Patent application document preparation system and patent examination system

Abstract

To provide a patent application document preparation system capable of creating high-quality claims.SOLUTION: A patent application document preparation system 11 comprises: a separation part 23 that separates common elements shared by both the prior art and the present invention from a description of advantages derived from differences between the prior art and the invention, and outputs common element data identifying the common elements; a machine learning part 13 that acquires the common element data from the separation part 23 and describes functions of the common elements and relations among them based on publicly available learning data; a presentation part 26 that acquires common requirement data specifying the functions and relations from the machine learning part 13 and presents the functions and relations so as to be perceptible by human senses; and a description part 28 that, in response to the presentation of the functions and relations, acquires linkage data describing the relations between the common elements and the differences, and creates claims based on the common requirement data and the linkage data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a patent application document drafting system and a patent examination system that utilize machine learning. [Background technology]

[0002] Patent Document 1 discloses a system for preparing patent application documents based on machine learning and rule-based algorithms. When preparing patent application documents, claims are input into the system. The quality of patent application documents is determined by the quality of the claims. Since claims are prepared by people, the quality of patent application documents depends on the skill of the people. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2020-510270 Summary of the Invention [Problem to be solved by the invention]

[0004] In Patent Document 1, machine learning such as deep learning is used in the preparation of patent application documents. For example, with deep learning, the more training data there is, the more likely it is that the quality of patent application documents can be improved. On the other hand, patent application documents must be kept confidential until they are filed. If claims are incorporated into deep learning, which is commonly used in the preparation of patent application documents, questions arise about their confidentiality. To begin with, Patent Document 1 does not use machine learning (artificial intelligence) in the preparation of claims.

[0005] An object of the present invention is to provide a patent application document drafting system that can draft good claims. [Means for solving the problem]

[0006] A patent application document drafting system according to one embodiment of the present invention includes a first acquisition unit that acquires difference data describing differences between the prior art and the invention; a second acquisition unit that acquires advantage data describing advantages resulting from the differences; a separation unit that separates common terms common to the prior art and the invention from the description of advantages based on the difference data and the advantage data and outputs common term data that identifies the common terms; a machine learning unit that acquires the common term data from the separation unit and describes functions of the common terms and relationships between the common terms based on publicly available learning data; a presentation unit that acquires common requirement data that identifies the functions and the relationships from the machine learning unit and presents the functions and the relationships in a manner that is perceptible to human senses; and a description unit that acquires linking data that describes the relationships between the common terms and the differences in response to the presentation of the functions and the relationships, and writes claims based on the common requirement data and the linking data.

[0007] The machine learning unit can identify the functions of common terms and the relationships between them based on accumulated learning data. The constituent elements of the prior art related to the differences can be identified. The constituent elements of the prior art can be presented to the writer (human). The writer can input the relationship between the constituent elements of the prior art and the differences in response to the presentation. In this way, the differences are linked to the common terms. The claims can be finalized based on this linking. Since machine learning is used in writing the claims, the burden on the writer can be reduced. Here, the common terms are separated from the description of advantages during machine learning. Rare "differences" can be excluded from the learning data. The machine learning unit can describe the functions of common terms and the relationships between common terms with high quality.

[0008] A patent application document drafting system according to one embodiment of the present invention includes a first acquisition unit that acquires difference data describing differences between the prior art and the invention; a second acquisition unit that acquires advantage data describing advantages resulting from the differences; a separation unit that separates common terms common to the prior art and the invention from the description of advantages based on the difference data and the advantage data and outputs common term data that identifies the common terms; a machine learning unit that acquires the common term data from the separation unit and describes functions of the common terms and relationships between the common terms based on publicly available learning data; and a description unit that acquires common requirement data that identifies the functions and relationships from the machine learning unit and describes claims by combining the differences with the common terms based on the description of advantages.

[0009] The machine learning unit can identify the functions of common terms and the relationships between them based on accumulated learning data. The constituent elements of prior art related to differences can be grasped. Since the relationship between common terms and differences is identified in the description of advantages, the differences are linked to the common terms. The claims can be finalized according to this linking. Since machine learning is used in writing claims, the burden on the writer can be reduced. Here, common terms are separated from the description of advantages during machine learning. Rare "differences" can be excluded from the learning data. The machine learning unit can describe the functions of common terms and the relationships between them with high quality.

[0010] A patent application document drafting system according to one embodiment of the present invention includes an identification unit that identifies common terms that describe constituent elements of conventional technology and difference terms that describe differences between the conventional technology and the invention; a machine learning unit that acquires common term data that identifies the common terms from the identification unit and identifies constituent elements necessary for the operation of the common terms based on publicly available learning data; a description unit that acquires constituent element data that identifies the constituent elements from the machine learning unit and combines the difference terms with the constituent elements to write claims; and a regulation unit that regulates the output of the difference terms from the identification unit and the description unit.

[0011] Because the output of difference terms (features) from the identifying and describing parts is restricted, the difference terms can be reliably concealed when writing claims. Meanwhile, the constituent elements of common terms can be identified with high quality based on the vast amount of publicly available training data. This improves the quality of claims. In this way, machine learning based on the vast amount of training data can be effectively used when drafting claims while ensuring the confidentiality of the invention.

[0012] A patent examination system according to one embodiment of the present invention comprises an acquisition unit that receives advantage data input from an input device and describes advantages resulting from differences between the prior art and the invention; an analysis unit that compares the advantages with the closest technology and identifies differences between the closest technology and the advantages; a machine learning unit that acquires difference data and advantage data that identify the differences, identifies effects resulting from the differences separated from the closest technology based on publicly available learning data, compares the advantages with the effects, and determines whether the effects are within the expected scope based on the publicly available learning data; and an examination unit that denies the inventive step of the invention when it receives determination data that recognizes the effects within the expected scope from the machine learning unit.

[0013] The machine learning unit identifies the effect produced by the difference based on accumulated learning data. When identifying the effect, the difference is separated from the closest technology, so the effect of the difference generally does not contribute to satisfying the requirement of inventive step (non-obviousness). If an unexpected effect is recognized in the advantage, the inventive step (non-obviousness) of the invention can be affirmed. In this way, the inventive step (non-obviousness) of the invention can be determined by the machine learning unit. The objectivity of the examination can be ensured. Machine learning can be used in the examination of inventive step (non-obviousness).

[0014] A patent examination system according to one embodiment of the present invention includes an acquisition unit that receives advantage data input from an input device and describes advantages resulting from differences between the prior art and the invention; an analysis unit that compares the advantages with the closest technology and identifies differences between the closest technology and the advantages; a machine learning unit that acquires difference data and advantage data that identify the differences, describes a mechanism leading from the differences to the advantages based on publicly available learning data, and determines the feasibility of the mechanism; and an examination unit that affirms the inventive step of the invention when it receives judgment data that recognizes the feasibility of the mechanism from the machine learning unit.

[0015] The machine learning unit describes the mechanism that leads from the difference to the advantage based on accumulated learning data. If the mechanism is established, the advantage can be judged as a reasonable claim. If the advantage is then recognized to have an unexpected effect, the inventive step (non-obviousness) of the invention can be affirmed. Even if the advantage has an unexpected effect, if the mechanism is not established, the advantage can be judged as an irrational claim. The inventive step (non-obviousness) of the invention can be denied.

[0016] A patent examination system according to one embodiment of the present invention comprises a first acquisition unit that acquires difference data describing differences between the prior art and the invention; a second acquisition unit that acquires advantage data describing advantages resulting from the differences; a separation unit that separates common terms shared by the prior art and the invention from the description of advantages based on the difference data and the advantage data and outputs common term data that identifies the common terms; a search unit that acquires the common term data, searches a prior art database based on the common terms, and extracts prior art containing the common terms; and an analysis unit that recognizes the prior art with the largest range of common terms as the closest prior art based on the advantage data.

[0017] The commonalities correspond to the prior art described in the advantage data. If the differences and the advantages resulting from the differences are understood, prior art can be extracted effectively. In this way, the closest prior art can be identified effectively. The burden on the researcher can be reduced because the researcher only needs to input the differences and advantages to extract the closest prior art. Here, machine learning can be used in the search section. Since there is a huge amount of publicly available training data, prior art can be extracted with high quality. Moreover, since only the commonalities need to be passed on to the machine learning, the confidentiality of the invention can be ensured. [Effects of the Invention]

[0018] As described above, the disclosed patent application drafting system allows for the drafting of good claims. Furthermore, the disclosed patent examination system allows for the determination of inventive step based on machine learning. Patent applications can be examined based on machine learning. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a block diagram showing an outline of the configuration of a patent application document drafting system according to a first embodiment of the present invention. [Figure 2] FIG. 1 is a conceptual diagram showing the structure of the invention as applied to plants. [Figure 3] FIG. 10 is a block diagram showing the outline of the configuration of a patent application document drafting system according to a second embodiment of the present invention. [Figure 4] FIG. 10 is a block diagram showing the outline of the configuration of a patent examination system according to a third embodiment of the present invention. [Figure 5] FIG. 1 is a conceptual diagram illustrating a pyramidal structure according to an embodiment. [Figure 6] FIG. 1 is a conceptual diagram showing the relationship between a pyramid structure and paragraphs. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings.

[0021] FIG. 1 shows a schematic configuration of a patent application drafting system 11 according to a first embodiment of the present invention. The patent application drafting system 11 includes a common machine learning unit 13 connected to the Internet 12 and a claim drafting unit 14 also connected to the Internet 12. The common machine learning unit 13 can collect training data from web pages 15 accessed via the Internet 12 and a prior art database 16 connected to the Internet 12. The training data can include text data, image data such as photographs and graphics, audio data, and other data. The prior art database 16 can store, for example, patent publications, patent lay publications, utility model registration publications, research papers, and other data. The region and language are not limited to a single region. The common machine learning unit 13 may support multiple languages with the support of a translation device. The common machine learning unit 13 performs machine learning based on the collected training data. The claim drafting unit 14 prepares claims, which are one element of patent application documents, using the machine learning of the common machine learning unit 13. Generally, patent application documents include an application, one or more claims, a specification or description, drawings, and an abstract. Other public networks may be used instead of the Internet 12. The more training data collected, the more accurate the machine learning can be.

[0022] The claim preparation unit 14 includes a first acquisition unit 21 that acquires difference data describing the differences between the prior art and the invention, and a second acquisition unit 22 that acquires advantage data describing the advantages resulting from the differences. The difference data can be composed of text data input from an input device, for example. Here, the difference is expressed by a single word. The difference only needs to reflect the inventor's subjective opinion. The input device can include a keyboard as well as a voice recognition device that generates text data from voice input from a microphone. The difference may also be expressed in sentences.

[0023] The advantage data may be composed of text data input from an input device, for example. The advantage includes commonalities describing the constituent features of the prior art and difference terms (feature terms or concealed terms) describing the differences between the prior art and the invention. The input device may include a keyboard and a speech recognition device that generates text data from speech input from a microphone. As shown in Figure 2, the invention 101 can be compared to a plant (e.g., corn). In plants, roots are essential for the growth of the stem. In the invention 101, the prior art 103 is essential for the operation of the difference 102. The difference 102 can exert advantage 104 depending on the operation of the prior art 103. The advantage 104 describes the relationship between the difference 102 and the prior art 103. Because the prior art is publicly known, machine learning can develop smoothly based on the use of abundant training data. On the other hand, because the invention 101 is generally not disclosed, the separation of the difference 102 from the prior art 103 is expected to facilitate rapid development of machine learning in the drafting of claims.

[0024] The claim creation unit 14 includes a separation unit 23 connected to the first acquisition unit 21 and the second acquisition unit 22. The separation unit 23 separates common terms common to the prior art and the invention from the description of advantages based on the difference data and advantage data. In separating the common terms, the separation unit 23 removes the differences and the actions linked to the differences. The difference terms may be specified in advance in the advantage data. An input device can be used for the specification. In this way, the separation unit 23 identifies common terms that describe the constituent elements of the prior art and difference terms that describe the differences between the prior art and the invention. The separation unit 23 generates common term data that identifies the common terms. The actions of the common terms may be input from the input device when generating such common term data.

[0025] The separation unit 23 may separate commonalities from descriptions of advantages based on machine learning. To perform the machine learning, an on-premise prior art database is connected to the separation unit 23. The on-premise prior art database is isolated from other machine learning. If the machine learning of the separation unit 23 is separated from the machine learning of the common machine learning unit 13, the output of differences and advantages from the separation unit 23 to the common machine learning unit 13 can be prevented. The common machine learning unit 13 cannot learn information about differences and advantages from the separation unit 23.

[0026] Separation unit 23 passes the common term data to common machine learning unit 13. When passing the common term data, it passes through restriction unit 24. Restriction unit 24 restricts data exchange between separation unit 23 and common machine learning unit 13. Restriction unit 24 prevents output of difference terms from separation unit 23 to common machine learning unit 13. Common machine learning unit 13 cannot learn information about the difference terms from separation unit 23. Restriction unit 24 can achieve security that ensures the confidentiality of the invention.

[0027] The common machine learning unit 13 describes the functions of the common terms and the relationships between the common terms based on the published learning data. In describing the common terms, the common machine learning unit 13 acquires the common term data from the separation unit 23. The common machine learning unit 13 generates common requirement data that identifies the functions of the common terms and the relationships between the common terms. The common requirement data can be composed of text data, for example.

[0028] The claim creation unit 14 includes a presentation unit 26 that acquires common requirement data from the common machine learning unit 13. The presentation unit 26 can display the functions of the common terms and the relationships between the common terms on a display screen based on the common requirement data. Instead of such a visual display, audio may be output from a speaker based on text data. The presentation unit 26 presents the functions of the common terms and the relationships between the common terms in a manner that is perceptible to human senses.

[0029] When acquiring the common requirement data, the common requirement data passes through restriction unit 24. Restriction unit 24 restricts data exchange between presentation unit 26 and common machine learning unit 13. Restriction unit 24 blocks the output of difference terms from presentation unit 26 to common machine learning unit 13. Common machine learning unit 13 cannot learn information about the difference terms from presentation unit 26. Restriction unit 24 can achieve security that ensures the confidentiality of the invention.

[0030] The claim creation unit 14 includes a third acquisition unit 27 that acquires linking data from an input device in response to presentation of the functions of the common terms and the relationships between the common terms. The linking data describes the relationships between the common terms and the differences. The linking data may be composed of text data input from an input device, for example. The input device may include a keyboard as well as a voice recognition device that generates text data from voice input from a microphone.

[0031] The claim creation unit 14 includes a description unit 28 connected to the presentation unit 26 and the third acquisition unit 27. The description unit 28 acquires common requirement data from the presentation unit 26 and acquires linking data from the third acquisition unit 27. The description unit 28 describes claims by combining the functions of common terms, the relationships between the common terms, and the relationships between the common terms and differences. The description unit 28 generates claim data that expresses the claims. The claim data can be composed of, for example, text data. The claim data can be stored in, for example, a storage device. The claims can be displayed on a display screen based on the text data. Instead of such visual display, audio can be output from a speaker based on the text data.

[0032] In this embodiment, the claim preparation unit 14 is implemented within a computer device connected to the Internet 12. The computer device includes a central processing unit (CPU) that functions as a first acquisition unit 21, a second acquisition unit 22, a separation unit 23, a presentation unit 26, a third acquisition unit 27, and a description unit 28; a memory connected to the CPU that temporarily stores programs and data when an application is executed; and a mass storage device connected to the CPU that stores programs and data. The mass storage device may include, for example, a hard disk drive (HDD). The mass storage device stores, for example, an application program that realizes the functions of the first acquisition unit 21, the second acquisition unit 22, the separation unit 23, the presentation unit 26, the third acquisition unit 27, and the description unit 28. The restriction unit 24 may be implemented as a firewall implemented in hardware or software.

[0033] Next, the operation of the patent application document preparation system 11 will be explained. Now, the string "amount" is input as the difference. The string "With this food product, Mucuna pruriens bean powder can be taken easily anywhere in the same way as honey is taken" is input as the advantage. The first acquisition unit 21 generates difference data describing the text data of "amount". The second acquisition unit 22 generates advantage data describing the text data "With this food product, Mucuna pruriens bean powder can be taken easily anywhere in the same way as honey is taken". Here, no difference item is specified in the advantage data.

[0034] The separation unit 23 acquires difference data from the first acquisition unit 21 and advantage data from the second acquisition unit 22. The separation unit 23 generates common term data based on the difference data and advantage data. The separation unit 23 removes the difference data "amount" from the advantage data. Since the advantage data does not contain the character string "amount", the separation unit 23 writes the character string "As a food product, Mucuna pruriens powder can be easily taken anywhere in the same way as honey is taken" into the common term data. The separation unit 23 requests the common machine learning unit 13 to create the specified description. The common term data is handed over to the common machine learning unit 13.

[0035] The common machine learning unit 13 receives common term data from the separation unit 23. The common machine learning unit 13 creates common requirement data based on the common term data. The common machine learning unit 13 performs machine learning based on the published learning data. The language in the common term data is analyzed according to the machine learning. The common machine learning unit 13 describes the functions of the common terms and the relationships between the common terms. For example, the action "Mucuna pruriens powder is transported" is recognized according to the common term "anywhere." Furthermore, the requirement "Mucuna pruriens powder is placed in a container" is understood from the action "Mucuna pruriens powder is transported." The requirement "Mucuna pruriens powder is mixed into honey" is set according to the common term "in the same way as consuming honey." The requirement "The container has an opening with an openable lid" is set according to the common term "ingested." The text data of "Mucuna pruriens powder is placed in a container," the text data of "Mucuna pruriens powder is mixed into honey," and the text data of "The container has an opening with an openable lid" are written to the common requirement data.

[0036] The presentation unit 26 acquires common requirement data from the common machine learning unit 13. The presentation unit 26 displays the common requirement data "Mucuna pruriens powder is placed in a container," "Mucuna pruriens powder is mixed into honey," and "The container has an opening with an openable lid." on the display screen. The presentation unit 26 prompts the worker to input linking data. The display screen may simply display an input frame for text data. The worker can use the input device to enter the string "Relationship between commonalities and differences" into the input frame. For example, when the string "Mucuna pruriens powder is mixed into honey in an amount that maintains the fluidity of the honey" is entered into the input frame, the text data "Mucuna pruriens powder is mixed into honey in an amount that maintains the fluidity of the honey" is incorporated into the linking data.

[0037] The description unit 28 obtains the common requirement data "Mucuna bean powder is contained in a container," "Mucuna bean powder is mixed into honey," and "The container has an opening with an openable lid," from the presentation unit 26, and obtains the linking data "Mucuna bean powder is mixed into honey in an amount that maintains the fluidity of the honey" from the third acquisition unit 27. The description unit 28 writes a claim based on the common requirement data and the linking data. Claim data "A food product comprising a container with an opening with an openable lid, and Mucuna bean powder contained in the container and mixed into honey in an amount that maintains the fluidity of the honey" is generated.

[0038] Now, the character string "unity" is input as the difference. The character string "With this diamond jewelry, the intended composition under natural light can be maintained as is even under ultraviolet light" is input as the advantage. The first acquisition unit 21 generates difference data describing the text data of "unity". The second acquisition unit 22 generates advantage data describing the text data of "With this diamond jewelry, the intended composition under natural light can be maintained as is even under ultraviolet light". Here, the difference term "can be maintained as is" is specified in the advantage data. When specifying the difference term, the character string "can be maintained as is" is input, for example, from an input device.

[0039] The separation unit 23 acquires difference data from the first acquisition unit 21 and advantage data from the second acquisition unit 22. The separation unit 23 generates common term data based on the difference data and advantage data. The separation unit 23 removes the difference data "unity" and the difference term "can be maintained as is" from the advantage data. The separation unit 23 writes "In this diamond jewelry, the composition intended under natural light appears under ultraviolet light" into the common term data. The separation unit 23 requests the common machine learning unit 13 to create a predetermined description. The common term data is handed over to the common machine learning unit 13.

[0040] The common machine learning unit 13 receives the common term data from the separation unit 23. The common machine learning unit 13 creates common requirement data based on the common term data. The common machine learning unit 13 performs machine learning based on the published learning data. The language in the common term data is analyzed according to the machine learning. The common machine learning unit 13 describes the functions of the common terms and the relationships between the common terms. For example, the requirement "diamond stones shine colorless and transparent" is understood according to the common term "under natural light." The requirement "all diamond stones fixed to a support" is set according to the common term "composition." The requirement "diamond stones exhibit blue fluorescence" is set according to the common term "under ultraviolet light." The text data of "diamond stones shine colorless and transparent," the text data of "all diamond stones fixed to a support," and the text data of "diamond stones exhibit blue fluorescence" are written into the common requirement data.

[0041] The presentation unit 26 acquires common requirement data from the common machine learning unit 13. The presentation unit 26 displays the common requirement data "Diamond stones shine colorless and transparent," "All diamond stones fixed to a support," and "Diamond stones exhibit blue fluorescence" on the display screen. The presentation unit 26 prompts the worker to input linking data. The display screen may simply display an input frame for text data. The worker can enter a string of characters "Relationship between commonalities and differences" into the input frame using an input device. For example, when the string "All diamond stones exhibit blue fluorescence graded to a unified grade 1" is entered into the input frame, the text data "All diamond stones exhibit blue fluorescence graded to a unified grade 1" is incorporated into the linking data.

[0042] The description unit 28 acquires the common requirement data "The diamond stones shine colorlessly and transparently," "All diamond stones fixed to the support," and "The diamond stones exhibit blue fluorescence" from the presentation unit 26, and acquires the linking data "All diamond stones exhibit blue fluorescence graded to a uniform one grade" from the third acquisition unit 27. The description unit 28 writes a claim based on the common requirement data and the linking data. The claim data "Diamond jewelry comprising a support and two or more diamond stones fixed to the support that shine colorlessly and transparently, wherein all diamond stones fixed to the support exhibit blue fluorescence graded to a uniform one grade" is generated.

[0043] The common machine learning unit 13 can identify the functions of common terms and the relationships between common terms based on accumulated learning data. The constituent elements of the prior art related to the differences can be identified. The constituent elements of the prior art can be presented to the writer (human). The writer can input the relationships between the constituent elements of the prior art and the differences in response to the presentation. In this way, the differences are linked to the common terms. The claims can be completed according to the linking. Since machine learning is used in writing the claims, the burden on the writer can be reduced. Here, common terms are separated from the description of advantages during machine learning. Rare "differences" can be excluded from the learning data. The common machine learning unit 13 can describe the functions of common terms and the relationships between common terms with high quality.

[0044] In this embodiment, the output of the difference terms (features) from the separation unit (identification unit) 23 and the presentation unit 26 is restricted, so that the difference terms can be reliably concealed when writing claims. On the other hand, the constituent elements of the common terms can be identified with high quality based on the vast amount of training data that has been made public. This improves the quality of the claims. In this way, while the confidentiality of the invention is ensured, machine learning based on the vast amount of training data can be effectively used when drafting claims.

[0045] FIG. 3 shows a schematic configuration of a patent application document drafting system 11a according to a second embodiment of the present invention. The claim drafting unit 14a includes a description unit 29 that acquires common requirement data from the common machine learning unit 13, instead of the description unit 28 of the first embodiment. The claim drafting unit 14a may be implemented within a computer device, as described above. The presentation unit 26 and third acquisition unit 27 of the first embodiment may be omitted from the claim drafting unit 14a. The description unit 29 describes claims by combining the functions of common terms, the relationships between common terms, and the relationships between common terms and differences. The description unit 29 generates claim data that expresses the claims. The claim data may be composed of, for example, text data. The claim data may be stored in, for example, a storage device. Claims may be displayed on a display screen based on the text data. Instead of such visual display, audio may be output from a speaker based on the text data.

[0046] Here, the description unit 29 identifies relationships between commonalities and differences based on difference data and advantage data. The description unit 29 may identify relationships between commonalities and differences based on machine learning. To perform machine learning, an on-premise prior art database is connected to the description unit 29. The on-premise prior art database is isolated from other machine learning. If the machine learning of the description unit 29 is separated from the machine learning of the common machine learning unit 13, the output of differences and advantages from the description unit 29 to the common machine learning unit 13 can be prevented. The common machine learning unit 13 cannot learn information about differences and advantages from the description unit 29.

[0047] In this embodiment, the common machine learning unit 13 can identify the functions of common terms and the relationships between the common terms based on accumulated learning data. The constituent elements of the prior art related to the differences can be grasped. Since the relationship between the common terms and the differences is identified in the description of the advantages, the differences are linked to the common terms. The claims can be finalized according to the linking. Since machine learning is used in writing the claims, the burden on the writer can be reduced. Here, the common terms are separated from the description of the advantages in machine learning. Rare "differences" can be excluded from the learning data. The common machine learning unit 13 can describe the functions of common terms and the relationships between the common terms with high quality.

[0048] In the claim drafting unit 14a according to this embodiment, the output of the difference terms (features) from the separation unit (identification unit) 23 and the description unit 29 is restricted, so that the difference terms can be reliably concealed when writing claims. Meanwhile, the constituent elements of the common terms can be identified with high quality based on the vast amount of publicly available learning data. The quality of the claims can be improved. Thus, while the confidentiality of the invention is ensured, machine learning based on the vast amount of learning data can be effectively utilized when drafting claims.

[0049] FIG. 4 shows a schematic configuration of a patent examination system 31 according to a third embodiment of the present invention. The patent examination system 31 includes a common machine learning unit 13 connected to the Internet 12 and a claim examination unit 32 also connected to the Internet 12. The common machine learning unit 13 can collect training data from web pages 15 accessed via the Internet 12 and a prior art database 16 connected to the Internet 12. The training data can include text data, image data such as photographs and graphics, audio data, and other data. The prior art database 16 can store, for example, patent publications, patent lay publications, utility model registration publications, research papers, and other data. The region and language are not limited to a single region or language. The common machine learning unit 13 may support multiple languages with the support of a translation device. The common machine learning unit 13 performs machine learning based on the collected training data. The claim examination unit 32 examines claims, which are one element of patent application documents, using machine learning from the common machine learning unit 13. The claims determine the scope of a patent right. Other public networks may be used instead of the Internet 12. The more training data that is collected, the more accurate the machine learning can be.

[0050] The claim examination unit 32 includes a first acquisition unit 33 that acquires difference data describing the differences between the prior art and the invention, and a second acquisition unit 34 that acquires advantage data describing the advantages resulting from the differences. The difference data can be composed of text data input from an input device, for example. Here, the difference is expressed by a single word. The difference only needs to reflect the inventor's subjective opinion. The input device can include a keyboard as well as a voice recognition device that generates text data from voice input from a microphone.

[0051] The advantage data can be composed of text data input from an input device, for example. The advantage includes common terms that describe the constituent elements of the prior art and difference terms (characteristic terms or hidden terms) that describe the differences between the prior art and the invention. The input device can include a keyboard as well as a voice recognition device that generates text data from voice input from a microphone.

[0052] The claim examination unit 32 includes a separation unit 35 connected to the first acquisition unit 33 and the second acquisition unit 34. The separation unit 35 separates common terms common to the prior art and the invention from the description of advantages based on the difference data and advantage data. In separating the common terms, the separation unit 35 removes the differences and the actions linked to the differences. The difference terms may be specified in advance in the advantage data. An input device can be used for the specification. In this way, the separation unit 35 identifies common terms that describe the constituent features of the prior art and difference terms that describe the differences between the prior art and the invention. The separation unit 35 generates common term data that identifies the common terms. The actions of the common terms may be input from the input device when generating such common term data.

[0053] The separation unit 35 may separate commonalities from descriptions of advantages based on machine learning. To perform the machine learning, an on-premise prior art database is connected to the separation unit 35. The on-premise prior art database is isolated from other machine learning. If the machine learning of the separation unit 35 is separated from the machine learning of the common machine learning unit 13, the output of differences and advantages from the separation unit 35 to the common machine learning unit 13 can be prevented. The common machine learning unit 13 cannot learn information about differences and advantages from the separation unit 35.

[0054] The common machine learning unit 13 describes the functions of the common terms and the relationships between the common terms based on the published learning data. In describing the functions, the common machine learning unit 13 acquires the common term data from the separation unit 35. The common machine learning unit 13 generates common requirement data that identifies the functions of the common terms and the relationships between the common terms. The common requirement data can be composed of text data, for example.

[0055] The common machine learning unit 13 passes the common requirement data to the search unit 36. The search unit 36 searches the prior art database 16 based on the common requirement data. In response to the search, the search unit 36 extracts prior art containing common terms. The search unit 36 generates adjacent technology data based on the extracted prior art. The adjacent technology data can include all (one or more) prior art that contains the common terms. The adjacent technology data can be composed of text data, for example.

[0056] The claim examination section 32 includes an analysis section 38 that selects the closest prior art from the extracted prior art. The analysis section 38 obtains adjacent art data from the search section 36 and advantage data from the second acquisition section 34 for the selection. The analysis section 38 compares the advantages with each prior art. When the largest commonality is found in one prior art, that prior art is set as the closest prior art. The closest prior art has the largest commonality among the prior arts.

[0057] When acquiring adjacent technology data, the adjacent technology data passes through a restriction unit 37. The restriction unit 37 restricts data exchange between the analysis unit 38 and the search unit 36. The restriction unit 37 prevents the analysis unit 38 from outputting difference terms to the search unit 36. The search unit 36 cannot learn information about the difference terms from the analysis unit 38. The restriction unit 37 can achieve security that ensures the confidentiality of the invention.

[0058] The analysis unit 38 compares the advantages with the nearest prior art and identifies the differences between the nearest prior art and the advantages. When identifying the differences, the analysis unit 38 compares the advantages with the nearest prior art. The analysis unit 38 generates difference data that identifies the differences. In the description of the differences, the relationship between the differences and the nearest prior art is omitted. The differences are separated from the nearest prior art.

[0059] If no difference is detected between the closest prior art and the advantage, the analysis unit 38 generates a first reconsideration notice. The first reconsideration notice is presented to the worker. In response to the first reconsideration notice, the worker is prompted to reconsider the differences. For example, the text "Not substantially different from the prior art" is displayed on the display screen.

[0060] The analysis unit 38 may identify the closest prior art and the difference in advantages based on machine learning. To perform the machine learning, an on-premise prior art database is connected to the analysis unit 38. The on-premise prior art database is isolated from other machine learning. If the machine learning of the analysis unit 38 is separated from the machine learning of the common machine learning unit 13, the output of the differences and advantages from the analysis unit 38 to the common machine learning unit 13 can be prevented. The common machine learning unit 13 cannot learn the information about the differences and advantages from the analysis unit 38.

[0061] The common machine learning unit 13 identifies the effect produced from the difference based on the published learning data. To identify the effect, the common machine learning unit 13 acquires difference data. Once the effect is identified, the common machine learning unit 13 compares the effect with the advantage. To compare, the common machine learning unit 13 acquires advantage data from the analysis unit 38. The common machine learning unit 13 determines whether the effect is within the expected range based on the published learning data. If the advantage is determined to be within the expected range, the common machine learning unit 13 generates differential effect data (determination data) that identifies the expected (common) effect. If the advantage is determined to be an unexpected effect, the common machine learning unit 13 generates differential effect data that identifies the superiority of the effect.

[0062] The claim examination division 32 includes a first examination division 41 that determines the inventive step based on differential effect data. When the first examination division 41 obtains differential effect data that recognizes an expected effect, it denies the inventive step. When the expected effect is recognized in the differential effect data, the first examination division 41 generates a second reconsideration notice. The second reconsideration notice is presented to the worker. In response to the presentation of the second reconsideration notice, the worker is prompted to reconsider the advantages. For example, the display screen may display the text "It is only a mundane effect." When the first examination division 41 obtains differential effect data that recognizes an unexpected effect, it affirms the inventive step. The first examination division 41 generates a positive notice specifying the affirmation of the inventive step.

[0063] The claim examination division 32 includes a second examination division 42 that determines the inventive step upon receiving a positive notification from the first examination division 41. The second examination division 42 refers to the mechanism that leads from the difference to the advantage when determining the inventive step. When referring to the mechanism, the second examination division 42 instructs the common machine learning unit 13 to analyze the mechanism. The second examination division 42 outputs the difference data and advantage data to the common machine learning unit 13. Upon obtaining the difference data and advantage data, the common machine learning unit 13 describes the mechanism that leads from the difference to the advantage based on the published learning data. The feasibility of the mechanism is determined.

[0064] When the second examination division 42 obtains feasibility data that recognizes the validity of the mechanism, it affirms the inventive step. When the feasibility data recognizes that the mechanism is not valid, the second examination division 42 generates a third reconsideration notice. The third reconsideration notice is presented to the worker. In response to the presentation of the third reconsideration notice, the worker is prompted to reconsider the advantages. For example, the display screen may display the text "The advantage is not valid due to the difference." When the second examination division 42 obtains feasibility data that recognizes the mechanism is valid, it affirms the inventive step. A patent notice is output from the second examination division 42.

[0065] The patent examination system 31 includes a patent publication issuing department 43 that issues a patent gazette when a patent notice is issued by the second examination department 42. Similar to the claim drafting department 14, the patent publication issuing department 43 writes claims based on difference data and advantage data. The patent publication issuing department 43 fleshes out the claims and describes embodiments of the product. As shown in Figure 5, when describing the embodiments, the patent publication issuing department 43 breaks down the product according to a pyramidal structure. The components of the product are grouped into higher-level and intermediate-level concepts according to their functions. As shown in Figure 6, the functions and relationships of each component are described in a paragraph.

[0066] The pyramid structure is built according to three principles: Principle [1] states that "within each group, the ideas must always be of the same kind;" Principle [2] states that "at any level, an idea must always be a summary of the ideas of its lower group;" Principle [3] states that "within each paragraph, the ideas must always be ordered logically;"

[0067] In this example, the claim examination unit 32 and the publication issuance unit 43 are implemented within a computer device connected to the Internet 12. The computer device includes a central processing unit (CPU) that functions as the first acquisition unit 41, the second acquisition unit 42, the separation unit 35, the analysis unit 38, the first examination unit 41, the second examination unit 42, and the publication issuance unit 43; a memory connected to the CPU that temporarily stores programs and data when executing an application; and a mass storage device connected to the CPU that stores programs and data. The mass storage device may include, for example, a hard disk drive (HDD). The mass storage device stores, for example, application programs that realize the functions of the first acquisition unit 41, the second acquisition unit 42, the separation unit 35, the analysis unit 38, the first examination unit 41, the second examination unit 42, and the publication issuance unit 43. The regulating unit 37 may be implemented as a firewall implemented in hardware or software.

[0068] Next, the operation of the patent examination system 31 will be explained. Now, the string "amount" is input as the difference. The string "With this food product, Mucuna pruriens powder can be taken easily anywhere in the same way as honey is taken" is input as the advantage. The first acquisition unit 33 generates difference data describing the text data of "amount". The second acquisition unit 34 generates advantage data describing the text data "With this food product, Mucuna pruriens powder can be taken easily anywhere in the same way as honey is taken". Here, no difference item is specified in the advantage data.

[0069] The separation unit 35 acquires difference data from the first acquisition unit 33 and advantage data from the second acquisition unit 34. The separation unit 35 generates commonality data based on the difference data and advantage data. The separation unit 35 removes the difference data "amount" from the advantage data. Since the advantage data does not contain the character string "amount," the separation unit 35 writes the character string "As a food product, Mucuna pruriens powder can be easily ingested anywhere in the same way as honey is ingested" into the commonality data. The separation unit 35 requests the search unit 36 to search for prior art.

[0070] The search unit 36 requests the common machine learning unit 13 to create a predetermined description. The common term data is handed over to the common machine learning unit 13. The common machine learning unit 13 creates common requirement data based on the common term data. The common machine learning unit 13 performs machine learning based on the published learning data. The language in the common term data is analyzed according to the machine learning. The common machine learning unit 13 describes the functions of the common terms and the relationships between the common terms. For example, the action "Mucuna pruriens powder is transported" is recognized according to the common term "anywhere." Furthermore, the requirement "Mucuna pruriens powder is placed in a container" is understood from the action "Mucuna pruriens powder is transported." The requirement "Mucuna pruriens powder is mixed into honey" is set according to the common term "in the same way as ingesting honey." The requirement "The container has an opening with a lid that can be opened and closed" is set according to the common term "ingested." The text data "The Mucuna pruriens powder is contained in a container," the text data "The Mucuna pruriens powder is mixed with honey," and the text data "The container has an opening with an openable and closable lid" are written into the common requirement data.

[0071] The search unit 36 receives common requirement data from the common machine learning unit 13. The search unit 36 searches the prior art database 16 based on the common requirement data "Mucuna pruriens powder is contained in a container," "Mucuna pruriens powder is mixed with honey," and "The container has an opening that can be closed and covered." The search unit 36 generates adjacent technology data based on the search results. The adjacent technology data describes all publications, papers, and products that contain the common requirement data "Mucuna pruriens powder is contained in a container," "Mucuna pruriens powder is mixed with honey," and "The container has an opening that can be closed and covered."

[0072] For example, if the benefit data identifies "As a food product, Mucuna pruriens powder can be easily ingested anywhere," the common machine learning unit 13 recognizes the action "Mucuna pruriens powder is transported" in accordance with the common term "anywhere." Furthermore, the requirement "Mucuna pruriens powder is contained in a container" is understood from the action "Mucuna pruriens powder is transported." The requirement "The container has an opening with an openable lid" is set in accordance with the common term "ingested." The search unit 36 searches the prior art database 16 based on the common requirement data "Mucuna pruriens powder is contained in a container" and "The container has an opening with an openable lid." When the search unit 36 generates neighboring technology data based on the search results, the neighboring technology data describes all publications, papers, and products that contain the common requirement data "Mucuna pruriens powder is contained in a container" and "The container has an opening with an openable lid."

[0073] The adjacent technology data is passed from the search unit 36 to the analysis unit 38. The analysis unit 38 selects the closest prior art from the extracted prior art. In making the selection, the analysis unit 38 compares the merits of each prior art. When a prior art with the greatest commonality is found, that prior art is set as the closest prior art. For example, when a patent publication containing "Mucuna pruriens powder mixed into honey" is found, the patent publication is set as the closest prior art.

[0074] The commonalities correspond to the prior art described in the advantage data. If the differences and the advantages resulting from the differences are understood, prior art can be extracted effectively. In this way, the closest prior art can be identified effectively. The burden on the researcher can be reduced because the researcher only needs to input the differences and advantages to extract the closest prior art. Here, machine learning can be used in the search unit 36. Since there is a huge amount of publicly available training data, prior art can be extracted with high quality. Moreover, since only the commonalities need to be handed over to the machine learning, the confidentiality of the invention can be ensured.

[0075] The analysis unit 38 identifies differences between the nearest prior art and the advantages. To identify differences, the analysis unit 38 compares the advantages with the nearest prior art. The analysis unit 38 generates difference data that identifies the differences. The difference data may, for example, describe "in an amount that maintains the fluidity of honey." In the difference data, the differences are described separately from the nearest prior art. If no differences are detected between the nearest prior art and the advantages, the analysis unit 38 generates a first reconsideration notice. For example, the text "Substantially no different from the prior art" is displayed on the display screen.

[0076] When the common machine learning unit 13 receives the differential data from the analysis unit 38, it identifies the effect produced by the difference. The common machine learning unit 13 performs machine learning based on the published learning data. The phrase "in an amount that maintains the fluidity of honey" in the differential data is analyzed according to the machine learning. If an effect such as "Mucuna pruriens powder can be ingested in the same way as honey is ingested" is identified, the common machine learning unit 13 compares the effect with the benefits. In the comparison, the common machine learning unit 13 acquires benefit data. The common machine learning unit 13 determines whether the effect is within the expected range. Since the benefit "As a food product, Mucuna pruriens powder can be easily ingested anywhere in the same way as honey is ingested" is within the expected range, the common machine learning unit 13 generates differential effect data identifying the expected effect. When the first review unit 41 acquires differential effect data identifying the expected effect, it denies the inventive step. The first review unit 41 generates a second reconsideration notice. For example, the text "It has only ordinary effects" is displayed on the display screen.

[0077] If the common machine learning unit 13 identifies an effect such as "the sweetness of honey is added to tasteless Mucuna pruriens powder," then the advantage "In the case of a food product, Mucuna pruriens powder can be easily ingested anywhere in the same way as honey is ingested" is determined to be an unexpected effect. The common machine learning unit 13 generates differential effect data that identifies the unexpected effect. When the first examination unit 41 obtains the differential effect data that identifies the unexpected effect, it affirms the inventive step. The first examination unit 41 generates a positive notice that identifies the inventive step.

[0078] If the effect of the difference "in an amount that maintains the fluidity of honey" is not found based on the published learning data, the advantage is determined to be an unexpected effect. The common machine learning unit 13 generates differential effect data that identifies the affirmation of inventive step. Upon receiving the differential effect data, the first examination unit 41 affirms the inventive step. The first examination unit 41 generates a positive notice that identifies the affirmation of inventive step.

[0079] The common machine learning unit 13 identifies the effect produced by the difference based on the accumulated learning data. When identifying the effect, the difference is separated from the closest technology, so the effect of the difference generally does not contribute to satisfying the requirement of inventive step (non-obviousness). If an unexpected effect is recognized in the advantage, the inventive step (non-obviousness) of the invention can be affirmed. In this way, the inventive step (non-obviousness) of the invention can be determined by the common machine learning unit 13. The objectivity of the examination can be ensured. Machine learning can be used in the examination of inventive step (non-obviousness).

[0080] When the second examination division 42 receives a positive notice from the first examination division 41, it determines whether the invention has an inventive step. The second examination division 42 instructs the common machine learning unit 13 to analyze the mechanism that leads from the difference to the advantage. The second examination division 42 outputs the difference data and advantage data to the common machine learning unit 13. Since the advantage "In a food product, Mucuna pruriens bean powder can be easily ingested anywhere in the same way as honey is ingested" is realized in accordance with the difference "in an amount that maintains the fluidity of honey," the common machine learning unit 13 generates feasibility data that recognizes the feasibility of the mechanism. The second examination division 42 affirms the inventive step based on the feasibility data. A patent notice is output from the second examination division 42.

[0081] If the advantage "In food products, the absorbability of L-dopa can be increased" is identified in accordance with the difference "in an amount that maintains the fluidity of honey," the common machine learning unit 13 generates feasibility data that recognizes the ineffectiveness of the mechanism, since the absorbability of L-dopa is unrelated to the fluidity of honey. The second examination unit 42 rejects the inventive step based on the feasibility data. The second examination unit 42 generates a third notice of reconsideration. For example, the display screen displays the text "The advantage is not effective due to the difference."

[0082] The common machine learning unit 13 describes the mechanism that leads from the difference to the advantage based on the accumulated learning data. If the mechanism is established, the advantage can be judged as a reasonable claim. If the advantage is then recognized to have an unexpected effect, the inventive step (non-obviousness) of the invention can be affirmed. Even if the advantage has an unexpected effect, if the mechanism is not established, the advantage can be judged as an irrational claim. The inventive step (non-obviousness) of the invention can be denied.

[0083] When a notice of patent is issued by the claim examination department 32, the publication issuing department 43 issues a patent publication. The publication issuing department 43 describes an embodiment of the product based on the claim description: "A food product comprising a container having an opening with an openable lid, and Mucuna pruriens powder contained in the container and mixed with honey in an amount that maintains the fluidity of the honey." For example, the container is disassembled into a body and a lid. The body is partitioned into a storage space and an opening. The publication issuing department 43 can instruct the common machine learning unit 13 to describe the embodiment.

[0084] In this embodiment, the patentability of an invention can be determined on the spot based on difference data describing the differences between the prior art and the invention, and advantage data describing the advantages derived from the differences. An applicant may be able to obtain a patent right on the spot based on the input of the differences and advantages. Therefore, even if the common machine learning unit 13 is used to identify the differences between the closest prior art and the advantages and the mechanism by which the differences lead to the advantages, it is considered that the impact on the confidentiality of the invention is small. On the other hand, an on-premise machine learning unit may be connected to the claim examination unit 32. If the on-premise machine learning unit is separated from other machine learning units, the output of the differences and advantages from the claim examination unit 32 to the common machine learning unit 13 can be prevented. The confidentiality of the differences and advantages can be ensured. [Explanation of symbols]

[0085] 11...Patent application document preparation system, 11a...Patent application document preparation system, 13...Machine learning unit (common machine learning unit), 21...First acquisition unit, 22...Second acquisition unit, 23...Separation unit (identification unit), 24...Regulation unit, 26...Presentation unit, 27...Third acquisition unit, 28...Description unit, 29...Description unit, 31...Patent examination system, 36...Search unit, 38...Analysis unit, 41...Examination unit (first examination unit), 42...Examination unit (second examination unit).

Claims

1. a first acquisition unit that acquires difference data describing differences between the prior art and the invention; a second acquisition unit that acquires advantage data that describes advantages generated from the differences; a separation unit that separates common features common to the prior art and the invention from the description of the advantages based on the difference data and the advantage data, and outputs common feature data that identifies the common features; a machine learning unit that acquires the common term data from the separation unit and describes functions of the common terms and relationships between the common terms based on publicly available learning data; a presentation unit that acquires common requirement data that identifies the functions and the relationships from the machine learning unit and presents the functions and the relationships in a manner that is perceptible to human senses; a description unit that acquires linking data describing the relationship between the common features and the differences in response to the presentation of the functions and the relationships, and describes claims based on the common feature data and the linking data; A patent application document drafting system comprising:

2. a first acquisition unit that acquires difference data describing differences between the prior art and the invention; a second acquisition unit that acquires advantage data that describes advantages generated from the differences; a separation unit that separates common features common to the prior art and the invention from the description of the advantages based on the difference data and the advantage data, and outputs common feature data that identifies the common features; a machine learning unit that acquires the common term data from the separation unit and describes functions of the common terms and relationships between the common terms based on publicly available learning data; a description unit that acquires common requirement data that identifies the functions and the relationships from the machine learning unit, and describes claims by combining the common features with the differences based on the description of the advantages; A patent application document drafting system comprising:

3. a specification section that specifies common features that describe the constituent elements of the prior art and different features that describe the differences between the prior art and the invention; a machine learning unit that acquires common term data that identifies the common term from the identification unit and identifies constituent elements necessary for the operation of the common term based on the published learning data; a description unit that acquires constituent feature data that identifies the constituent features from the machine learning unit, and describes the claim by combining the constituent features with the difference terms; a restriction unit that restricts the output of the difference terms from the specification unit and the description unit; A patent application document drafting system comprising:

4. an acquisition unit that receives advantage data input from an input device and describes advantages derived from the differences between the prior art and the invention; an analysis unit that compares the advantages with the closest technologies and identifies differences between the closest technologies and the advantages; a machine learning unit that acquires difference data that identifies the difference and the advantage data, identifies an effect produced by the difference separated from the closest technology based on publicly available learning data, compares the effect with the advantage, and determines whether the effect is within the expected range based on the publicly available learning data; When the machine learning department obtains judgment data that recognizes the expected effect, the examining department denies the inventive step of the invention. A patent examination system comprising:

5. an acquisition unit that receives advantage data input from an input device and describes advantages derived from the differences between the prior art and the invention; an analysis unit that compares the advantages with the closest technologies and identifies differences between the closest technologies and the advantages; a machine learning unit that acquires difference data that identifies the difference and the advantage data, describes a mechanism that leads from the difference to the advantage based on the published learning data, and determines the feasibility of the mechanism; When the machine learning department obtains the judgment data that confirms the establishment of the mechanism, the examining department will confirm the inventive step of the invention. A patent examination system comprising:

6. a first acquisition unit that acquires difference data describing differences between the prior art and the invention; a second acquisition unit that acquires advantage data that describes advantages generated from the differences; a separation unit that separates common features common to the prior art and the invention from the description of the advantages based on the difference data and the advantage data, and outputs common feature data that identifies the common features; a search unit that acquires the common term data, searches a prior art database based on the common term, and extracts prior art containing the common term; an analysis unit that identifies the prior art with the largest range of commonality as the closest prior art based on the advantage data; A patent examination system comprising:

Images