JP2023025933A - Similarity degree determination device, similarity degree determination system, similarity degree determination method, and program - Google Patents

Abstract

To provide a similarity degree determination device, a similarity degree determination system, a similarity degree determination method, and a program, capable of calculating a similarity degree of texts with higher accuracy.SOLUTION: A similarity degree determination device includes: a sentence vector acquisition unit that acquires a sentence vector that is a feature vector for each sentence obtained by decomposing a first text that is an analysis source text and a second text that is a text to be compared; a key phrase acquisition unit that acquires a key phrase that is included in each of the first text and the second text and is an important element constituting the texts; and a similarity degree calculation unit that calculates a comprehensive similarity degree of the first text and the second text based on a similarity degree between the sentence vectors between the first text and the second text and a similarity degree of an appearance degree of the same key phrase.SELECTED DRAWING: Figure 2

Description

There is an application for the application of Article 30, Paragraph 2 of the Patent Act. Disclosure through telecommunication lines Website publication date August 25, 2020 Website URL https://kagemusya.com biz-samurai. com/

There is an application for the application of Article 30, Paragraph 2 of the Patent Act. Disclosure through telecommunication lines Website publication date August 27, 2020 Website URL https://cro-co. co. jp/information/news/service/418/

The present invention relates to a similarity determination device, a similarity determination system, a similarity determination method, and a program.

With the enormous number of online documents accessible via the Internet, techniques related to searching for similar documents are being utilized in many fields such as document plagiarism searching.

In relation to this, a document similarity derivation device has been proposed that derives the similarity between a group of documents classified into a plurality of categories and an input document (see Patent Document 1). Specifically, the document similarity deriving device calculates a feature vector of the input document with the weight as an element based on the result of morphological analysis of the sentence included in the input document, and calculates the feature vector of the input document, and Calculate the average feature vector of the document group from the feature vector of each document in the input document and the average feature vector of each document group to determine which document group the input document is most similar to judge.

JP-A-2009-53743

In the technique described in Patent Document 1, the similarity of sentences is calculated mainly based on feature vectors, and elements other than feature vectors are not taken into account. .

The present invention has been made in consideration of such circumstances, and provides a similarity determination device, a similarity determination system, a similarity determination method, and a program capable of calculating the similarity of sentences with higher accuracy. intended to provide

In order to achieve the above object, the similarity determination device of the present invention provides a sentence vector, which is a feature vector for each sentence obtained by decomposing a first sentence, which is a sentence to be analyzed, and a second sentence, which is a sentence to be compared. a sentence vector acquisition unit that acquires a key phrase that is included in each of the first sentence and the second sentence and is an important element that constitutes the sentence; a key phrase acquisition unit that acquires the first sentence and the Similarity calculation for calculating a total similarity between the first sentence and the second sentence based on the similarity between the sentence vectors with the second sentence and the similarity of the appearance of the same key phrase and

Further features and aspects of the present invention will become apparent from the detailed description of the embodiments set forth below, with reference to the accompanying drawings.

According to the present invention, it is possible to calculate the similarity of sentences with higher accuracy.

It is a figure which shows an example of the whole structure of the similarity determination system 1 which concerns on embodiment. It is a figure which shows the structure of the similarity determination apparatus 100 which concerns on embodiment. FIG. 10 is a diagram showing an example of how the sentence vector comparison unit 132 calculates the similarity between sentence vectors. FIG. 10 is a diagram showing an example of how the key phrase comparison unit 134 calculates the degree of occurrence of matching key phrases in the first sentence and the second sentence in each sentence. FIG. 10 is a diagram schematically showing the content of a process of calculating a total similarity between a first sentence and a second sentence by a similarity calculating unit 130; 4 is a flow chart showing an example of a process of calculating a total similarity between a first sentence and a second sentence by the similarity determination device 100; FIG. 10 is a diagram showing an example of a screen displayed on the terminal device 200 by the similarity determination device 100 after the process of calculating the total similarity is performed.

Hereinafter, a similarity determination device, a similarity determination system, a similarity determination method, and a program according to embodiments will be described with reference to the drawings. The similarity determination device is a device that determines the degree of similarity between a text to be analyzed and a text to be compared based on feature vectors and key phrases of each sentence that constitutes the text. The similarity determination device, for example, calculates the degree of similarity between the analysis source sentence and the comparison target sentence, and transmits the calculation result to the terminal device. The terminal device is, for example, a personal computer, a tablet computer, a smart phone, or the like. The text to be analyzed and the text to be compared may be obtained from a terminal device, or may be obtained by other methods such as being automatically obtained from an external environment such as the Internet. The similarity determination device may use functions of other servers when acquiring the feature vector of each sentence.

FIG. 1 is a diagram showing an example of the overall configuration of a similarity determination system 1 according to an embodiment. A similarity determination system 1 includes a similarity determination device 100 and a terminal device 200 . Similarity determination device 100, terminal device 200, and external server 300 communicate with each other via a network. The network NW includes, for example, a part or all of a WAN (Wide Area Network), a LAN (Local Area Network), the Internet, a provider device, a wireless base station, a dedicated line, and the like. The role of the external server 300 will be described later.

FIG. 2 is a diagram showing the configuration of the similarity determination device 100 according to the embodiment. The similarity determination device 100 includes, for example, a communication unit 110, a text acquisition unit 120, a sentence vector acquisition unit 122, a key phrase acquisition unit 124, a similarity calculation unit 130, and a storage unit 150.

Each unit other than the communication unit 110 and the storage unit 150 is implemented by executing a program (software) by a hardware processor such as a CPU (Central Processing Unit). Some or all of these components are hardware (circuit part; circuitry) or by cooperation of software and hardware. The program may be stored in advance in a storage device (a storage device with a non-transitory storage medium) such as a HDD (Hard Disk Drive) or flash memory, or may be stored in a removable storage such as a DVD or CD-ROM. It may be stored in a medium (non-transitory storage medium) and installed in the storage device by loading the storage medium into the drive device.

The communication unit 110 includes a network interface such as a NIC (Network Interface Card). Each unit of the similarity determination device 100 uses the communication unit 110 to communicate with the terminal device 200 and the external server 300 via the network NW.

The storage unit 150 is implemented by, for example, a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD, a flash memory, or a hybrid storage device in which a plurality of these are combined. The storage unit 150 stores, for example, data such as an acquired sentence 152, an acquired sentence vector 154, an acquired key phrase 156, and the number 158 of pairs of sentence vectors.

The text acquisition unit 120 acquires, for example, a text to be analyzed and a text to be compared from the terminal device 200 . In the following, the original text to be analyzed is referred to as the first text, and the text to be compared is referred to as the second text. The sentence acquisition unit 120 causes the storage unit 150 to store the acquired first sentence and second sentence.

Here, before describing the sentence vector acquisition unit 122, the role of the external server 300 will be described. The external server 300 has various analyzers such as segmenting sentences, specifying the appearance positions of words from sentences, acquiring feature vectors (sentence vectors), and acquiring key phrases. Provide libraries, etc. For example, the external server 300 can use a natural language processing model of BERT (Bidirectional Encoder Representations from Transformers) to create a corresponding sentence vector for each sentence based on delimiters such as full stops. In addition, the external server 300 can acquire important elements (key phrases) that make up a sentence for each sentence using, for example, spaCy/GiNZA (a type of Japanese morphological analyzer).

The external server 300, for example, clusters a corpus prepared in advance and calculates the score of each word for each classified cluster. When extracting key phrases from a sentence, the score of each word in the cluster to which the sentence belongs is used to obtain the score of the phrase candidate. The external server 300 can obtain a more accurate word score by collecting many similar sentences. Here, the external server 300 may extract key phrases using genre-specific corpora. For example, the external server 300 prepares a corpus in advance for each genre such as sports and cooking, and uses the corpus corresponding to the genre specified by the sentence vector acquisition unit 122 to calculate the word score. In this case, the sentence vector acquisition unit 122 may receive designation of genre from the terminal device 200 .

The sentence vector acquisition unit 122 transmits the first sentence and the second sentence stored in the storage unit 150 to the external server 300 to request creation of sentence vectors, and obtains sentences corresponding to the first sentence and the second sentence, respectively. Get a vector. In this embodiment, a sentence vector created based on the first sentence is called a first sentence vector, and a sentence vector created based on the second sentence is called a second sentence vector. The sentence vector acquisition unit 122 outputs the acquired first sentence vector and second sentence vector to the similarity calculation unit 130 .

The key phrase acquisition unit 124 acquires the first and second sentences stored in the storage unit 150, transmits them to the external server 300, and extracts key phrases from each of the first and second sentences. Request and obtain key phrases corresponding to each of the first and second sentences. In this embodiment, a key phrase created based on the first sentence is called a first key phrase, and a key phrase created based on the second sentence is called a second key phrase. The key phrase acquisition unit 124 outputs the acquired first key phrase and second key phrase to the similarity calculation unit 130 .

The similarity calculator 130 calculates the total similarity between the first sentence and the second sentence. The similarity calculation unit 130 includes, for example, a sentence vector comparison unit 132, a key phrase comparison unit 134, an overall similarity calculation unit 136, and a similarity display control unit 138.

The sentence vector comparison unit 132 compares the first sentence vector and the second sentence vector based on the inter-vector distance (Euclidean distance) or the cosine similarity between the first sentence vector and the second sentence vector created based on the first sentence and the second sentence, respectively. and the second sentence vector are calculated. The sentence vector comparison unit 132 comprehensively calculates the similarity between the first sentence vector and the second sentence vector, for example, and calculates a first index value for calculating the total similarity between the first sentence and the second sentence. do.

FIG. 3 is a diagram showing an example of how the sentence vector comparison unit 132 calculates the degree of similarity between sentence vectors. As illustrated, each of the first sentence and the second sentence is broken down into a plurality of sentences such as sentence 1, sentence 2, sentence 3, and so on. The similarity between the sentence vector of sentence 1 of the first sentence and the sentence vector of sentence 2 of the second sentence is 97%. is 35%, and the similarity between the sentence vector of sentence 3 of the first sentence and the sentence vector of sentence 4 of the second sentence is 97%. In this way, the sentence vector comparison unit 132 calculates the degree of similarity between the sentence vector of each sentence included in the first sentence and the sentence vector of each sentence included in the second sentence.

For example, after calculating the degree of similarity between the sentence vectors, the sentence vector comparison unit 132 determines whether the calculated degree of similarity is equal to or greater than a threshold. Then, the sentence vector comparison unit 132 counts the similarities (including the maximum value of the similarities) greater than or equal to the threshold and the number of sets of sentence vectors whose similarities are greater than or equal to the threshold. In this embodiment, for example, the similarity threshold between sentence vectors is set to 85%. In this case, the similarities of sentence vector set (1), sentence vector set (4), and sentence vector set (5) in FIG. 3 are 97%, 88%, and 97%, respectively. Since these are equal to or greater than the threshold of 85%, the sentence vector comparison unit 132 calculates the similarity between the sentence vectors and the number of sets “3” and stores them in the storage unit 150 .

The sentence vector comparison unit 132 may simply calculate the first index value based on the similarity of the calculated similarity between sentence vectors equal to or higher than the threshold, or the similarity equal to or higher than the threshold and the similarity to the threshold. The first index value may be calculated based on the number of sets of sentence vectors as described above. Further, the sentence vector comparison unit 132 may obtain a weighted sum of the number of sets by increasing the weight as the value exceeding the threshold increases, for example.

Furthermore, the sentence vector comparison unit 132 may calculate the first index value based on the maximum similarity value. In this embodiment, in order to calculate the overall similarity between the first sentence and the second sentence with high accuracy, the sentence vector comparison unit 132 calculates the maximum value of the similarity between sentence vectors, Suppose that the first index value is calculated based on the number of pairs of . For example, in the case of FIG. 3, the sentence vector comparison unit 132 uses "97%" and "3" to calculate the first index value.

Note that the sentence vector comparison unit 132 may simply calculate the first index value based on the number of pairs of sentence vectors whose similarity is equal to or greater than the threshold, or may simply calculate the first index value based on the maximum value of the similarity. value may be calculated. For example, the sentence vector comparison unit 132 sets the first index value so that the larger the number of pairs of sentence vectors whose similarity is equal to or higher than the threshold value, the larger the first index value, and the larger the maximum value of the similarity, the larger the first index value. Comprehensive similarity is calculated so as to increase. As long as this tendency exists, the sentence vector comparison unit 132 may calculate the first index value by any method.

The key phrase comparison unit 134 compares the key phrases extracted from the first sentence and the second sentence based on the degree of occurrence of matching key phrases in the first sentence and the second sentence. A second index value for calculating the total similarity between the first sentence and the second sentence is calculated. The degree of appearance is the amount of information that indicates the relative frequency or weight of appearance in consideration of the lengths of the first and second sentences.

FIG. 4 is a diagram showing an example of how the key phrase comparison unit 134 calculates the degree of occurrence of matching key phrases in the first and second sentences. First, the key phrase comparison unit 134 counts the number of appearances of key phrases in the first and second sentences. Then, the keyphrase comparison unit 134 divides the number of appearances of the keyphrase by a coefficient (first coefficient or second coefficient) corresponding to the length of the first sentence or the second sentence, and calculates the degree of appearance of the keyphrase. Calculate as This coefficient is a value that increases as the length of the first sentence or the second sentence increases.

For example, as shown, key phrases appearing in each of the first and second sentences are "dribbling", "youth", "ball", "playing style", "2019" and "2015". . The key phrase "dribble" appears five times in the first sentence and eight times in the second sentence. On the other hand, if the coefficient α1 corresponding to the first sentence is calculated to be 50 and the coefficient α2 corresponding to the second sentence is calculated to be 100, the degree of occurrence of "dribble" in the first sentence is calculated to be 0.1. , the degree of appearance in the second sentence is calculated as 0.08. There may also be key phrases that appear in only one of the first and second sentences. In that case, the number of appearances of key phrases and the degree of appearance of key phrases corresponding to the sentence in which the key phrase does not appear are zero.

Then, the key phrase comparison unit 134, for example, virtually arranges key phrases that appear in at least one of the first sentence and the second sentence, and generates a vector (key phrase vector) whose elements are the number of occurrences of the key phrases in the first sentence. Define each of the sentence and the second sentence. Assuming that the keyphrase vector corresponding to the first sentence is the first keyphrase vector V1 and the keyphrase vector corresponding to the second sentence is the second keyphrase vector V2, the example in FIG. be done.
V1 = (0.1, 0.06, 0.2, 0.14, 0.2, 0.04, ...)
V2=(0.08, 0, 0.1, 0.05, 0.02, 0.08, ...) (1)

The key-phrase comparison unit 134, for example, compares the inter-vector distance (Euclidean distance) between the first key-phrase vector V1 and the second key-phrase vector V2 and the cosine similarity (that is, the first key-phrase vector V1 and the second key-phrase vector V2). (which is an example of the degree of similarity between the first sentence and the second sentence of the key phrase) is calculated as the second index value.

The total similarity calculator 136 calculates the total similarity between the first sentence and the second sentence based on the first index value and the second index value. For example, the total similarity calculation unit 136 calculates the total similarity such that the larger the first index value, the larger the total similarity, and the larger the second index value, the larger the total similarity. . As long as this tendency exists, the total similarity calculation unit 136 may calculate the total similarity by any method.

FIG. 5 is a diagram schematically showing the content of processing for calculating the total similarity between the first sentence and the second sentence by the similarity calculation unit 130. As shown in FIG. In this embodiment, for example, the number of pairs of sentence vectors whose similarity is equal to or greater than the threshold is calculated as n, the maximum similarity is calculated as m, and the number of appearances of the key phrase in the first sentence is P ₁ , the key phrase The number of occurrences of the phrase in the second sentence is _P2 , the normalization factor for the first sentence is set to _α1 , and the normalization factor for the second sentence is set to _α2 .

The total similarity calculation unit 136 sets F to the first index value calculated by the sentence vector comparison unit 132 based on n and m, and the key phrase comparison unit 134 calculates F based on P ₁ , P ₂ , α ₁ , and α _{2 .} Let Q be the second index value calculated by A method of calculating the first index value F is represented by, for example, F=f(n,m). As described above, f(n,m) is set such that the larger the number n of sentence vector sets whose similarity is equal to or higher than the threshold, the larger the first index value F becomes, and the maximum value m of similarity is It is a function for calculating the first index value F so that the first index value F increases as the value increases. Moreover, the calculation method of the 2nd index value Q is represented by Q=q( _P1 ,P2,( _{alpha )1} ,(alpha) ₂ ), for example. Q = q (P ₁ , P ₂ , α ₁ , α ₂ ) is obtained by dividing the number of appearances P ₁ of key phrases in the first sentence by the coefficient α ₁ according to the length of the first sentence, as described above. and the value obtained by dividing the number of appearances _P2 of the key phrase in the second sentence by a coefficient _α2 corresponding to the length of the second sentence, are calculated as the degree of appearance of the key phrase in each sentence, and the calculated value is It is a function for calculating the second index value Q based on.

Then, based on the first index value F and the second index value Q, the total similarity calculator 136 calculates the total similarity S between the first sentence and the second sentence. A method of calculating the total similarity S is represented by, for example, S=h(F, Q). S=h(F, Q) is such that the larger the first index value F, the larger the total similarity S, and the larger the second index value Q, the larger the total similarity S. is a function for calculating the total similarity S as follows.

The similarity display control unit 138 displays, for example, the overall similarity, the first index value, the second index value, and a graph (radar chart) is displayed on the display unit of the terminal device 200 (not shown). Details will be described later.

FIG. 6 is a flow chart showing an example of a process of calculating the total similarity between the first sentence and the second sentence by the similarity determination device 100 .

First, the sentence acquisition unit 120 acquires the full sentences of the first sentence and the second sentence from, for example, the terminal device 200 (step S200). The sentence vector acquisition unit 122 transmits the first sentence and the second sentence to the external server 300 and requests creation of sentence vectors. Then, the sentence vector acquisition unit 122 acquires the sentence vector of the first sentence and the sentence vector of the second sentence from the external server 300, and the key phrase acquisition unit 124 acquires the key phrase of the first sentence and the key phrase of the second sentence. is obtained from the external server 300 (step S202).

Next, the sentence vector comparison unit 132 calculates the degree of similarity between sentence vectors for each set created based on each of the first sentence and the second sentence (step S204). The sentence vector comparison unit 132 selects one sentence vector from the first sentence vectors, selects one sentence vector from the second sentence vectors, and calculates the similarity A _i,j between the selected sentence vectors. Run on combinations. Then, the sentence vector comparison unit 132 determines whether or not the calculated similarity A _i,j between the sentence vectors is equal to or greater than the threshold Th (step S208). A _i,j represents the degree of similarity between the i-th first sentence vector and the j-th second sentence vector. If the calculated similarity A _i,j between sentence vectors is not equal to or greater than the threshold Th, the sentence vector comparison unit 132 proceeds to step S216. When the calculated similarity A _i,j between sentence vectors is equal to or greater than the threshold Th, the sentence vector comparison unit 132 counts up the number n of sentence vector pairs equal to or greater than the threshold Th, and causes the storage unit 150 to store the number n. (Step S212).

Then, the sentence vector comparison unit 132 determines whether or not the similarity A _i,j has been calculated for all combinations of sentence vectors (step S216). If the similarity A _i,j has not been calculated for all combinations of sentence vectors, the sentence vector comparison unit 132 returns to step S204, selects the next set of sentence vectors, and calculates the similarity A _i,j. do. When the similarities A _i,j are calculated for all sentence vector combinations, the sentence vector comparison unit 132 extracts the maximum value m among the calculated similarities (step S220). Next, the sentence vector comparison unit 132 calculates a first index value F based on the number n counted in step S212 and the maximum similarity value m calculated in step S220 (step S224).

In this embodiment, the similarity determination device 100 may extract matching key phrases in each of the first sentence and the second sentence in parallel with the process of step S204 (step S206).

The key phrase comparison unit 134 calculates the number of appearances P ₁ and P ₂ of key phrases in each of the first sentence and the second sentence (step S210). Next, the key phrase comparison unit 134 calculates normalization coefficients α ₁ and α ₂ for the first and second sentences (step S214). Based on P ₁ , P ₂ , α ₁ , and α ₂ calculated above, the key phrase comparison unit 134 calculates the second index value Q (step S218). Then, the key phrase comparison unit 134 creates a radar chart regarding the degree of similarity of appearance of key phrases, and the similarity degree display control unit 138 causes the display unit of the terminal device 200 to display the radar chart (step S222).

Finally, the total similarity calculator 136 calculates the total similarity S of the first sentence and the second sentence based on the first index value F and the second index value Q (step S226).

According to the comparison of key phrases, since key phrases are extracted from the entire sentences, it is possible to determine the similarity of the entire sentences. However, if the second sentence to be compared uses only a part of the first sentence to be analyzed, it may be determined that the similarity is low only by comparing the key phrases. In contrast, similarity can be determined with higher accuracy by combining comparison of key phrases and comparison of sentence vectors as in the present embodiment.

That is, in the present embodiment, by calculating the degree of similarity between sentence vectors, the user can determine the degree of local similarity between the first sentence to be analyzed and the second sentence to be compared. Further, by calculating the degree of similarity of the degree of occurrence of key phrases, the user can determine the degree of similarity between the first sentence to be analyzed and the second sentence to be compared.

As described above, in the present embodiment, the first sentence to be analyzed and the second sentence to be compared have the same purpose of showing an outline of the sentence, that is, creation of sentence vectors and extraction of key phrases. However, the overall similarity is calculated by performing two types of analysis that are completely different from each other. This makes it possible to detect the diversion of sentences by simple batch conversion, such as replacing the Olympics with the Olympics.

FIG. 7 is a diagram showing an example of a screen displayed on the terminal device 200 by the similarity determination device 100 after the process of calculating the total similarity is performed. This screen includes, for example, a display area A1 displaying the "total similarity", a display area A2 displaying the "first index value", a display area A3 displaying the "second index value", and " A display area A4 is provided in which "Radar Chart Concerning Similarity of Appearance of Key Phrases" is displayed. As shown in the figure, after the processing related to the calculation of the total similarity is performed, the similarity display control unit 138 displays the “analysis result detailed report” including the items displayed in each display area on the display screen of the terminal device 200. display. In the "radar chart for similarity of key phrase appearance", a line L1 representing the appearance of key phrases in the first sentence and a line L2 representing the appearance of key phrases in the second sentence are displayed. Key phrases with the highest appearance frequency in the first sentence are arranged in the 12 o'clock direction of the radar chart, that is, directly above, and the key phrases are arranged counterclockwise in order of appearance frequency in the first sentence. Therefore, the line L1 indicating the first sentence is drawn as a spiral curve whose radius gradually decreases counterclockwise from directly above. On the other hand, if the similarity is low, the line L2 indicating the second sentence does not have a spiral shape but has an distorted shape. In addition, when the second sentence is used in part of the first sentence, a specific key phrase is drawn with a prominent high or low frequency, but many of the key phrases have similar appearance frequencies. Therefore, it becomes close to a spiral shape. This radar chart allows the user to visually compare the frequency of appearance of key phrases in the first sentence and the frequency of appearance of key phrases in the second sentence.

Some or all of the functions of the external server 300 described above may be included in the functions of the similarity determination device 100 . For example, the sentence vector acquisition unit 122 of the similarity determination device 100 may have a function of generating sentence vectors, and the key phrase acquisition unit 124 of the similarity determination device 100 may have a function of extracting key phrases. You may

Also, the first sentence and the second sentence acquired by the sentence acquisition unit 120 may be any copyrighted material such as a thesis, article, lyrics, haiku, poem, novel, and the like. As a result, the similarity determination device 100 can be used to detect plagiarism of works such as papers, articles, lyrics, haiku, poems, novels, and the like.

Further, the similarity determination device 100 determines the degree of similarity between sentences (similarity determination) based on both sentence vectors and key phrases corresponding to the sentences. Or if it is specified that the whole is quoted (or vice versa, or if both the first and second sentences quote a different third sentence), the similarity determination device 100 determines that the first sentence and the quoted part of the second sentence may be excluded to calculate the total similarity between the first sentence and the second sentence. For example, when the first sentence and the second sentence are input in HTML (Hyper Text Markup Language) format, the similarity determination device 100 may determine the range of citation based on the citation tag. For the purpose of detecting plagiarism, it is natural that the quoted part of the text is the same, but if the similarity is determined including the quoted part, the similarity will be unnecessarily high. This is because the purpose of whether or not texts are plagiarized on a regular basis cannot be achieved. Even if the citation is not entered in HTML, etc., if the citation is correct, it is normal to change the font or clearly indicate the citation source with an asterisk (*), etc. Therefore, such citations may be machine-detected using natural language processing.

When such citations are made, the similarity score of a specific sentence is extremely high, even though the similarity of the sentence as a whole is not high. Since natural language processing places a heavy load on the similarity determination device 100, machine detection may be operated only when such a local similarity discontinuity is found. When the chart is displayed, the terminal device 200 may display the chart so that the user can manually point out the quoted portion and prompt re-determination.

In addition, when creating an article based on objective facts such as news or the sale of a popular product, similar articles are created in many news media regarding the facts. In such a case, if similarity judgment is performed on the part that conveys the facts, it will be judged that the degree of similarity is high even if it is not plagiarism, and as a result, the purpose of whether or not plagiarism has occurred cannot be achieved. Therefore, portions that convey objective facts may be excluded from similarity determination targets. Unlike citations, in which the places of citation are specified, it is possible to specify the factual part by natural language processing, but it is necessary to judge from the meaning of the sentence, and the factual description is not necessarily the same sentence. , more advanced natural language processing is required. In the case of such news articles, there is a tendency that a plurality of extracted second sentence candidates are relatively new and have similar dates. Therefore, only when a plurality of documents dated close to each other are extracted as candidates for the second sentence, it is possible to detect excluded points by natural language processing, or check whether the article conveys the facts to the user. may be displayed on the terminal device 200 to prompt the user to manually exclude the part of the article that conveys the fact. Similar to the citation, when displaying the radar chart shown in FIG. 7, the user may be prompted to re-determine. , it is preferable to be able to specify an exclusion point before natural language processing.

According to the embodiment described above, the sentence vector acquisition unit acquires the sentence vector, which is the feature vector for each sentence obtained by decomposing the first sentence, which is the sentence to be analyzed, and the second sentence, which is the sentence to be compared. 122, a key phrase acquisition unit 124 that acquires key phrases that are included in each of the first and second sentences and are important elements that constitute the sentences, and a sentence vector between the first and second sentences. By providing a similarity calculation unit 130 that calculates the total similarity between the first sentence and the second sentence based on the similarity between them and the similarity in the appearance of the same key phrase, sentences can be calculated with higher accuracy. can be calculated.

Further, according to the embodiment, by calculating the first index value based on the number of pairs of sentence vectors whose similarity is equal to or higher than a threshold among pairs of sentence vectors, the similarity of sentences can be calculated with higher accuracy. can be calculated.

Furthermore, according to the embodiment, by further including the similarity display control unit 138 that displays a graph representing the degree of similarity of appearance of key phrases between the first sentence and the second sentence, the user can The frequency of occurrence of key phrases in the sentence and the frequency of occurrence of key phrases in the second sentence can be compared.

Although preferred embodiments of the invention have been described above, the invention is not limited to these embodiments. Configuration additions, omissions, substitutions, and other changes are possible without departing from the scope of the present invention. The present invention is not limited by the foregoing description, but only by the scope of the appended claims.

100 Similarity determination device 120 Sentence acquisition unit 122 Sentence vector acquisition unit 124 Key phrase acquisition unit 130 Similarity calculation unit 132 Sentence vector comparison unit 134 Key phrase comparison unit 136 Comprehensive similarity calculation unit 200 Terminal device 300 External server

In order to achieve the above object, the similarity determination device of the present invention provides a sentence vector, which is a feature vector for each sentence obtained by decomposing a first sentence, which is a sentence to be analyzed, and a second sentence, which is a sentence to be compared. a sentence vector acquisition unit that acquires a key phrase that is included in each of the first sentence and the second sentence and is an important element that constitutes the sentence; a key phrase acquisition unit that acquires the first sentence and the Similarity calculation for calculating a total similarity between the first sentence and the second sentence based on the similarity between the sentence vectors with the second sentence and the similarity of the appearance of the same key phrase wherein the similarity calculation unit calculates a first key phrase vector whose elements are the number of appearances of the key phrases included in the first sentence, and the number of appearances of the key phrases included in the second sentence. A second key phrase vector having elements of is obtained, and based on the first key phrase vector and the second key phrase vector, a second index value representing the degree of similarity between the first sentence and the second sentence is calculated A key phrase comparison unit for calculating is provided .

In order to achieve the above object, the similarity determination device of the present invention provides a sentence vector, which is a feature vector for each sentence obtained by decomposing a first sentence, which is a sentence to be analyzed, and a second sentence, which is a sentence to be compared. a sentence vector acquisition unit that acquires a key phrase that is included in each of the first sentence and the second sentence and is an important element that constitutes the sentence; a key phrase acquisition unit that acquires the first sentence and the Similarity calculation for calculating a total similarity between the first sentence and the second sentence based on the similarity between the sentence vectors with the second sentence and the similarity of the appearance of the same key phrase and a display unit for displaying a graph representing the degree of similarity of appearance between the first sentence and the second sentence of the key phrase acquired by the key phrase acquisition unit. and the similarity calculation unit calculates a first key phrase vector whose elements are the number of appearances of the key phrases included in the first sentence, and the number of appearances of the key phrases included in the second sentence. A second key phrase vector as an element is obtained, and a second index value representing the degree of similarity between the first sentence and the second sentence is calculated based on the first key phrase vector and the second key phrase vector. wherein the display control unit includes: a first line connecting points representing degrees of appearance of each of the plurality of key phrases in the first sentence; A radar chart displaying a second line connecting points representing respective appearance degrees is displayed on the terminal device as the graph.

Claims (9)

a sentence vector acquisition unit that acquires a sentence vector that is a feature vector for each sentence obtained by decomposing a first sentence that is an analysis source sentence and a second sentence that is a comparison target sentence;
a key phrase acquisition unit that acquires a key phrase that is included in each of the first sentence and the second sentence and is an important element that constitutes the sentence;
total similarity between the first sentence and the second sentence based on the similarity between the sentence vectors between the first sentence and the second sentence and the similarity of the occurrence of the same key phrases; a similarity calculation unit that calculates
A similarity determination device comprising: The similarity calculation unit
comprehensively comparing one or more first sentence vectors, which are the sentence vectors obtained from the first sentence, and one or more second sentence vectors, which are the sentence vectors obtained from the second sentence; a sentence vector comparison unit that calculates the similarity between the sentence vectors and calculates a first index value representing the similarity between the first sentence and the second sentence based on the similarity between the sentence vectors;
The key-phrase that matches between the key-phrase obtained from the first sentence and the key-phrase obtained from the second sentence occurs in each of the first sentence and the second sentence. a key phrase comparison unit that calculates a second index value representing the degree of similarity between the first sentence and the second sentence based on the degree of appearance;
a total similarity calculation unit that calculates a total similarity between the first sentence and the second sentence based on the first index value and the second index value,
The similarity determination device according to claim 1. The sentence vector comparison unit calculates the first index value based on the number of pairs of sentence vectors having a similarity equal to or higher than a threshold among the pairs of sentence vectors.
The similarity determination device according to claim 2. The sentence vector comparison unit calculates the first index value based on a maximum value of similarities between the sentence vectors.
4. The similarity determination device according to claim 2 or 3. The key phrase comparison unit normalizes the number of occurrences of the matching key phrase in the first sentence by a first coefficient based on the length of the first sentence, and the number of occurrences of the matching key phrase in the second sentence. Calculate the number of occurrences as the second index value based on a value normalized by a second coefficient based on the length of the second sentence;
The similarity determination device according to any one of claims 2 to 4. Further comprising a display control unit for displaying a graph representing a similarity degree of appearance between the first sentence and the second sentence of the key phrase acquired by the key phrase acquisition unit, on the terminal device comprising the display unit. ,
The similarity determination device according to any one of claims 1 to 5. a similarity determination device according to any one of claims 1 to 6;
A terminal device that displays the total similarity calculated by the similarity calculation unit of the similarity determination device,
Similarity determination system. the computer
Obtaining a sentence vector, which is a feature vector for each sentence obtained by decomposing the first sentence, which is the original sentence to be analyzed, and the second sentence, which is the sentence to be compared, and
Obtaining key phrases, which are important elements constituting a sentence, from each of the first sentence and the second sentence,
With respect to the first sentence and the second sentence, a total similarity between the first sentence and the second sentence is calculated based on the similarity between the sentence vectors and the similarity of appearance of the same key phrase. do,
Similarity determination method. to the computer,
Acquire a sentence vector, which is a feature vector for each sentence obtained by decomposing the first sentence, which is the sentence to be analyzed, and the second sentence, which is the sentence to be compared,
Acquiring key phrases, which are important elements constituting a sentence, from each of the first sentence and the second sentence;
With respect to the first sentence and the second sentence, a total similarity between the first sentence and the second sentence is calculated based on the similarity between the sentence vectors and the similarity of appearance of the same key phrase. let
program.

Images