeping from the upper left to the lower right of the matrix, so that only the path that records the highest point along the matrix is traced, resulting in an optimal alignment.

For example, it demonstrates in detail.

Sentence 1

I am a boy who is a hansome guy and he is handsome and powerful.

Word pattern

boy-handsome-guy-handsome-powerful

Character matching

A-C-T-C-G

Sentence 2

I am a boy who is a hansome boy and a powerful guy and a boy is powerful.

Word pattern

boy-handsome-boy-powerful-guy-boy-powerful

Character matching

A-C-A-G-T-A-G

Each score is assigned to the scores of the words that match the word, the scores of the discrepancy, and the scores of the spaces, and the sequence sorting result with the highest possible sum of the various sorted scores is found.

In the present invention, if it matches, it is +1 point, if it is inconsistent, 0 point is set, and -1 point is provided for a space.

When the ACTCG sequence and the ACAGTAG sequence are aligned using Equation 5, the result shown in Table 1 is obtained.

Where a [0,0] = 0 and score (i, j) is the score according to whether the i, j th character matches. The last value at the bottom right is the result of global sort. Thus, the global alignment value of the two sequences is two.

A (boy) C (handsome T (guy) C (handsome G (powerful 0 -One -2 -3 -4 -5 A (boy) -One One 0 -One -2 -3 C (handsome -2 0 2 One 0 -One A (boy) -3 -One One 2 One 0 G (powerful -4 -2 0 One 2 2 T (guy) -5 -3 -One One One 2 A (boy) -6 -4 -2 0 One One G (powerful -7 -5 -3 -One 0 2

Geosorting is a way to search a sequence database for a sequence that is not yet known, or to find parts that match a query sequence in very long DNA, such as genomes. In other words, the most optimal subsequence is found by the Smith-Waterman algorithm.

Applying Equation 6 to the local alignment sequence, [Table 2], the following results are produced. In Table 2, the maximum value is the longest subsequence, and the value is 2.

A (boy) C (handsome T (guy) C (handsome G (powerful 0 -One -2 -3 -4 -5 A (boy) -One One 0 0 0 0 C (handsome -2 0 2 One 0 0 A (boy) -3 0 One 0 0 0 G (powerful -4 0 0 0 0 One T (guy) -5 0 0 One 0 0 A (boy) -6 One 0 0 0 0 G (powerful -7 0 0 0 0 One

Semi-global sorting is a global sorting technique that adds the ability to allow unscored gaps at the beginning and end of a sequence.

Applying [Equation 7] to make [Table 3] is as follows.

a [i, 0] = a [0, j] = a [0,0] = 0. The value of the semi-global alignment is 3.

A (boy) C (handsome T (guy) C (handsome G (powerful 0 0 0 0 0 0 A (boy) 0 One 0 -One 0 0 C (handsome 0 0 2 One 0 0 A (boy) 0 One One 2 One 0 G (powerful 0 0 One One 2 2 T (guy) 0 0 0 2 One 2 A (boy) 0 One 0 0 2 One G (powerful 0 0 One 0 One 3

As explained above, global sorting is characterized by how consistent it is in the overall view of the sequence, local sorting is characterized by finding the optimal subsequence, and semi-global sorting is used for sequences with many differences in length. Efficient for comparison.

Therefore, the text plagiarism search system and the text plagiarism checking method according to the present invention have the advantage of coping with partial copying of sentences, changes in sentence arrangement, and deletion or insertion of words in sentences.

A text plagiarism check method using the text plagiarism search technique according to the present invention described above will be described. In the following embodiments, although English is the target text, the present invention is not limited thereto, and sentences written in Korean and various other languages may also be the target of plagiarism search.

3 is a block diagram illustrating a text plagiarism search system according to the present invention, and FIG. 4 is a flowchart sequentially showing a text plagiarism checking method according to the present invention.

S10; Classifying the original document by sentence

The text plagiarism search system 100 according to the present invention guides the input method when a user inputs an original document and a comparative text document in order to confirm the degree of plagiarism between the original document and the comparative original document, in addition to the original document and the comparison. It includes a user interface 110 for performing any process, such as outputting the degree of plagiarism between the text to the user.

The original document input to the text plagiarism search system 100 through the user interface 110 is transmitted to the document analysis module 120 and classified into sentence units.

In general, since the sentences are separated by a period (.), The document analysis module 120 classifies the original document including a plurality of sentences based on this.

S20; Classifying the classified original sentences in word units

Each sentence classified through step S10 is classified by word by the document analysis module 120. At this time, among the words constituting a sentence, there are stop words, stems, etc. to fit the phrase, the document analysis module 120 performs a process of searching and deleting them to extract only the complete word.

If the sentence is in English, it may further include converting uppercase letters to lowercase letters.

5 is a diagram illustrating a document analysis process performed in the document analysis module 120 by way of example.

As shown in FIG. 5, They, are, who, are, etc. are stopwords for matching phrases, and 's' after nouns is a stem for expressing plurals of nouns. ) Removes them and sorts the sentences by pure word units.

In the case of Hangul, 'search' and 'pronoun' may be the stopwords.

The pronouns in English and Korean, which are disclosed as examples of the above-mentioned stopwords, may be set as stopwords as needed, and may also be taken as words, and the selection of stopwords may be changed as necessary.

The classified words are collected in a word-based vector format and classified and stored for each sentence.

S30; Step of classifying the comparative document into sentence units

The user inputs a comparison document to compare plagiarism with the original document classified above through the user interface 110, and the input comparison document is classified by sentence unit by the document analysis module 120.

S40; Classifying the classified comparative sentence by word unit

Each sentence classified through the step 'S30' is classified in word units through the document analysis module 120. Since this is described above, the description thereof will be omitted.

S50; Measuring the similarity between the original and the comparative classified by word unit by sentence unit

The original document and the comparative document, which are each set in a word-based vector format by the document analysis module 120 and classified by sentence, are checked for overlapping degree by the OMCUS measurement by the word plural number checking module 130.

As described above, the vector set of all sentences of the original document and the vector set of all sentences of the comparison document are individually measured by the OMCUS measurement, and when the result value does not exceed a predetermined threshold, the comparison target (S60, S70)

The threshold is a predetermined value within the range of the cosine similarity result value (minimum value 0, maximum value 1) of the sentence of the original document and the sentence of the comparative document. have.

That is, when the threshold value is set to 0.75, if the result of checking the degree of redundancy through the OMCUS measurement results in a result smaller than 0.75, the corresponding original sentence and the comparison sentence set in a word-based vector form do not perform further comparison. If the result is 0.75 or more, the original text and the comparative text are subjected to a sequence alignment process.

S80; Sorting by word unit

The original sentence and the comparative sentence having a result value higher than the threshold value through the step 'S50' performs the sequence alignment of the words by the sequence alignment check module 140.

In this case, the sequence tack relation identification module 140 performs sequence alignment between the aggregated words by applying the global sorting, local sorting and anti-global sorting techniques.

S90; Identifying the sum of digitized values through sequencing

The plagiarism determination module 150 adds all of the result values of the sequence alignment relationship verification module 140 and compares the result with a threshold value.

That is, the result value of all the aggregated sentences of the original document and the comparative document with global sorting, the result value of all the aggregated sentences of the original document and the comparative document with local sorting, and semi-global sorting. The sum of the result values of all the sentences of the applied original document and the compared original document is compared with the threshold value.

However, by applying only one or more methods selected from global sorting, local sorting, or anti-global sorting, the sum of only the result values of the applied methods may be compared.

In addition, after applying one or more methods selected from global sorting, local sorting, or semi-global sorting, the resultant may be averaged by sentence and applied sorting method, and the average value may be compared with a threshold associated with the average.

At this time, if the result value is smaller than the threshold value, the original document and the comparative document are judged to have a low degree of plagiarism (S120). If the result value is higher than the threshold value, the original document and the comparative document are determined to have a high degree of plagiarism. S110)

As described above, the threshold may be arbitrarily determined or determined by statistics, which is not a fixed value and may be changed as the type and content of the document and various factors.

The experimental example according to the text plagiarism check method according to the present invention will be described below.

First, check if the original document can be detected as it is.

Document 1 is the original news, and Document 2 is a copy of the original news as it is, applying both cosine similarity and OMUCS according to the present invention.

As a result, the cosine similarity as well as the OMUCS according to the present invention, all of the results were 1, through which the document 1 and the document 2 was identified as the same document. The reason why both the cosine similarity and the OMUCS according to the present invention resulted in the same result is that the sentences of the two documents coincide. As a result of the experiment, the improvement of the OMUCS according to the present invention was not confirmed.

In addition, the position of the same sentence of the document 1 and the document 2 is also displayed along with the result value. This is because the text plagiarism checking method according to the present invention can specify the order of the classified sentences to distinguish them from other sentences while going through the step of classifying the sentences.

Next, an experiment is performed by attaching an arbitrary sentence in Document 1, which is the original document, to Document 2, which is another document.

As a result, the cosine similarity and the result of the OMUCS according to the present invention is 1, and the position of the sentence is also displayed.

Next, in the case where the original document Document 1 is copied to another document Document 2, which is smaller than a sentence, the cosine similarity is 0.67, while the result of OMUCS according to the present invention is 0.98.

Subsequently, the contents of the experiments and the results that can clearly show the effect of the text plagiarism check method according to the present invention are disclosed.

6 and 7 are tables describing original English texts and comparative English texts applied for the implementation of the method for checking text plagiarism according to the present invention, and FIG. 8 is a comparative text with original English texts described in FIGS. 6 and 7. The graph shows the results of applying the English text to the conventional cosine similarity and OMUCS according to the present invention, which will be described with reference to this.

As shown in FIG. 8, the original English text and the comparative English text shown in FIGS. 6 and 7 show different similarities according to cosine similarity and OMUCS. In fact, the comparative English text shown in FIG. 7 is only the same as the original English text shown in FIG. 6, but the contents are almost the same, so that the similarity between the original English text and the comparative English text is increased to obtain an accurate comparison result. Bar, it can be seen that the text plagiarism check method according to the present invention is improved than the conventional.

9 is a graph showing the results of applying the original English text and the comparative English text described in FIGS. 6 and 7 in the sequence alignment process according to the present invention.

In analyzing the results of FIG. 9, the values of global sorting, local sorting, and anti-global sorting are different. This is because the original English sentences were not plagiarized, but changed the sentence. The total number of patterns in 124 was 67 points for global alignment, 75 points for local alignment, and 75 points for anti-global alignment. Therefore, it can be seen that the text plagiarism check method according to the present invention is effective in checking plagiarism for a document in which the original sentence is changed.

Thresholds of sequence alignment values that can identify plagiarism may be changed through statistics as described above, and may also be changed depending on the contents of a document.

According to the present invention as described above, it is easy to numerically determine whether the comparative document has copied the original document, and even if the sentence format is changed, it is possible to check whether the actual sentence is copied more precisely plagiarism. Identification is possible.

Claims (5)

An original document sentence classification step of classifying the input original document in sentence units; An original document word classification step of classifying sentences classified through the original document sentence classification step in word units; A comparative document sentence classification step of classifying the input comparative document into sentence units; A comparison document word classification step of classifying sentences classified by the comparison document sentence classification step in word units; Comparing the original sentences classified in word units and the comparative text sentences through the original document word classification step and the comparative original document word classification step, and identifying the same words in the original sentence and the comparison text sentence compared to each other; An OMUCS calculation step of applying the same word identified in the same word checking step to an OMUCS [original text] [comparative text] with a modified cosine similarity; And A similarity determination step of a sentence comparing the result output through the OMUCS operation step with a first threshold value and determining whether or not the similarity is based on the similarity between the original sentence and the comparative body sentence; Text plagiarism check method comprising a. The method of claim 1, The method further includes a sequence sorting step of obtaining a maximum value by applying one or more methods selected from a global sorting method, a local sorting method, and a semi-global sorting method to the original sentence and the comparative sentence identified as similar through the similarity determination step. Text plagiarism check method characterized in that. The method of claim 2, And comparing the sum of the maximum values calculated through the sequence alignment step with a second threshold value, further comprising a plagiarism determination step of determining whether plagiarism is based on similarity between the original sentence and the comparative sentence. How to check for plagiarism. An original document sentence classification step of classifying the input original document in sentence units; An original document word classification step of classifying sentences classified through the original document sentence classification step in word units; A comparative document sentence classification step of classifying the input comparative document into sentence units; A comparison document word classification step of classifying sentences classified by the comparison document sentence classification step in word units; The original document classified in word units and the comparative text, respectively, through the word classification step of the original document and the word classification step of the comparative document, respectively, and applied to a COS similar to the cosine [original text] [comparative text]. Operation step; A similarity determination step of determining whether or not the similarity is based on the similarity between the original sentence and the comparative text sentence, by comparing the result output through the COS operation step with a third threshold value; And A sequence sorting step of obtaining a maximum value by applying one or more methods selected from a global sorting method, a local sorting method, and a semi-global sorting method to the original text and the comparative text that are identified as similar through the similarity determining step; Text plagiarism check method comprising a. The method of claim 4, wherein And comparing the sum of the maximum values calculated through the sequence sorting step with a fourth threshold value, further including a plagiarism determination step of determining a plagiarism according to similarity between the original sentence and the comparative sentence. How to check for plagiarism.

Images