Relevance analysis system, relevance analysis method, and program

Abstract

The information based on the analysis results of the semantic relevance of image components and text components is output. [Solution] An image component acquisition unit (101) of a relevance analysis system (1) acquires image components that are elements that make up an image. A text component acquisition unit (102) acquires text components that are elements that make up text corresponding to an image. A relevance analysis unit (104) analyzes the semantic relevance of the image components and text components based on natural language processing. An output unit (105) outputs information based on the relevance analysis results.

Description

[Technical Field]

[0001] The present disclosure relates to a relevance analysis system, a relevance analysis method, and a program. [Background technology]

[0002] Techniques for analyzing documents based on natural language processing have been known for some time. For example, Patent Document 1 describes a program that performs natural language processing on claims included in patent documents (e.g., application documents or publications) related to patents, thereby acquiring combinations of keywords having technical meanings for each element of the claims, which are elements that make up the claims. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-079142 Summary of the Invention [Problem to be solved by the invention]

[0004] However, while the technology of Patent Document 1 can analyze claims, it cannot analyze drawings included in patent documents. Therefore, the technology of Patent Document 1 cannot analyze the semantic relationship between claim components and drawing components (e.g., symbols, numbers, or annotations) that constitute drawings. This point is not limited to patent documents, but also applies to other documents (e.g., documents related to utility models, contracts, or technical documents). Conventional technologies cannot analyze the semantic relationship between text components that constitute text and image components that constitute images, and output information based on the analysis results.

[0005] One of the objectives of the present disclosure is to output information based on the results of an analysis of the semantic relevance of image components and text components. [Means for solving the problem]

[0006] The relevance analysis system according to the present disclosure includes an image component acquisition unit that acquires image components, which are elements that make up an image; a text component acquisition unit that acquires text components, which are elements that make up text corresponding to the image; a relevance analysis unit that analyzes the semantic relevance between the image components and the text components based on natural language processing; and an output unit that outputs information based on the relevance analysis results. [Effects of the Invention]

[0007] The present disclosure outputs information based on the results of analyzing the semantic relevance of image components and text components. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating an example of a hardware configuration of a relevance analysis system. [Figure 2] FIG. 1 is a diagram illustrating an example of functions realized by a relevance analysis system. [Figure 3] FIG. 10 is a diagram illustrating an example of a method for acquiring image components. [Figure 4] FIG. 10 is a diagram showing an example of a method for searching specification components. [Figure 5] FIG. 10 is a diagram showing an example of a method for acquiring claim components. [Figure 6] FIG. 10 is a diagram illustrating an example of processing by an output unit. [Figure 7] FIG. 10 is a diagram illustrating an example of processing executed in the relevance analysis system. DETAILED DESCRIPTION OF THE INVENTION

[0009] [1. Hardware configuration of relevance analysis system] An example of an embodiment of a relevance analysis system, a relevance analysis method, and a program according to the present disclosure will be described. Fig. 1 is a diagram showing an example of a hardware configuration of a relevance analysis system. For example, the relevance analysis system 1 includes a server 10 and a user terminal 20. Each of the server 10 and the user terminal 20 is connected to a network N such as the Internet or a LAN.

[0010] The server 10 is a server computer. For example, the server 10 includes a control unit 11, a storage unit 12, and a communication unit 13. The control unit 11 includes at least one processor. The storage unit 12 includes at least one of a volatile memory such as RAM and a non-volatile memory such as a flash memory. The communication unit 13 includes at least one of a communication interface for wired communication and a communication interface for wireless communication.

[0011] The user terminal 20 is a user's computer. For example, the user terminal 20 is a personal computer, a smartphone, a tablet, or a wearable terminal. The user terminal 20 includes a control unit 21, a memory unit 22, a communication unit 23, an operation unit 24, and a display unit 25. The hardware configurations of the control unit 21, the memory unit 22, and the communication unit 23 may be similar to those of the control unit 11, the memory unit 12, and the communication unit 13, respectively. The operation unit 24 is an input device such as a keyboard. The display unit 25 is a display such as a liquid crystal or organic electroluminescence (EL) display.

[0012] The program stored in the storage units 12, 22 may be supplied to the server 10 or the user terminal 20 via the network N. Also, at least one of a reading unit (e.g., a memory card slot) that reads a computer-readable information storage medium and an input / output unit (e.g., a USB port) that inputs and outputs data to and from an external device may be included in the server 10 or the user terminal 20. For example, a program stored in an information storage medium may be supplied to the server 10 or the user terminal 20 via at least one of the reading unit and the input / output unit.

[0013] Furthermore, the relevance analysis system 1 may include at least one computer. The computers included in the relevance analysis system 1 are not limited to the example of FIG. 1. For example, the relevance analysis system 1 may include only the server 10. In this case, the user terminal 20 exists outside the relevance analysis system 1. The relevance analysis system 1 may include the server 10 and another computer not shown in FIG. 1. The relevance analysis system 1 may include only another computer not shown in FIG. 1.

[0014] [2. Overview of the relevance analysis system] In this embodiment, the server 10 analyzes a document containing images and text. The images and text to be analyzed do not have to be included in a single document. That is, the images and text do not have to be collected into a single document.

[0015] An image is a picture that shows some meaningful content. An image may show content written in natural language. For example, an image may show content related to at least a portion of the text to be analyzed together with the image. An image may show supplementary explanation of at least a portion of the text. An image may be a drawing, a table, a mathematical formula, a chemical formula, or a combination thereof. The data format of an image may be any format. For example, the data format of an image may be JPEG format, GIF format, PDF format, or another format. When letters, numbers, or other symbols are shown in an image, the letters, numbers, or other symbols are shown only as images. Although information in the form of letters, numbers, or other symbols (information as text) may be included in an image, in this embodiment, such information is not included in the image.

[0016] Text is a string of characters that indicates some meaningful content. Text can also be considered a sentence written in a natural language. For example, text may be letters, numbers, other symbols, or a combination of these. Text may be of any length. That is, text may be any number of characters. Text may indicate any content. For example, text may indicate content related to an image, or may not specifically indicate content related to the image. Text may be included in the same document as the image, or may not be included in the same document as the image. Text may not be included in a specific document. Similarly, images may not be included in a specific document.

[0017] A document includes at least one image and at least one piece of text. The number of images included in a document may be any number. The number of texts included in a document may also be any number. In this embodiment, a Japanese document is used as an example, but the document may be in a language other than Japanese. The document may be in any language. For example, the document may be in English, Chinese, Korean, French, German, Spanish, or another language. Hereinafter, the term "document" simply refers to the document to be analyzed. The term "image" simply refers to the image included in the document to be analyzed. The term "text" simply refers to the text included in the document to be analyzed.

[0018] In this embodiment, a patent or utility model document will be described as an example of a document. A patent or utility model document is a document that describes content related to a patent or utility model. In other words, a patent or utility model document is a document that describes an invention or idea. A patent or utility model document may be in any format. For example, a patent or utility model document may be a gazette issued by a public institution such as a patent office or WIPO, a file distributed by a private intellectual property-related service, a web page displayed in a browser, a document created with application software, or a draft document before filing. A patent or utility model document may also indicate pre-filing or pre-registration content rather than the content that has been applied for or registered.

[0019] In this embodiment, a drawing of a patent or utility model will be described as an example of an image. When a patent or utility model document includes multiple drawings, each of the multiple drawings corresponds to an image. For example, if a patent or utility model document includes 10 drawings, the document includes 10 images. Images are not limited to drawings. Images may also be other images included in the patent or utility model document. For example, an image may be a table, a mathematical formula, or a chemical formula. When a table, a mathematical formula, or a chemical formula corresponds to an image, the image may be inserted in the description, claims, or abstract.

[0020] In this embodiment, the claims of a patent or utility model are described as an example of text. The text may be the entire claim or a part of the claim. For example, if the claim includes multiple claims, the entire multiple claims may correspond to the text, or only some of the claims may correspond to the text. The text is not limited to the claims. The text may also be other text included in the document of the patent or utility model. For example, the text may be the specification or the abstract.

[0021] For example, the server 10 performs image analysis on drawings included in a document to obtain information such as symbols shown in the drawings. The server 10 obtains an explanation corresponding to the information such as symbols from a specification included in the same document. The server 10 associates the information such as symbols with the explanation corresponding to the information. Based on these associations, the server 10 analyzes the semantic association between the information such as symbols and the elements constituting the claims. The server 10 provides the information based on the analysis results of the association to the user terminal 20, thereby providing the information to the user. The relevance analysis system 1 will be described in detail below.

[0022] [3. Functions realized by the relevance analysis system] 2 is a diagram showing an example of functions realized by the relevance analysis system 1. For example, the server 10 includes a data storage unit 100, an image component acquisition unit 101, a description component search unit 102, a text component acquisition unit 103, a relevance analysis unit 104, and an output unit 105. The data storage unit 100 is realized by the memory unit 12. Each of the image component acquisition unit 101, the description component search unit 102, the text component acquisition unit 103, the relevance analysis unit 104, and the output unit 105 is realized by the control unit 11.

[0023] [Data storage section] The data storage unit 100 stores various data necessary for the processing described in this embodiment. For example, the data storage unit 100 stores a document database in which patent or utility model documents are stored. In this embodiment, an example is given in which patent or utility model documents in Japan are stored in the document database, but patent or utility model documents in countries or regions other than Japan may also be stored in the document database. The document database may simply store data of documents to be analyzed.

[0024] For example, the document database stores data on each of a plurality of documents included in a patent or utility model document. In this embodiment, the document database stores, for each patent or utility model document, data on drawings, which are an example of an image, and data on claims, which are an example of text. These data may be in any format. The document database may also store data on specifications. The document database may also store data on documents other than drawings and claims (e.g., abstracts, applications, petitions, amendments, or opinions). The text to be analyzed may be from these other documents.

[0025] The data stored in the data storage unit 100 is not limited to a document database. The data storage unit 100 may store other data necessary for other processes described in this embodiment. For example, the data storage unit 100 may store an image analysis program, actual AI data (described later), training data used in AI learning, or other data.

[0026] [Image component acquisition section] The image component acquisition unit 101 acquires image components, which are elements that make up an image. An image component is all or part of an image. An image component can also be referred to as an object shown in an image. An image component may be information in a text format that allows natural language processing. For example, an image component may be a character, a number, another symbol, a figure (e.g., a line, an arrow, or a polygon), a graph, a table, a photograph, a CG (Computer Graphic), a hand-drawn illustration, or an icon.

[0027] In this embodiment, since the drawings of a patent or utility model correspond to images, the image component acquisition unit 101 acquires drawing components, which are elements that make up the drawings, as image components. Drawing components are an example of image components. Drawing components indicate some content related to an invention or device. In other words, drawing components have some meaning related to the invention or device. For example, drawing components may be symbols, annotations, numbers, or figures. Annotations are strings of natural language characters shown on the drawings. Annotations can also be considered explanations related to the invention or device. Numbers are identification numbers assigned to drawings (e.g., numbers such as Figure 1, Figure 2, etc.). Figures are lines, arrows, or polygons used to explain the invention or device. Figures may indicate the exterior or interior of the invention or device.

[0028] FIG. 3 is a diagram showing an example of a method for acquiring image components. As shown in FIG. 3, the image component acquisition unit 101 acquires at least one drawing included in a document to be analyzed from a document database. In the example of FIG. 3, the image component acquisition unit 101 acquires multiple drawings, but the image component acquisition unit 101 may acquire only one drawing. For example, the image component acquisition unit 101 performs image analysis of the drawing to acquire each of the symbols and annotations written on the drawing as drawing components. The image component acquisition unit 101 may also acquire other drawing components such as numbers by performing image analysis of the drawing.

[0029] In this embodiment, an example is given in which the image component acquisition unit 101 acquires drawing components by performing optical character recognition on a drawing. A known algorithm may be used for the optical character recognition. For example, the image component acquisition unit 101 may acquire drawing components by performing optical character recognition using template matching or AI (artificial intelligence). When acquiring a figure or the like that cannot be recognized by optical character recognition as a drawing component, the image component acquisition unit 101 may acquire the drawing component by template matching based on a template of a sample figure or the like, or by AI that has learned training data of various figures or the like.

[0030] There are various definitions of AI, and the AI ​​of this embodiment may be AI identified by various known definitions. For example, the AI ​​may be a large-scale language model (e.g., a Transformer-based model such as GPT (Generative Pre-trained Transformer) or BERT (Bidirectional Encoder Representations from Transformers)), a machine learning model not classified as a large-scale language model (e.g., a neural network), a rule-based model not classified as a machine learning model, or other models. In this embodiment, an example is given in which the AI ​​is stored in the data storage unit 100, but the AI ​​may also be stored in an external system that cooperates with the relevance analysis system 1. If the AI ​​is stored in an external system, the server 10 may use the AI ​​by communicating with the external system.

[0031] In this embodiment, a large-scale language model in which various documents have been trained corresponds to the AI. The AI ​​includes a program that indicates a series of processes such as calculation of embedded expressions, and parameters (e.g., weighting coefficients or biases) referenced by the program. The embedded expressions are information that indicates the meaning of natural language. For example, the embedded expressions may be multidimensional vectors. The internal structure of the AI ​​may be the same as that of known AI. The AI ​​may be fine-tuned for the relevance analysis system 1, or may be a general-purpose AI that is not particularly fine-tuned.

[0032] In this embodiment, a case where a patent or utility model document is analyzed is taken as an example, and therefore, it is assumed that the AI ​​has learned training patent or utility model documents. For example, the AI ​​may have learned training claims, specifications, abstracts, petitions, procedural amendments, or other documents. The AI ​​learning may be performed by the server 10 or another computer.

[0033] For example, various training images may be learned by the AI ​​as training data. If the AI ​​is a large-scale language model, the AI ​​is assumed to be a multimodal type capable of processing not only natural language documents but also images. The image component acquisition unit 101 inputs to the AI ​​a prompt indicating that the AI ​​is to acquire drawing components from a drawing (for example, a sentence such as "You are an AI that acquires drawing components from a drawing. Please acquire drawing components from the drawing input to you.") and the drawing itself.

[0034] The prompt used to acquire the drawing components may be stored in advance in the data storage unit 100, or may be input by the user or another person. The prompt used to acquire the drawing components may be any sentence indicating that the AI ​​will acquire the drawing components from the drawing. The prompt used to acquire the drawing components may specifically indicate what in the drawing to acquire as the drawing components, such as "Please acquire symbols as drawing components" or "Please acquire annotations as drawing components."

[0035] For example, the AI ​​calculates an embedded representation of a prompt input to itself based on parameters adjusted by pre-learning. The AI ​​may also calculate an embedded representation of a drawing. The AI ​​acquires drawing components from the drawing input to itself based on the embedded representation. In this embodiment, the AI ​​acquires drawing components by performing optical character recognition on the drawing. The optical character recognition program may be built into the AI ​​or may be an external program available to the AI. The AI ​​outputs drawing components indicating the results of reading by optical character recognition. The image component acquisition unit 101 acquires the drawing components output from the AI.

[0036] For example, when the image component acquisition unit 101 acquires a drawing component from a drawing, it assigns a drawing component ID for identification to the drawing component. The image component acquisition unit 101 generates drawing component data DT1 in which the drawing component ID, the drawing number of the drawing in which the drawing component is shown, and the drawing component are associated with each other. The drawing number information may be associated with the drawing in advance or may be acquired by optical character recognition. The image component acquisition unit 101 records the drawing component data DT1 in the data storage unit 100.

[0037] In addition, the prompt used to acquire the drawing component may indicate that the figure number of the drawing containing the drawing component should be output. The prompt used to acquire the drawing component may indicate that a drawing component ID should be assigned. Based on the prompt, the AI ​​may output drawing component data DT1 indicating the combination of the drawing component ID, figure number, and drawing component. The drawing component ID and figure number may be acquired by information processing performed by the image component acquisition unit 101, rather than being output by the AI.

[0038] Furthermore, if an AI specialized in acquiring drawing components is prepared, no particular prompt may be input. In this case, when a drawing to be analyzed is input, the AI ​​acquires drawing components from the drawing. The image component acquisition unit 101 may acquire drawing components from a drawing by executing an optical character recognition program without using an AI. When acquiring a shape such as an arrow as a drawing component, the image component acquisition unit 101 may acquire the drawing component by using template matching or AI without using optical character recognition.

[0039] In addition, the image component acquisition unit 101 may acquire drawing components without performing image analysis such as optical character recognition on the drawing. For example, if a symbol explanation is provided at the end of the specification, the image component acquisition unit 101 may acquire the symbol explanation as a drawing component. If the symbol explanation indicates a pair of a number indicating the symbol and a character string representing the configuration indicated by the symbol, the image component acquisition unit 101 may acquire only the symbol in the symbol explanation as a drawing component, or may acquire only the character string in the symbol explanation as a drawing component, or may acquire both of them as drawing components.

[0040] Furthermore, the image component acquisition unit 101 may acquire drawing components from parts of the specification other than the symbol explanations. For example, the image component acquisition unit 101 may acquire drawing components by acquiring at least one of a symbol and a character string representing the configuration indicated by the symbol from the specification. The image component acquisition unit 101 may acquire drawing components from the specification based on a known specification analysis tool, or may acquire drawing components from the specification based on AI.

[0041] Furthermore, the image components acquired by the image component acquisition unit 101 are not limited to drawing components. The image component acquisition unit 101 may acquire image components from images other than drawings. For example, the image component acquisition unit 101 may acquire image components by performing optical character recognition on a table, mathematical formula, or chemical formula included in a patent or utility model document. The image component acquisition unit 101 may acquire image components from a table, mathematical formula, or chemical formula based on a method other than optical character recognition (e.g., template matching or AI).

[0042] [Specification Element Search Section] The specification component search unit 102 searches the specification of a patent or utility model for specification components that are elements that make up the specification and that correspond to drawing components, and associates the drawing components and specification components.

[0043] A specification element is all or part of a specification. A specification element can also be referred to as a description of an invention or device. A specification element can be a sentence, or a word or phrase that is not a sentence. A specification element can include multiple sentences or multiple paragraphs of a specification. A specification element can be multiple contiguous parts of a specification, or multiple non-contiguous parts. For example, a specification element can be letters, numbers, other symbols, a table, a mathematical formula, or a chemical formula. A specification element is information in text format that is amenable to natural language processing.

[0044] A specification element corresponding to a drawing element is a specification element that provides an explanation of the drawing element. In other words, a specification element corresponding to a drawing element can also be said to be a specification element that has the meaning indicated by the drawing element. For example, if a symbol, annotation, or number corresponds to a drawing element, the description in the specification that includes the symbol, the description that includes the annotation, or the description that includes the number corresponds to the specification element corresponding to the drawing element. If a shape such as a line corresponds to a drawing element, the description in the specification that indicates the meaning of the shape corresponds to the specification element corresponding to the drawing element.

[0045] FIG. 4 is a diagram showing an example of a method for searching specification components. As shown in FIG. 4, the specification component search unit 102 acquires the specification contained in the document to be analyzed from the document database. In this embodiment, an example is given in which the entire specification is the subject of the search, but only a portion of the specification (for example, only some items such as "Problem to be solved by the invention" or "Mode for carrying out the invention") may also be the subject of the search. For each drawing component, the specification component search unit 102 searches the specification for a specification component corresponding to that drawing component.

[0046] For example, suppose that a symbol or annotation that can be keyword searched corresponds to a drawing component. For each drawing component, the specification component search unit 102 performs a keyword search on the specification using the drawing component as a query, and acquires sentences that are hit by the keyword search as specification components corresponding to the drawing component. The keyword search may be the same as a known method. For example, the keyword search may be an exact match, a partial match, or a fuzzy search. The specification component search unit 102 may acquire one sentence as one specification component, or may acquire multiple sentences together as one specification component.

[0047] For example, when the specification component search unit 102 acquires a specification component corresponding to each drawing component, it associates the specification component with the drawing component ID of the drawing component and stores the specification component and specification portion identification information for identifying the portion in which the specification component is described in the drawing component data DT1, as shown in FIG. 4. In the example of FIG. 4, the specification portion identification information is a paragraph of the specification. The specification portion identification information may be information other than a paragraph of the specification. The specification portion identification information may be information that can identify a specific portion of the specification. For example, the identification information may be a page number, column number, line number, or other information of the specification.

[0048] Note that the search method used by the specification component search unit 102 is not limited to keyword search. The specification component search unit 102 may search for specification components corresponding to drawing components from within the specification based on search methods other than keyword search. For example, the specification component search unit 102 may search for specification components corresponding to drawing components from within the specification based on AI. The AI ​​used to search for specification components may be the same as or different from the AI ​​used to acquire drawing components. The AI ​​used to search for specification components may be fine-tuned for searching for specification components.

[0049] For example, let us assume that the AI ​​corresponds to a large-scale language model that has been trained using various training specifications as training data. The specification component search unit 102 inputs to the AI ​​a prompt indicating that the AI ​​is to search for specification components corresponding to drawing components from the specification (for example, a sentence such as "You are an AI that searches for specification components corresponding to drawing components from the specification. Please search for specification components corresponding to drawing components based on the specification and drawing components that you input."), as well as the specification and the drawing components.

[0050] The prompt used in searching for specification components may be stored in advance in the data storage unit 100, or may be input by the user or another person. The prompt used in retrieving specification components may be any sentence indicating that the AI ​​should search the specification components corresponding to the drawing components from the specification. The prompt used in retrieving specification components may specifically indicate what specification components from the specification should be retrieved, such as "Please search for the sentences of the specification components corresponding to the drawing components."

[0051] For example, the AI ​​calculates embedded representations of the prompt, specification, and drawing components input to it based on parameters adjusted by pre-learning. The AI ​​outputs search results for specification components corresponding to the drawing components based on the embedded representations. The specification component search unit 102 acquires the specification components output from the AI. If the prompt indicates that paragraphs of the specification components should also be output, the AI ​​also outputs the paragraphs of the specification components based on the embedded representations. The specification component search unit 102 acquires the paragraphs output from the AI.

[0052] For example, when the specification component search unit 102 acquires a specification component corresponding to a drawing component, it stores the paragraph and specification component output from the AI ​​in the drawing component data DT1 in association with the drawing component ID of the drawing component. In the example of Fig. 4, the specification component search unit 102 associates the drawing component with the specification component corresponding to the drawing component by combining the paragraph and specification component output from the AI ​​with the drawing component data DT1 generated by the image component acquisition unit 101.

[0053] Note that if an AI specialized in searching for specification components is prepared, no particular prompt need be entered. In this case, when the specification to be searched and the query drawing component are input, the AI ​​searches the specification for specification components corresponding to the drawing component and outputs them together with the paragraph. The specification component search unit 102 may search for specification components corresponding to the drawing component based on a vector expression indicating the meaning of the drawing component and a vector expression indicating the meaning of the specification, rather than using AI. In this case, the specification component search unit 102 may search for specification components corresponding to the drawing component based on the vector expression, for specification components having a similar meaning to the drawing component.

[0054] [Text component acquisition section] The text component acquisition unit 103 acquires text components that are elements that make up text corresponding to an image. Text corresponding to an image is text that is analyzed together with the image to be analyzed. In this embodiment, an example is given in which text included in the same document as the image corresponds to the text corresponding to the image, but text that has been previously associated with the image may also correspond to the text corresponding to the image. The image and text do not need to be included in the same document.

[0055] A text component is all or part of a text. For example, a text component may be a letter, a number, another symbol, or a combination of these contained in the text. A text component may have the same meaning as or a meaning unrelated to the image, or may have a meaning unrelated to the image. The semantic relevance of the image and the text component is analyzed by the relevance analysis unit described below.

[0056] In this embodiment, since the claims of a patent or utility model correspond to text, the text component acquisition unit 103 acquires claim components, which are elements that make up the claims, as text components. Claim components are an example of text components. Claim components indicate some content related to an invention or device. In other words, claim components have some meaning related to an invention or device. A claim component may be the entirety of a single claim, or a part of a single claim. If the claims include multiple claims, a claim component may be the entirety of multiple claims.

[0057] For example, a claim element may be a sentence, or a word or phrase that is not specifically a sentence. A claim element may include multiple sentences. A claim element may be multiple contiguous portions of a claim, or multiple non-contiguous portions. For example, a claim element may be letters, numbers, other symbols, tables, mathematical formulas, or chemical formulas. A claim element may be a portion of a claim called a feature.

[0058] Fig. 5 is a diagram showing an example of a method for acquiring claim components. As shown in Fig. 5, the text component acquisition unit 103 acquires claims included in a document to be analyzed from a document database. The text component acquisition unit 103 only needs to acquire at least one claim described in the claims. In this embodiment, an example is given in which the text component acquisition unit 103 acquires all claims included in the claims, but it is also possible to acquire only some of the claims included in the claims.

[0059] For example, the text component acquisition unit 103 may acquire claim components based on line break positions in a claim. In this case, the text component acquisition unit 103 acquires the portion from the beginning of the claim to the first line break position as the first claim component. The text component acquisition unit 103 acquires the portion from the first line break position to the next line break position as the second claim component. Similarly, the text component acquisition unit 103 may acquire the portion from one line break position to the next line break position as a claim component.

[0060] The method by which the text component acquisition unit 103 acquires claim components is not limited to the method based on line break positions. When the entire claim corresponds to a claim component, the text component acquisition unit 103 may acquire all claims included in the claim as a single claim component. When the entire claim corresponds to a claim component, the text component acquisition unit 103 may acquire that single claim as a claim component. The text component acquisition unit 103 may also perform morphological analysis on the claim and acquire individual morphemes as claim components.

[0061] For example, the text component acquisition unit 103 may acquire claim components from the specification based on AI. The AI ​​used to acquire the claim components may be the same as or different from at least one of the AI ​​used to acquire drawing components and the AI ​​used to search for specification components. The AI ​​used to acquire the claim components may be fine-tuned for acquiring the claim components.

[0062] For example, suppose that the AI ​​is a large-scale language model that has learned various training claims as training data. The text component acquisition unit 103 inputs to the AI ​​a prompt indicating that it will acquire claim components from the claims (for example, a sentence such as "You are an AI that searches for claim components from the claims. Please acquire claim components based on the claims you entered.") and the claims.

[0063] The prompt used to acquire the claim components may be stored in advance in the data storage unit 100, or may be input by the user or another person. The prompt used to acquire the claim components may be any sentence indicating that the AI ​​will acquire the claim components from the claims. The prompt used to acquire the claim components may specifically indicate what elements of the claims to acquire as the claim components, such as "Please acquire the constituent features of each claim as the claim components."

[0064] For example, the AI ​​calculates the embedded representation of the prompt and claim input to itself based on parameters adjusted by pre-learning. The AI ​​outputs the claim components based on the embedded representation. The text component acquisition unit 103 acquires the claim components output from the AI. If the prompt indicates that claim identification information (e.g., claim number) that can identify the claim components should also be output, the AI ​​also outputs the claim identification information.

[0065] For example, when the text component acquisition unit 103 acquires a claim component from a claim, it assigns a claim component ID for identification to the claim component. The text component acquisition unit 103 generates claim component data DT2 in which the claim component ID, the claim identification information of the claim component (claim number in the example of FIG. 5), and the claim component are associated with each other. The claim number may be identified based on a predetermined mark such as inked parentheses. The text component acquisition unit 103 records the claim component data DT2 in the data storage unit 100.

[0066] In Japan, the claims and the specification are separate documents, but the specification may contain the same content as the claims. In this case, the text component acquisition unit 103 may acquire the claim components from the specification. The claims may also be contained in other documents, such as a written amendment or a written petition. In this case, the text component acquisition unit 103 may acquire the claim components from other documents, such as a written amendment or a written petition. Similarly, for countries or regions other than Japan, the text component acquisition unit 103 may acquire the claim components in a manner appropriate to the format of the other country or region. This also applies to the image component acquisition unit 101 and the specification component search unit 102.

[0067] [Relevance Analysis Department] The relevance analysis unit 104 analyzes the semantic relevance of image components and text components based on natural language processing. Natural language processing is information processing that analyzes information written in natural language. The natural language processing performed by the relevance analysis unit 104 may be any processing that can analyze the semantic relevance between one piece of information and another piece of information. For example, the natural language processing may be processing that analyzes relevance using vector representation, processing that uses AI, which is a large-scale language model, syntactic analysis, matching of domain-specific vocabulary, or other processing. The relevance analysis unit 104 may analyze the semantic relevance of image components and text components based on a known algorithm used in semantic relevance analysis.

[0068] The semantic relevance between an image component and a text component is the relevance between the meaning of the image component and the meaning of the text component. The semantic relevance between an image component and a text component can also be expressed as whether the image component and the text component are semantically similar. When an image component indicates content related to a text component, this corresponds to the existence of a semantic relevance between the image component and the text component. Conversely, when a text component indicates content related to the image component, this may correspond to the existence of a semantic relevance between the image component and the text component. The relevance may be expressed as a binary value indicating the presence or absence of relevance, or may be expressed as information with three or more levels indicating the degree (level) of relevance. For example, the relevance may be expressed by letters, numbers, other symbols, or a combination of these.

[0069] In this embodiment, drawing components correspond to image components, and claim components correspond to text components, so the relevance analysis unit 104 analyzes the relevance between drawing components and claim components based on natural language processing. When there are multiple combinations of drawing components and claim components, the relevance analysis unit 104 may analyze the relevance of all combinations, or may analyze the relevance of some combinations. The some combinations to be analyzed may be determined randomly.

[0070] In this embodiment, a description component corresponding to a drawing component is searched for, and the relevance analysis unit 104 analyzes the relevance between the drawing component and the claim component based on natural language processing and the description component associated with the drawing component. For example, the relevance analysis unit 104 analyzes the semantic relevance between each drawing component and the claim component based on natural language processing. A description component associated with an image component indicating content related to a text component corresponds to a semantic relevance between the image component and the text component. Conversely, a text component indicating content related to the description component associated with the image component may correspond to a semantic relevance between the image component and the text component.

[0071] For example, the relevance analysis unit 104 analyzes the relevance between the drawing components and the claim components by calculating a matching score related to the semantic match between the description components associated with the drawing components and the claim components based on natural language processing. That is, an example will be given in which the calculation of the matching score by the relevance analysis unit 104 corresponds to analyzing the relevance between the drawing components and the claim components.

[0072] In this embodiment, a case where the matching score is expressed numerically will be taken as an example. For example, the higher the matching score between a drawing component and a claim component, the more semantically related the drawing component and the claim component are. In other words, the higher the matching score, the more similar the meaning of the drawing component and the meaning of the claim component are. The calculation method of the matching score may be the same as a known calculation method. The relevance analysis unit 104 may calculate the matching score between the drawing component and the claim component based on a known calculation method.

[0073] It should be noted that there may be a correlation between the matching score and the relevance, and the lower the matching score between the drawing element and the claim element, the more semantically related the drawing element and the claim element may be. The matching score may also be expressed in a form other than numbers. The matching score may be expressed by letters or other symbols, or a combination thereof.

[0074] For example, the relevance analysis unit 104 may calculate a matching score based on a vector expression indicating the meaning of a description component associated with a drawing component and a vector expression indicating the meaning of a claim component. These vector expressions are calculated based on natural language processing, so that the relevance analysis unit 104's calculation of a matching score based on the vector expressions corresponds to analyzing relevance based on natural language processing.

[0075] The calculation method of the vector expression may be the same as a known calculation method. The vector expression may be calculated by the relevance analysis unit 104, or may be calculated in advance and stored in the data storage unit 100. That is, the data storage unit 100 may store a vector expression database in which vector expressions calculated from patent or utility model documents are stored. The relevance analysis unit 104 may acquire the vector expression from the vector expression database. The vector expression database may be stored in a computer or information storage medium other than the server 10.

[0076] For example, the relevance analysis unit 104 calculates the matching score so that the closer the distance between the vector expression indicating the meaning of the description component associated with the drawing component and the vector expression indicating the meaning of the claim component, the higher the matching score. A calculation formula indicating the relationship between these vector expressions and the matching score is stored in advance in the data storage unit 100. The relevance analysis unit 104 calculates the matching score by substituting these vector expressions into the calculation formula.

[0077] For example, the relevance analysis unit 104 calculates a matching score for each combination of drawing component and claim component to be analyzed, and generates relevance data DT3. In the example of Fig. 5, the relevance analysis unit 104 calculates a matching score for each combination of drawing component ID and claim component ID. The relevance analysis unit 104 generates relevance data DT3 that associates the matching score with the combination. The relevance analysis unit 104 records the relevance data DT3 in the data storage unit 100, associating it with the document to be analyzed.

[0078] Note that the method by which the relevance analysis unit 104 calculates the matching score is not limited to the above example. The relevance analysis unit 104 may calculate the matching score based on some kind of natural language processing. For example, the relevance analysis unit 104 may calculate the matching score based on AI. In this case, the relevance analysis unit 104 inputs to the AI ​​a prompt indicating that the matching score will be calculated (for example, a sentence such as "You are an AI that calculates the matching score between a description component associated with a drawing component and a claim component. Please calculate the matching score based on the description component and claim component associated with the drawing component you entered."), along with the description component and claim component associated with the drawing component.

[0079] For example, the AI ​​calculates embedded representations of the prompt, description components, and claim components input thereto based on parameters adjusted by pre-learning. The AI ​​calculates and outputs a matching score based on the embedded representations. The relevance analysis unit 104 acquires the matching score output from the AI. The relevance analysis unit 104 may input not only the description components associated with the drawing components but also the drawing components to the AI. The AI ​​may also calculate the embedded representations of the drawing components to calculate the matching score. The relevance analysis unit 104 may calculate the matching score by calculating the number of words common to the description components associated with the drawing components and the claim components without using AI. The relevance analysis unit 104 can calculate the matching score based on various methods used in the field of natural language processing as a method for calculating the semantic relevance between certain information and other information.

[0080] Furthermore, the relevance analysis unit 104 may analyze the relevance based on the drawing components and the claim components, rather than based on the specification components associated with the drawing components. In this case, the relevance analysis system 1 may not include the specification component search unit 102. For example, the relevance analysis unit 104 may analyze the relevance by calculating a matching score between a vector expression indicating the meaning of the drawing component and a vector expression indicating the meaning of the claim component. If the drawing component is a symbol, it may be difficult to analyze the relevance because it does not have a linguistic meaning. However, if the drawing component is an annotation, it is possible to analyze the relevance because it has a linguistic meaning.

[0081] Furthermore, the method by which the relevance analysis unit 104 analyzes the relevance is not limited to calculating a matching score. The relevance analysis unit 104 may analyze the relevance by acquiring binary information indicating whether or not there is semantic relevance, instead of a matching score. In this case, the relevance analysis unit 104 may acquire binary information indicating there is relevance when the distance of the above-mentioned vector representation is less than a threshold, and may acquire binary information indicating there is no relevance when the distance is equal to or greater than the threshold.

[0082] [Output section] The output unit 105 outputs information based on the relevance analysis result. Hereinafter, the information output by the output unit 105 will be referred to as output information. The output unit 105 may output the output information as data, or may output the output information as a screen displayed on a computer such as the user terminal 20. The output information may be any information that indicates the relevance analysis result itself, or may be information generated based on the relevance analysis result.

[0083] In this embodiment, since a matching score is calculated as a result of the relevance analysis, the output unit 105 outputs output information based on the matching score. The output unit 105 may output the matching scores themselves of the drawing components and the claim components as output information. The output unit 105 may also generate and output output information based on the matching scores of the drawing components and the claim components.

[0084] For example, the output unit 105 may not generate output information for drawing components and claim components with relatively low matching scores, but may generate output information based on drawing components and claim components with relatively high matching scores. A relatively low matching score means that the matching score is less than a threshold, or that the ranking when sorted in descending order of matching score is equal to or lower than a predetermined rank. A relatively high matching score means that the matching score is equal to or higher than a threshold, or that the ranking when sorted in descending order of matching score is equal to or higher than a predetermined rank.

[0085] In the example of the relationship data DT3 in FIG. 5, suppose the threshold is 50. In this case, the combination of drawing components and claim components with a matching score of 80 has a relatively high matching score. The combination of drawing components and claim components with matching scores of 20, 15, and 14 has a relatively low matching score. Therefore, the output unit 105 outputs output information based on the combination of drawing components and claim components with a matching score of 80, rather than based on the combination of drawing components and claim components with a matching score of 20, 15, and 14.

[0086] 6 is a diagram showing an example of processing by the output unit 105. As shown in FIG. 6, the output unit 105 may output summary information regarding the summary of at least one of the drawing components and the claim components as output information based on the analysis results. For example, the output unit 105 identifies, based on the relevance data DT3, combinations of drawing components and claim components with relatively high matching scores from among combinations of drawing components and claim components for which matching scores have been calculated. The output unit 105 generates summary information for the identified combination.

[0087] For example, the output unit 105 may generate summary information for claim components associated with drawing components included in a combination with a relatively high matching score, based on the drawing components. The output unit 105 generates summary information based on description components associated with the drawing components. The output unit 105 may generate summary information based on an AI. In this case, the output unit 105 inputs to the AI ​​a prompt indicating that the AI ​​will generate a summary of information input to the AI ​​(e.g., a sentence such as "You are an AI that generates a summary. Please generate a summary of the information input to you.") and the description components associated with drawing components included in a combination with a relatively high matching score.

[0088] For example, the output unit 105 may generate summary information of drawing components associated with claim components included in a combination with a relatively high matching score, based on the claim components included in the combination. In this case, the output unit 105 inputs to the AI ​​a prompt indicating that a summary of the information input to the AI ​​will be generated, and the claim components included in the combination with a relatively high matching score.

[0089] The prompt used in generating the summary information may be stored in advance in the data storage unit 100, or may be input by a user or another person. The prompt used in generating the summary information may be any sentence indicating that the AI ​​will generate a summary of the information input to it. The prompt used in generating the summary information may indicate what information will be input to the AI. The prompt may indicate the format of the summary information (for example, a standard format such as "This configuration achieves XX"). The AI ​​may generate the summary information in the format indicated by the prompt.

[0090] For example, the AI ​​calculates the embedded representation of the prompt input to itself and the information that forms the basis of the summary information (e.g., a specification component or a claim component associated with a drawing component) based on parameters adjusted by pre-learning. The AI ​​outputs the summary information based on the embedded representation. The output unit 105 acquires the summary information output from the AI. The output unit 105 records the summary information in the data storage unit 100 and outputs the summary information to the user terminal 20 in response to a request from the user terminal 20. In the example of Figure 6, the output unit 105 outputs summary information data DT4 in which summary information is indicated for each drawing component. The summary information data DT4 may indicate at least one of the drawing component, the specification component, and the claim component.

[0091] In the example of FIG. 6, the output unit 105 outputs, in association with the summary information, basis identification information that can identify the portion of the document that served as the basis for the summary information. For example, the prompt may indicate that claim identification information (e.g., paragraph, figure number, or claim number) that can identify the basis identification information should also be output. The AI ​​acquires not only the summary information but also the basis identification information. The basis identification information may be incorporated as part of the summary information. The output unit 105 records the basis identification information in the data storage unit 100 and outputs the basis identification information to the user terminal 20 in response to a request from the user terminal 20. The output unit 105 may acquire, as basis identification information, the figure number of a drawing component used in generating the summary information, the description portion identification information of a specification component used in generating the summary information, or the claim identification information of a claim component used in generating the summary information.

[0092] The output unit 105 may output output information other than the summary information. For example, the output unit 105 may output output information in which the contents of the relevance data DT3 are displayed in a table format. That is, the output unit 105 may output the relevance data DT3 as output information, rather than the summary information data DT4. The output unit 105 may output a combination of drawing components and claim components with a relatively high matching score as output information. The output unit 105 may output basis identification information (e.g., a drawing number, a description portion identification information, or a claim identification information) along with the combination of drawing components and claim components with a relatively high matching score.

[0093] [4. Processing performed by the relevance analysis system] Fig. 7 is a diagram showing an example of processing executed by the relevance analysis system 1. The processing in Fig. 7 is executed by the control units 11 and 21 executing programs stored in the storage units 12 and 22, respectively. Each step in Fig. 7 is an example of a relevance analysis method.

[0094] As shown in Figure 7, the server 10 acquires drawings of a document to be analyzed from the document database (S1). The server 10 acquires drawing components based on the drawings acquired in S1 (S2). Details of the processes of S1 and S2 are as explained in the processing of the image component acquisition unit 101. The server 10 acquires the specification of the patent or utility model for which drawings were acquired in S1 from the document database (S3). The server 10 searches the specification acquired in S3 for specification components corresponding to the drawing components acquired in S2 (S4). Details of the processes of S3 and S4 are as explained in the processing of the specification component search unit 102.

[0095] The server 10 retrieves from the document database the claims of the patent or utility model for which drawings were retrieved in S1 (S5). The server 10 retrieves claim components based on the claims retrieved in S5 (S6). Details of the processes of S5 and S6 have been described as the processes of the text component retrieval unit 103. Based on the description components searched in S4 and the claim components retrieved in S6, the server 10 analyzes the semantic relevance between the drawing components corresponding to the description components and the claim components (S7). Details of the process of S7 have been described as the processes of the relevance analysis unit 104.

[0096] The server 10 generates summary information based on the analysis results of S7 and outputs summary information data DT4 (S8). The summary information data DT4 may be stored in a document database and provided as needed in response to a user request. The server 10 executes processing between the server 10 and the user terminal 20 to display the summary information output in S8 on the display unit 25 (S9), and this processing ends. Details of the processing of S8 and S9 are as described above for the processing of the output unit 105. For example, the user terminal 20 displays on the display unit 25 the summary information of a patent or utility model document that the user has searched for in the document database.

[0097] [5. Summary of embodiments] The relevance analysis system 1 of this embodiment acquires image components. The relevance analysis system 1 acquires text components. The relevance analysis system 1 analyzes the semantic relevance of image components and text components based on natural language processing. The relevance analysis system 1 outputs information based on the relevance analysis results. This allows the relevance analysis system 1 to analyze the semantic relevance of image components and text components. For example, by analyzing the semantic relevance of image components and text components, the relevance analysis system 1 can detect whether content unrelated to the content of the text is shown in the image, or conversely, whether content unrelated to the image is shown in the text. When a user visually checks the semantic relevance of image components and text components, the user's burden increases and the checking process takes time. However, the relevance analysis system 1 can reduce the user's burden and shorten the time required. For example, each process performed by the relevance analysis system 1 passes the output of the previous step to the next step as context information based on the concept of MPC (Model Context Protocol), thereby improving the accuracy of relevance analysis.

[0098] The text is the claims of a patent or utility model. The image is a drawing of the patent or utility model. The relevance analysis system 1 acquires drawing components as image components. The relevance analysis system 1 acquires claim components as text components. The relevance analysis system 1 analyzes the relevance between the drawing components and the claim components based on natural language processing. This allows the relevance analysis system 1 to analyze the semantic relevance between the drawing components and the claim components. For example, the relevance analysis system 1 can assist a user in confirming whether the drawings and the claims are consistent. The relevance analysis system can provide work support to users, whether they are patent attorneys working at patent offices, patent examiners at patent offices, intellectual property officers at companies, or others. The relevance analysis system 1 can also detect ambiguous or missing descriptions by analyzing the semantic relevance between the drawing components and the claim components, thereby automating quality control of manuscripts before filing.

[0099] The relevance analysis system 1 also searches for specification components corresponding to drawing components in the specification of a patent or utility model and associates the drawing components with the specification components. The relevance analysis system 1 analyzes the relevance between drawing components and claim components based on natural language processing and specification components associated with the drawing components. This allows the relevance analysis system 1 to improve the accuracy of analyzing the semantic relevance between drawing components and claim components. For example, when a symbol corresponds to a drawing component, the symbol itself does not have meaning related to the invention or device, but the relevance analysis system 1 can search for specification components that indicate the content corresponding to the symbol and analyze the relevance based on the specification components. In other words, even if the drawing component itself has no meaning, the specification components corresponding to the drawing component have meaning related to the invention or device, so the relevance analysis system 1 can analyze the relevance.

[0100] Furthermore, the relevance analysis system 1 analyzes the relevance between the drawing components and the claim components by calculating a matching score based on natural language processing, which indicates the degree to which the description components associated with the drawing components and the claim components semantically match. The relevance analysis system 1 outputs information based on the matching score. This allows the relevance analysis system 1 to output information based on a clear indicator called the matching score. For example, the relevance analysis system 1 can evaluate discrepancies or inconsistencies in the description that are difficult to detect manually, based on a clear indicator called the matching score.

[0101] Furthermore, the relevance analysis system 1 outputs summary information relating to the summary of at least one of the drawing components and the claim components based on the analysis results. This allows the relevance analysis system 1 to output summary information that is useful for the user to understand the content of at least one of the drawing components and the claim components. The summary information allows the user to easily understand the content of at least one of the drawing components and the claim components.

[0102] Furthermore, the relevance analysis system 1 outputs, in association with the summary information, basis identification information that can identify the portion of the document that served as the basis for the summary information. This allows the relevance analysis system 1 to output basis identification information that is useful for the user to understand the basis for the summary information. The basis identification information allows the user to confirm the portion that served as the basis for generating the summary information. For example, by checking the portion indicated by the basis identification information, the user can easily confirm whether the summary information is indeed correct.

[0103] Furthermore, the relevance analysis system 1 acquires each of the symbols and annotations written on the drawings as drawing components by performing image analysis of the drawings. This allows the relevance analysis system 1 to analyze the semantic relevance between the symbols and annotations acquired as drawing components and the claim components. For example, while analysis of drawings has not been given much attention in the past, the relevance analysis system 1 can also analyze the semantic content of drawings.

[0104] [6. Modifications] The present disclosure is not limited to the above-described embodiments, and may be modified as appropriate without departing from the spirit of the present disclosure.

[0105] For example, the documents to be analyzed are not limited to patent or utility model documents. The relevance analysis system 1 may analyze documents other than patents or utility models. The relevance analysis system 1 may analyze a contract. The contract includes text indicating the contract content and an image indicating supplementary explanations of the contract content. The relevance analysis system 1 acquires image components of the contract image and text components of the contract text in the same manner as in the embodiment. The relevance analysis system 1 may analyze the semantic relevance of these and output the analysis results.

[0106] For example, the relevance analysis system 1 may analyze technical documents such as papers or design documents. The technical documents include text showing technical content and images showing supplementary explanations of the technical content. The relevance analysis system 1 acquires image components of the images of the technical documents and text components of the text of the technical documents in the same manner as in the embodiment. The relevance analysis system 1 may analyze the semantic relevance of these and output the analysis results. In addition to contracts and technical documents, the relevance analysis system 1 may also analyze internal documents such as internal regulations or manuals, medical documents used in hospitals, teaching materials used in schools, or other documents. The relevance analysis system 1 may analyze documents containing images and text that have a common meaning. The relevance analysis system 1 may also analyze images and text that are not combined into a single document.

[0107] For example, the functions described as being realized by the server 10 may be realized by the user terminal 20 or another computer. The functions described as being realized by the server 10 may be shared among multiple computers. [Explanation of symbols]

[0108] 1 Relevance analysis system, N network, 10 server, 11,21 control unit, 12,22 memory unit, 13,23 communication unit, 20 user terminal, 24 operation unit, 25 display unit, 100 data memory unit, 101 image component acquisition unit, 102 specification component search unit, 103 text component acquisition unit, 104 relevance analysis unit, 105 output unit, DT1 drawing component data, DT2 claim component data, DT3 relevance data, DT4 summary information data.

Claims

1. A drawing component acquisition unit that acquires drawing components that are elements that make up drawings of a patent or utility model; a claim element acquisition unit that acquires claim elements that constitute the claims of the patent or utility model corresponding to the drawings; a relevance analysis unit that analyzes semantic relevance between the drawing components and the claim components based on natural language processing; an output unit that outputs summary information regarding a summary of at least one of the drawing components and the claim components based on the analysis result of the relevance; A relevance analysis system including:

2. The relevance analysis system further includes a specification component search unit that searches the specification of the patent or utility model for specification components that are elements that constitute the specification and that correspond to the drawing components, and associates the drawing components and the specification components, the relevance analysis unit analyzes the relevance between the drawing components and the claim components based on the natural language processing and the specification components associated with the drawing components; The relevance analysis system according to claim 1 .

3. the relevance analysis unit analyzes the relevance between the drawing component and the claim component by calculating a matching score relating to the degree of semantic match between the description component associated with the drawing component and the claim component based on the natural language processing; the output unit outputs the summary information based on the matching score. The relevance analysis system according to claim 2 .

4. the output unit outputs basis identification information that can identify a portion that serves as a basis for the summary information in association with the summary information. The relevance analysis system according to any one of claims 1 to 3.

5. the drawing component acquisition unit acquires each of the symbols and annotations written on the drawing as the drawing components by performing image analysis on the drawing; The relevance analysis system according to any one of claims 1 to 3.

6. The computer Obtaining drawing elements that are elements that make up the drawings of a patent or utility model; Obtain claim elements that constitute the scope of claims of the patent or utility model corresponding to the drawings; analyzing the semantic relationships between the drawing components and the claim components based on natural language processing; outputting summary information relating to a summary of at least one of the drawing components and the claim components based on the analysis result of the relevance; Relevance analysis methods.

7. A drawing component acquisition unit that acquires drawing components that are elements that constitute drawings of a patent or utility model; a claim element acquisition unit for acquiring claim elements that constitute the claims of the patent or utility model corresponding to the drawings; a relevance analysis unit that analyzes semantic relevance between the drawing components and the claim components based on natural language processing; an output unit that outputs summary information regarding a summary of at least one of the drawing components and the claim components based on the analysis result of the relevance; A program that allows a computer to function as a

Images