Device for analyzing technical document, method, and program for same

Abstract

According to the present invention, in a method for analyzing technical documents, effort for imparting of classification that can be used as an analysis axis is minimized. First, specification of a p count (p > 1) of technical documents D = {D1, D2, ..., Dp} to be analyzed is received from a user terminal (S201). Next, a device 100 requests a generative AI model to generate a q count of classifications C = {C1, C2, ..., Cq} capable of classifying the p count of technical documents D, on the basis of a p count of features F = {F1, F2, ..., Fp} associated respectively with the p count of technical documents D (S202). Before the generation of the classifications C is requested, the generation AI model may be requested to generate a feature Fi by inputting at least a part of technical documents. The device 100 then receives the generated classifications C (S203), and requests the AI model to associate, from among the classifications C, for i that is no less than 1 and no more than p, one or a plurality of classifications representing technical documents Di, as C'i (S204). The device 100 then receives and stores the classifications C'i associated with the technical documents Di for i that is no less than 1 and no more than p (S205).

Description

Apparatus, method and program for analyzing technical literature

[0001] The present invention relates to an apparatus, a method and a program for analyzing technical literature.

[0002] Patent research serves as a fundamental resource for understanding the state of research and development to date. Patent documents are assigned patent classifications according to their content, and various analyses are carried out by combining these patent classifications with keywords as needed.

[0003] However, patent classifications are updated only once every few years, making them inadequate as analytical axes in fields undergoing rapid change. For example, even if researchers wanted to analyze the R&D focus of the rapidly expanding metaverse field, there is no patent classification that subdivides the metaverse field. This would require them to read each patent document and interpret its contents, which would require a significant amount of effort. For example, a 100KB text file written in English can be estimated to contain around 20,000 words. If there are 50 patent documents to analyze, without a clue, this would mean reading around 1 million words of English text. The same burden of analysis applies to technical documents that do not have a classification system other than patent documents.

[0004] The present invention has been made in light of these points, and its first object is to reduce the effort required to generate classifications that can be used as analytical axes in an apparatus, method, or program for analyzing technical literature. An alternative second object of the present invention is to reduce the effort required to assign classifications that can be used as analytical axes in an apparatus, method, or program for analyzing technical literature. A further alternative third object of the present invention is to provide, in an apparatus, method, or program for analyzing technical literature, the generation or application of information that enables the content of the technical literature to be grasped.

[0005] In order to achieve this object, a first aspect of the present invention is a method for analyzing technical documents, comprising the steps of receiving, from a user terminal, a designation of p (p>1) technical documents D = {D1, D2, ..., Dp} to be analyzed; making a first request to a first AI model to generate q classifications C = {C1, C2, ..., Cq} capable of classifying the p technical documents D based on p features F = {F1, F2, ..., Fp} associated with each of the p technical documents D, where the first AI model is a generative AI model; making a second request to a second AI model to associate, for i between 1 and p, one or more classifications C'i representing technical document Di from among the q classifications C or classifications corresponding to at least some of them; and receiving and storing the classifications C'i associated with technical document Di, for i between 1 and p.

[0006] Furthermore, a second aspect of the present invention is a method of the first aspect, further comprising a step of transmitting to the user terminal classification information including at least a portion of the p classifications C'={C'1, C'2..., C'p} or literature information identifying one or more technical literature to which they are assigned.

[0007] A third aspect of the present invention is the method of the first or second aspect, wherein the p features F are included in the specification.

[0008] Furthermore, a fourth aspect of the present invention is the first or second aspect, further comprising a step of making a third request to a third AI model to generate features Fi by summarizing at least a portion of technical documents Di for i greater than or equal to 1 and less than or equal to p, before requesting the first AI model to generate the classification C, wherein the third AI model is a generative AI model.

[0009] A fifth aspect of the present invention is the method according to the fourth aspect, wherein the p features F are of the same predetermined format.

[0010] Furthermore, a sixth aspect of the present invention is a method of the first or second aspect, further comprising a step of making a third request to a third AI model to input at least a portion of technical documents Di and generate features Fi for i greater than or equal to 1 and less than p, before requesting the first AI model to generate the classification C, wherein the third AI model is a generative AI model and the features Fi have a length equal to or less than a predetermined length.

[0011] A seventh aspect of the present invention is the method of the sixth aspect, wherein the predetermined length is 125 English words or 250 Japanese characters.

[0012] An eighth aspect of the present invention is the method of the sixth or seventh aspect, wherein the at least part is within the first half of the entire length of the technical document Di.

[0013] A ninth aspect of the present invention is the method of any one of the fourth to eighth aspects, wherein the technical document is a patent document, and at least a part of the technical document Di is at least a part of the summary of the invention, the brief description of the drawings, or the description of the preferred embodiment of the invention of the patent document Di.

[0014] Furthermore, a tenth aspect of the present invention is a method according to any one of the fourth to eighth aspects, wherein the technical document is a patent document, and at least a portion of the technical document Di includes one or more claims of the patent document Di.

[0015] An eleventh aspect of the present invention is the method of any one of the fourth to tenth aspects, wherein the third requirement includes specifying a third condition to be imposed on the third AI model in generating the p features F.

[0016] A twelfth aspect of the present invention is the method according to any one of the fourth to tenth aspects, wherein the third request includes specifying the nature of the feature to be extracted from the at least part.

[0017] Furthermore, a thirteenth aspect of the present invention is a method according to any one of the first to ninth aspects, wherein the technical documents are patent documents, and each of the p features F is a summary of the problem to be solved by the invention described in each of the patent documents included in the p patent documents D.

[0018] Furthermore, a fourteenth aspect of the present invention is the method of any one of the first to thirteenth aspects, wherein the first AI model is a model to which a Transformer architecture is applied.

[0019] A fifteenth aspect of the present invention is the method of the fourteenth aspect, wherein the second AI model is identical to the first AI model.

[0020] Furthermore, a sixteenth aspect of the present invention is the method of the fourteenth aspect, wherein the second AI model is provided on the same platform as the first AI model.

[0021] Furthermore, a seventeenth aspect of the present invention is the method of the fourth or sixth aspect, wherein the third AI model is a model to which a Transformer architecture is applied.

[0022] Furthermore, an 18th aspect of the present invention is a method according to the 4th or 6th aspect, wherein the third AI model is provided on the same platform as at least one of the first AI model and the second AI model.

[0023] A nineteenth aspect of the present invention is a method according to any one of the first to eighteenth aspects, wherein the first request includes specifying a first condition to be imposed on the first generative AI model in generating the q classifications C.

[0024] A twentieth aspect of the present invention is the method of the nineteenth aspect, wherein the first condition includes the technical field to which the p patent documents D belong.

[0025] A twenty-first aspect of the present invention is the method of the nineteenth aspect, wherein the first condition includes one or more expressions to be included in at least one of the q classes C.

[0026] A twenty-second aspect of the present invention is the method according to the twenty-first aspect, wherein the first condition includes that the q classes C include an expression meaning "other."

[0027] A twenty-third aspect of the present invention is the method according to the nineteenth aspect, wherein the first condition does not fix the value of q.

[0028] A twenty-fourth aspect of the present invention is the method according to the nineteenth aspect, wherein the first condition includes a range of values ​​of q that is less than p.

[0029] A 25th aspect of the present invention is the method of the 19th aspect, wherein the first condition includes a range or value in which the value of q is equal to or greater than a lower limit of 2 or more and less than or equal to an upper limit of 10 or less.

[0030] Furthermore, a 26th aspect of the present invention is a method of any one of the 1st to 25th aspects, wherein the transmission of classification information including at least a part of the p classifications C' or literature information identifying one or more technical literature to which they are assigned to the user terminal is the transmission of viewing screen display information for displaying a viewing screen for the classification information or the literature information on the user terminal.

[0031] Furthermore, a 27th aspect of the present invention is the method of the 26th aspect, wherein the viewing screen allows a user using the user terminal to input feedback for one or more categories included in the p categories C'.

[0032] A twenty-eighth aspect of the present invention is the method of the twenty-seventh aspect, wherein the feedback is a positive or negative selection or evaluation of the one or more classifications.

[0033] A twenty-ninth aspect of the present invention is the method of the twenty-seventh aspect, wherein the feedback is a suggestion or designation of a modification to the one or more classifications.

[0034] A 30th aspect of the present invention is the method of the 27th aspect, wherein the feedback is a suggestion or designation of a new classification by the user.

[0035] A thirty-first aspect of the present invention is the method of the twenty-seventh aspect, wherein the feedback is a proposal or designation of abstraction or concretization of the p classifications C'.

[0036] Furthermore, a 32nd aspect of the present invention is a method according to any one of the 27th to 31st aspects, further comprising, after receiving the feedback, requesting the first AI model to generate r classifications C={C1, C2..., Cr} capable of classifying the p technical documents D in accordance with the feedback.

[0037] Furthermore, a 33rd aspect of the present invention is the method of the 26th aspect, wherein the viewing screen is capable of displaying analysis results in which at least a portion of the p classifications C' is used as a first axis among a plurality of analysis axes.

[0038] A thirty-fourth aspect of the present invention is the method according to the thirty-third aspect, wherein a second axis of the plurality of analysis axes, which is different from the first axis, is an axis representing a change over time.

[0039] A thirty-fifth aspect of the present invention is the method according to the thirty-third aspect, wherein a second axis of the plurality of analysis axes, which is different from the first axis, is a country axis.

[0040] A thirty-sixth aspect of the present invention is the method of the thirty-third aspect, wherein a second axis of the plurality of analysis axes, which is different from the first axis, is an axis representing a patent classification.

[0041] A thirty-seventh aspect of the present invention is the method of any one of the first to thirty-sixth aspects, wherein the technical documents are patent documents, and the q classes C do not include patent classes.

[0042] A thirty-eighth aspect of the present invention is a method according to any one of the first to thirty-seventh aspects, wherein the first request to the first AI model is made by obtaining pre-stored code for making the first request, and setting the feature F to a variable included in the code and executing the code obtained.

[0043] Moreover, the 39th aspect of the present invention is a method for analyzing technical documents, comprising the steps of: receiving, from a user terminal, a specification of p (p>1) technical documents D={D1, D2, …, Dp} to be analyzed; for i ranging from 1 to p, making a request k to an AI model (k) to generate a feature Fi by inputting at least a part of the technical document Di, wherein the AI model (k) is a generation AI model and the feature Fi includes one or more sentences longer than a single term and not included in the input; and making a request l to an AI model (l) to generate q (1<q<p) classifications C={C1, C2, …, Cq} capable of classifying the p features F={F1, F2, …, Fp} associated with each of the p technical documents D or a modified feature F’ in which at least any of them is modified, wherein the AI model (l) is a generation AI model.

[0044] Moreover, the 40th aspect of the present invention is a method for analyzing technical documents, comprising the steps of: receiving, from a user terminal, a specification of p (p>1) technical documents D={D1, D2, …, Dp} to be analyzed; requesting a generation AI model to generate q (1<q<p) classifications C={C1, C2, …, Cq} capable of classifying the p technical documents D in association with each of the p features F={F1, F2, …, Fp} associated with each of the p technical documents D; and receiving and storing the classifications associated with the technical document Di for i ranging from 1 to p.

[0045] In addition, a forty-first aspect of the present invention is a method for analyzing technical documents, comprising the steps of receiving from a user terminal designation of p (p>1) technical documents D = {D1, D2, ..., Dp} to be analyzed; making a request k to an AI model (k) for i between 1 and p inclusive, requesting that the AI ​​model (k) input at least a portion of the technical documents Di and generate features Fi, wherein the AI ​​model (k) is a generative AI model, and the features Fi include one or more sentences not included in the input and longer than a single term; making a request l to an AI model (l) to generate q classifications C = {C1, C2 ..., Cq} associated with each technical document, capable of classifying p features F = {F1, F2 ..., Fp} associated with each of the p technical documents D or modified features F' obtained by modifying at least one of them, wherein the AI ​​model (l) is a generative AI model; and receiving and storing the classifications associated with the technical documents Di, for i between 1 and p inclusive.

[0046] In addition, a forty-second aspect of the present invention is a method for analyzing technical literature, comprising the steps of receiving from a user terminal a designation of a first set of p (p>1) technical literatures; requesting a generative AI model to generate q classifications C capable of classifying the first set of technical literatures based on p features associated with each of the first technical literatures; receiving from the user terminal a designation of a second set of s (s>1) technical literatures D = {D1, D2, ..., Ds}; for i between 1 and s, requesting the AI ​​model to associate one or more classifications C'i representing technical literature Di from among the q classifications C or classifications corresponding to at least some of them; and receiving and storing the classifications C'i associated with technical literature Di, for i between 1 and s.

[0047] Also, a 43rd aspect of the present invention is a method for analyzing technical documents, including the steps of receiving, from a user terminal, a specification of p (p > 1) first sets of technical documents D = {D1, D2, …, Dp}; making a request k to an AI model (k) to input at least a part of a technical document Di for 1 ≤ i ≤ p to generate a feature Fi, where the AI model (k) is a generation AI model and the feature Fi includes one or more sentences longer than a single term and not included in the input; making a request l to an AI model (l) to generate q (1 < q < p) classifications C = {C1, C2, …, Cq} that can classify modified features F' in which p features associated with each of the first sets of technical documents or at least any of them are modified, where the AI model (l) is a generation AI model; receiving, from the user terminal, a specification of s (s > 1) second sets of technical documents D = {D1, D2, …, Ds}; making a request m to an AI model (m) to associate, for 1 ≤ j ≤ s, one or more classifications representing the technical document Dj as C'j from among the classifications corresponding to at least a part of the q classifications C or at least any of them; and receiving and storing, for 1 ≤ j ≤ s, the classification C'j associated with the technical document Dj.

[0048] Also, a 44th aspect of the present invention is a method for analyzing technical documents, including the steps of receiving, from a user terminal, a specification of p (p > 1) first sets of technical documents; making a request to a generation AI model to generate q classifications C that can classify the first sets of technical documents based on p features associated with each of the first technical documents; transmitting at least a part of the generated q classifications C to the user terminal; receiving, from the user terminal, a specification of s (s > 1) second sets of technical documents D = {D1, D2, …, Ds}; and receiving, from the user terminal, a request to determine, from among the s technical documents D, technical documents related to a classification specified by the user from among at least a part of the q classifications C or classifications corresponding thereto, and transmitting a request to the AI model to make the determination.

[0049] Further, a forty-fifth aspect of the present invention is a method for analyzing technical documents, comprising the steps of: receiving, from a user terminal, a specification of p (p>1) first sets of technical documents D={D1, D2, …, Dp}; for i ranging from 1 to p, making a request k to an AI model (k) to generate a feature Fi by inputting at least a part of the technical document Di, wherein the AI model (k) is a generation AI model and the feature Fi includes one or more sentences longer than a single term and not included in the input; making a request l to an AI model (l) to generate q (1<q<p) classifications C={C1, C2, …, Cq} that can classify the p features associated with each of the first sets of technical documents or at least some of the modified features F' obtained by modifying at least some of them, wherein the AI model (l) is a generation AI model; sending at least a part of the generated q classifications C to the user terminal; receiving, from the user terminal, a specification of s (s>1) second sets of technical documents D={D1, D2, …, Ds}; and receiving, from the user terminal, a request to determine, from among the s technical documents D, the technical documents related to the classification specified by the user from at least a part of the q classifications C or the corresponding classifications, and making a request m to an AI model (m) to make the determination.

[0050] In addition, a forty-sixth aspect of the present invention is a method for analyzing technical documents, comprising the steps of acquiring designations of p (p>1) technical documents D = {D1, D2, ..., Dp} to be analyzed; making a request k to an AI model (k) for i between 1 and p inclusive, requesting that the AI ​​model (k) input at least a portion of the technical documents Di and generate features Fi, wherein the AI ​​model (k) is a generative AI model and the features Fi include one or more sentences longer than a single term that are not included in the input; making a request l to an AI model (l) to assign one of q classifications C = {C1, C2 ..., Cq} to each of p features F = {F1, F2 ..., Fp} associated with each of the p technical documents D or to modified features F' obtained by modifying at least one of them; and receiving and storing the classifications assigned to the technical documents Di for i between 1 and p inclusive.

[0051] Furthermore, a 47th aspect of the present invention is the method of the 46th aspect, wherein the q classifications C represent relevance to a given description of each of the feature F or the modified feature F′.

[0052] Furthermore, the forty-seventh or forty-eighth aspect of the present invention is the method of the forty-seventh aspect, wherein q is 2 or 3.

[0053] A 49th aspect of the present invention is the method according to any one of the 46th to 48th aspects, wherein the AI ​​model (l) is a generative AI model.

[0054] A 50th aspect of the present invention is the method of the 49th aspect, wherein the request l includes specifying that the reason for assigning the classification to at least one of the q classifications C be described.

[0055] Further, a 51st aspect of the present invention is a program for causing a computer to execute a method for analyzing technical documents, the method comprising: receiving, from a user terminal, a designation of p (p > 1) technical documents D = {D1, D2, …, Dp} to be analyzed; making a first request to a first AI model to generate q classifications C = {C1, C2, …, Cq} capable of classifying the p technical documents D based on p features F = {F1, F2, …, Fp} associated with each of the p technical documents D, the first AI model being a generation AI model; for i from 1 to p, making a second request to a second AI model to associate, as C'i, one or more classifications represented by the technical document Di from among the classifications corresponding to the q classifications C or at least a part thereof; and for i from 1 to p, receiving and storing the classification C'i associated with the technical document Di.

[0056] Further, a 52nd aspect of the present invention is a program for causing a computer to execute a method for analyzing technical documents, the method comprising: receiving, from a user terminal, a designation of p (p > 1) technical documents D = {D1, D2, …, Dp} to be analyzed; for i from 1 to p, making a request k to an AI model (k) to input at least a part of the technical document Di to generate a feature Fi, the AI model (k) being a generation AI model, the feature Fi including one or more sentences longer than a single term not included in the input; and making a request l to an AI model (l) to generate q classifications C = {C1, C2, …, Cq} (1 < q < p) capable of classifying the p features F = {F1, F2, …, Fp} associated with each of the p technical documents D or a modified feature F' in which at least any of them is modified, the AI model (l) being a generation AI model.

[0057] In addition, a fifty-third aspect of the present invention is a program for causing a computer to execute a method for analyzing technical documents, the method including the steps of receiving, from a user terminal, designation of p (p>1) technical documents D={D1, D2, ..., Dp} to be analyzed; requesting a generative AI model to generate q classifications C={C1, C2..., Cq} capable of classifying the p technical documents D based on p features F={F1, F2..., Fp} associated with each of the p technical documents D, in association with each technical document; and receiving and storing the classifications associated with technical documents Di for i between 1 and p.

[0058] In addition, a 54th aspect of the present invention is a program for causing a computer to execute a method for analyzing technical literature, the method comprising the steps of: receiving from a user terminal the designation of a first set of p (p>1) technical literature; requesting a generative AI model to generate q classifications C capable of classifying the first set of technical literature based on p features associated with each of the first technical literature; receiving from the user terminal the designation of a second set of s (s>1) technical literature D = {D1, D2, ..., Ds}; for i between 1 and s, requesting the AI ​​model to associate one or more classifications C'i representing technical literature Di from among the q classifications C or classifications corresponding to at least some of them; and receiving and storing the classifications C'i associated with technical literature Di, for i between 1 and s.

[0059] In addition, a 55th aspect of the present invention is a program for causing a computer to execute a method for analyzing technical literature, the method including the steps of receiving, from a user terminal, designation of a first set of p (p>1) technical literatures; requesting a generative AI model to generate q classifications C capable of classifying the first set of technical literatures based on p features associated with each of the first technical literatures; transmitting at least some of the generated q classifications C to the user terminal; receiving from the user terminal designation of a second set of s (s>1) technical literatures D = {D1, D2, ..., Ds}; and receiving from the user terminal a request to determine from the s technical literatures D which technical literature is related to a classification specified by the user from at least some of the q classifications C or a classification corresponding thereto, and transmitting a request to the AI ​​model to make the determination.

[0060] In addition, a fifty-sixth aspect of the present invention is a program for causing a computer to execute a method for analyzing technical documents, the method including the steps of: acquiring specification of p (p>1) technical documents D = {D1, D2, ..., Dp} to be analyzed; and issuing a request k to an AI model (k) for i between 1 and p inclusive, requesting that an AI model (k) input at least a portion of a technical document Di and generate features Fi, wherein the AI ​​model (k) is a generative AI model and the features Fi include one or more sentences longer than a single term that are not included in the input; issuing a request l to an AI model (l) to assign one of q classifications C = {C1, C2 ..., Cq} to each of p features F = {F1, F2 ..., Fp} associated with each of the p technical documents D or to modified features F' in which at least one of them is modified; and receiving and storing the classifications assigned to the technical documents Di for i between 1 and p inclusive.

[0061] A fifty-seventh aspect of the present invention is the program according to the fifty-sixth aspect, wherein q is 2 or 3.

[0062] Further, a 58th aspect of the present invention is an apparatus for analyzing technical documents, which receives, from a user terminal, a specification of p (p>1) technical documents D={D1, D2, …, Dp} to be analyzed, and based on p features F={F1, F2, …, Fp} associated with each of the p technical documents D, generates q classifications C={C1, C2, …, Cq} capable of classifying the p technical documents D using a first AI model, which is an AI generation model. A first request is made to the first AI model to generate the q classifications C. For i ranging from 1 to p, a second request is made to a second AI model to associate one or more classifications represented by the technical document Di as C'i from among the classifications corresponding to the q classifications C or at least a part thereof. For i ranging from 1 to p, the received and stored classification C'i associated with the technical document Di is received.

[0063] Further, a 59th aspect of the present invention is an apparatus for analyzing technical documents, which receives, from a user terminal, a specification of p (p>1) technical documents D={D1, D2, …, Dp} to be analyzed. For i ranging from 1 to p, a request k is made to an AI model (k), which is a generation AI model, to input at least a part of the technical document Di to generate a feature Fi. And a request l is made to an AI model (l), which is a generation AI model, to generate q classifications C={C1, C2, …, Cq} (1<q<p) capable of classifying the p features F={F1, F2, …, Fp} associated with each of the p technical documents D or a modified feature F' in which at least any one of them is modified. The feature Fi includes one or more sentences longer than a single term that are not included in the input.

[0064] Furthermore, a sixtieth aspect of the present invention is an apparatus for analyzing technical documents, which receives from a user terminal a specification of p (p>1) technical documents D={D1, D2, ..., Dp} to be analyzed, and requests a generative AI model to generate q classifications C={C1, C2..., Cq} capable of classifying the p technical documents D based on p features F={F1, F2..., Fp} associated with each of the p technical documents D, and receives and stores the classifications associated with the technical documents Di for i between 1 and p.

[0065] Furthermore, a sixty-first aspect of the present invention is an apparatus for analyzing technical documents, which receives from a user terminal a designation of a first set of p (p>1) technical documents, and requests a generative AI model to generate q classifications C that can classify the first set of technical documents based on p features associated with each of the first technical documents, receives from the user terminal a designation of a second set of s (s>1) technical documents D = {D1, D2, ..., Ds}, and requests the AI ​​model, for i between 1 and s, to associate one or more classifications C'i representing technical document Di from among the q classifications C or classifications corresponding to at least some of them, and receives and stores the classifications C'i associated with technical document Di for i between 1 and s.

[0066] Furthermore, a 62nd aspect of the present invention is an apparatus for analyzing technical literature, which receives from a user terminal a designation of a first set of p (p>1) technical literature, and requests a generative AI model to generate q classifications C that can classify the first set of technical literature based on p features associated with each of the first technical literature, transmits at least some of the generated q classifications C to the user terminal, receives from the user terminal a designation of a second set of s (s>1) technical literature D = {D1, D2, ..., Ds}, receives from the user terminal a request to determine from the s technical literature D which technical literature is related to a classification specified by the user from at least some of the q classifications C or a classification corresponding thereto, and transmits a request to make the determination to the AI ​​model.

[0067] In addition, a 63rd aspect of the present invention is an apparatus for executing a method for analyzing technical documents, which obtains specification of p (p>1) technical documents D = {D1, D2, ..., Dp} to be analyzed, and for i between 1 and p, makes a request k to an AI model (k) that is a generative AI model to input at least a portion of the technical document Di to generate features Fi, and makes a request l to an AI model (l) to assign one of q classifications C = {C1, C2 ..., Cq} to each of p features F = {F1, F2 ..., Fp} associated with each of the p technical documents D or to modified features F' in which at least one of them is modified, wherein the features Fi include one or more sentences that are not included in the input and are longer than a single term.

[0068] A sixty-fourth aspect of the present invention is the device according to the sixty-third aspect, wherein q is 2 or 3.

[0069] In addition, a 65th aspect of the present invention is a method for analyzing technical documents, comprising the steps of obtaining specifications of p (p>1) technical documents D={D1, D2, ..., Dp}, creating instructions for i between 1 and p to generate features Fi using at least a part of the technical documents Di as input, making a request k including the instructions to an AI model (k), wherein the AI ​​model (k) is a generative AI model, and storing the generated features Fi in association with the technical documents Di, wherein the features Fi include one or more sentences longer than a single term that are not included in the input.

[0070] A 66th aspect of the present invention is the method according to the 65th aspect, wherein the p features F are of the same predetermined format.

[0071] A 67th aspect of the present invention is the method of the 66th aspect, wherein the predetermined format is two sentences.

[0072] A 68th aspect of the present invention is a method according to any one of the 65th to 67th aspects, wherein each technical document Di is a patent document Di, and at least a part of the technical document Di includes other than the abstract contained in the patent document Di itself.

[0073] A 69th aspect of the present invention is the method of the 68th aspect, wherein at least a part of the technical document Di is at least a part of the patent document Di up to the Summary of the Invention section, the Brief Description of the Drawings section, or the Mode for Carrying Out the Invention section, and the predetermined format is a format beginning with "The problem is to" or an expression synonymous therewith.

[0074] Furthermore, a 70th aspect of the present invention is a method according to any one of aspects 65 to 69, further comprising the step of making a request l to an AI model (l) to generate one or more new problems to be solved based on at least a portion of the generated p features F, wherein the AI ​​model (l) is a generative AI model, and each of the one or more new problems comprises one or more sentences longer than a single term.

[0075] A seventy-first aspect of the present invention is the method of the sixty-fifth aspect, wherein the specification is a search formula for a technical literature database.

[0076] Also, a 72nd aspect of the present invention is a program for causing a computer to execute a method for analyzing technical literature, the method comprising the steps of: obtaining specifications of p (p>1) technical literatures D={D1, D2, ..., Dp}; creating instructions for i between 1 and p to generate features Fi using at least a portion of the technical literature Di as input; making a request k including the instructions to an AI model (k), wherein the AI ​​model (k) is a generative AI model; and storing the generated features Fi in association with the technical literature Di, wherein the features Fi include one or more sentences longer than a single term that are not included in the input.

[0077] A 73rd aspect of the present invention is the program according to the 72nd aspect, wherein the p number of features F are in the same predetermined format.

[0078] In addition, a 74th aspect of the present invention is an apparatus for analyzing technical documents, which is configured to obtain specifications of p (p>1) technical documents D={D1, D2, ..., Dp} for which summaries are to be generated, create instructions for generating features Fi using at least a part of the technical documents Di as input for i between 1 and p, make a request k including the instructions to an AI model (k) which is a generative AI model, and store the generated features Fi, which include one or more sentences longer than a single term that are not included in the input, in association with the technical documents Di.

[0079] A 75th aspect of the present invention is the method according to the 74th aspect, wherein the p features F are of the same predetermined format.

[0080] In addition, a 76th aspect of the present invention is a method for analyzing technical literature, comprising the steps of receiving from a user terminal a description describing technology; making a request k to an AI model (k) to generate one or more search formulas based on the description, wherein the AI ​​model (k) is a generative AI model; determining whether the number of documents obtained by searching a technical literature database using at least one of the generated one or more search formulas or the one or more revised search formulas satisfies a predetermined condition; if the determination result is positive, obtaining data identifying p (p>1) technical literatures D = {D1, D2, ..., Dp} determined by searching the technical literature database using the search formula or the revised search formula that satisfies the predetermined condition; and making a request l to an AI model (l) to input at least a portion of technical literature Di and generate features Fi, for i between 1 and p, wherein the AI ​​model (l) is a generative AI model and the features Fi include one or more sentences that are longer than a single term and are not included in the input.

[0081] A 77th aspect of the present invention is the method of the 76th aspect, further comprising the step of making the request k again if the determination result is negative.

[0082] A 78th aspect of the present invention is the method of the 76th or 77th aspect, wherein the predetermined condition is equal to or greater than a lower limit of 50 or more items and is less than or equal to an upper limit of 100 or less items.

[0083] A 79th aspect of the present invention is the method of the 76th or 77th aspect, wherein the predetermined condition is equal to or greater than one or more lower limit values ​​and less than or equal to 50 or 100 or less upper limit values.

[0084] In addition, an eightieth aspect of the present invention is a program for causing a computer to execute a method for analyzing technical literature, the method comprising the steps of receiving a description describing technology from a user terminal, and making a request k to an AI model (k) to generate one or more search formulas based on the description, wherein the AI ​​model (k) is a generative AI model; determining whether the number of documents obtained by searching a technical literature database using at least one of the generated one or more search formulas or the one or more revised search formulas satisfies a predetermined condition; if the determination result is positive, obtaining data identifying p (p>1) technical documents D = {D1, D2, ..., Dp} determined by searching the technical literature database using the search formula or the revised search formula that satisfies the predetermined condition; and making a request l to an AI model (l) to generate features Fi by inputting at least a portion of technical documents Di for i between 1 and p, wherein the AI ​​model (l) is a generative AI model, and the features Fi include one or more sentences that are longer than a single term and are not included in the input.

[0085] In addition, an 81st aspect of the present invention is an apparatus for analyzing technical literature, which receives from a user terminal a description describing technology, makes a request k to an AI model (k) that is a generative AI model to generate one or more search formulas based on the description, determines whether the number of documents obtained by searching a technical literature database using at least one of the generated one or more search formulas or one or more modified search formulas satisfies predetermined conditions, and if the determination result is positive, obtains data identifying p (p>1) technical documents D = {D1, D2, ..., Dp} determined by searching the technical literature database using the search formula or the modified search formula that satisfies the predetermined conditions, and makes a request l to an AI model (l) that is a generative AI model to input at least a portion of the technical documents Di and generate features Fi for i between 1 and p, wherein the features Fi include one or more sentences that are longer than a single term and are not included in the input.

[0086] In addition, an 82nd aspect of the present invention is a method for analyzing technical literature, comprising the steps of receiving a description describing technology from a user terminal, and making a request k to an AI model (k) to generate one or more search expressions based on the description, wherein the AI ​​model (k) is a generative AI model; obtaining data identifying p (p>1) technical documents D = {D1, D2, ..., Dp} determined by searching a technical literature database using at least one of the generated one or more search expressions or the one or more modified search expressions; and making a request l to an AI model (l) for i between 1 and p to generate features Fi by inputting at least a portion of technical documents Di, wherein the AI ​​model (l) is a generative AI model, and the features Fi include one or more sentences longer than a single term that are not included in the input.

[0087] In addition, an 83rd aspect of the present invention is a program for causing a computer to execute a method for analyzing technical literature, the method comprising the steps of receiving a description describing technology from a user terminal, and making a request k to an AI model (k) to generate one or more search formulas based on the description, wherein the AI ​​model (k) is a generative AI model; obtaining data identifying p (p>1) technical documents D = {D1, D2, ..., Dp} determined by searching a technical literature database using at least one of the generated one or more search formulas or the one or more modified search formulas; and making a request l to an AI model (l) for i between 1 and p inclusive to input at least a portion of technical document Di and generate features Fi, wherein the AI ​​model (l) is a generative AI model, and the features Fi include one or more sentences longer than a single term that are not included in the input.

[0088] In addition, an 84th aspect of the present invention is an apparatus for analyzing technical literature, configured to receive from a user terminal a description describing a technology, make a request k to an AI model (k) that is a generative AI model to generate one or more search formulas based on the description, obtain data identifying p (p>1) technical documents D = {D1, D2, ..., Dp} determined by searching a technical literature database using at least one of the generated one or more search formulas or one or more modified search formulas, and make a request l to the generative AI model (l) for i between 1 and p inclusive to input at least a portion of the technical document Di and generate features Fi, wherein the features Fi include one or more sentences longer than a single term that are not included in the input.

[0089] Also, an 85th aspect of the present invention is a method for researching prior art documents, comprising the steps of receiving a description describing a technology from a user terminal, making a request k to an AI model (k) to generate one or more search formulas based on the description, wherein the AI ​​model (k) is a generative AI model, and obtaining data identifying p (p>1) technical documents D={D1, D2, ..., Dp} determined by searching a technical document database using at least one of the generated one or more search formulas or one or more modified search formulas, and for i between 1 and p, obtaining data identifying p of technical documents D={D1, D2, ..., Dp} determined by searching a technical document database using at least one of the generated one or more search formulas or one or more modified search formulas. The method includes the steps of: making a request l to an AI model (l) to generate features Fi from at least a portion of an input, wherein the AI ​​model (l) is a generative AI model, and the features Fi include one or more sentences longer than a single term that are not included in the input; making a request m to an AI model (m) to assign, for i between 1 and p inclusive, one of q (q>1) classifications C={C1, C2..., Cq} as Ci', which represent the relevance of the description of the features Fi associated with the technical document Di; and receiving and storing the classifications Ci' assigned to the technical document Di, for i between 1 and p inclusive.

[0090] Furthermore, an 86th aspect of the present invention is the method according to the 85th aspect, wherein the p number of features F are of the same predetermined format.

[0091] Furthermore, an 87th aspect of the present invention is the method according to the 85th or 86th aspect, wherein q is 2 or 3.

[0092] Furthermore, an 88th aspect of the present invention is the method of any one of the 85th to 87th aspects, wherein the AI ​​model (m) is a generative AI model.

[0093] Furthermore, an 89th aspect of the present invention is the method of the 88th aspect, wherein the AI ​​model (m) is identical to the AI ​​model (l).

[0094] Furthermore, a 90th aspect of the present invention is a method of any one of the 85th to 88th aspects, further comprising a step of transmitting to the user terminal classification information including at least a portion of the p classifications C' or literature information identifying one or more technical literature to which they are assigned.

[0095] Furthermore, a 91st aspect of the present invention is a method of the 90th aspect, wherein the transmission of classification information including at least a part of the p classifications C' or literature information identifying one or more technical literature to which they are assigned to the user terminal is the transmission of viewing screen display information for displaying a viewing screen for the classification information or the literature information on the user terminal.

[0096] A 92nd aspect of the present invention is the method of the 91st aspect, wherein the viewing screen includes only one or more technical documents out of the p technical documents D that have been assigned any one of the q classifications C.

[0097] A 93rd aspect of the present invention is the method according to the 91st or 92nd aspect, wherein the viewing screen is capable of displaying a search formula that defines p technical documents D.

[0098] A ninety-fourth aspect of the present invention is the method according to any one of the ninety-first to ninety-third aspects, wherein said viewing screen includes a document in the form of an invention notification form or an invention proposal form.

[0099] A 95th aspect of the present invention is the method of the 94th aspect, wherein the document includes the description or a part thereof, or a description corresponding thereto.

[0100] Furthermore, a 96th aspect of the present invention is the method of the 95th aspect, wherein the document further includes document information identifying one or more technical documents among the p technical documents D that have been assigned any one of the q classifications C.

[0101] Furthermore, a 97th aspect of the present invention is a method according to any one of the 94th to 96th aspects, wherein the document further includes features associated with one or more technical documents among the p technical documents D that have been assigned any one of the t classifications C.

[0102] A 98th aspect of the present invention is the method according to any one of the 94th to 97th aspects, wherein the document further includes a search formula that defines p technical documents D.

[0103] A 99th aspect of the present invention is a program for causing a computer to execute a method for researching prior art documents, the method comprising the steps of: receiving a description describing a technology from a user terminal; making a request k to an AI model (k) to generate one or more search formulas based on the description, the AI ​​model (k) being a generative AI model; and acquiring data identifying p (p>1) technical documents D={D1, D2, ..., Dp} determined by searching a technical document database using at least one of the generated one or more search formulas or the one or more modified search formulas; and The method includes the steps of: making a request l to an AI model (l) to input at least a part of technical document Di and generate features Fi for i, wherein the AI ​​model (l) is a generative AI model, and the features Fi include one or more sentences longer than a single term that are not included in the input; making a request m to an AI model (m), for i between 1 and p and including p, to assign as Ci' one of q (q > 1) classifications C = {C1, C2 ..., Cq} that represent the relevance of the description of the features Fi associated with the technical document Di; and receiving and storing the classification Ci' assigned to the technical document Di for i between 1 and p and including p.

[0104] In addition, a hundredth aspect of the present invention is an apparatus for researching prior art documents, the apparatus being configured to receive from a user terminal a description describing technology, and make a request k to an AI model (k) that is a generative AI model to generate one or more search formulas based on the description, and to obtain data identifying p (p>1) technical documents D = {D1, D2, ..., Dp} determined by searching a technical document database using at least one of the generated one or more search formulas or one or more modified search formulas, and to input, for i between 1 and p, at least a portion of a technical document Di, and make a request l to an AI model (l) that is a generative AI model to generate features Fi, the features Fi including one or more sentences longer than a single term that are not included in the input, and a request m to an AI model (m) that assigns, as Ci', one of q (q>1) classifications C = {C1, C2 ..., Cq} that represent the relevance of the feature Fi associated with the technical document Di to the description, and to receive and store the classifications Ci' assigned to the technical documents Di for i between 1 and p.

[0105] According to one aspect of the present invention, q classifications C capable of classifying p specified technical documents D are generated in response to a request to a generative AI model based on p features F associated with each of the p specified technical documents D, and from the generated classifications C, those representing each technical document are stored in association with the technical document, thereby reducing the labor-intensive task of classification and enabling a variety of analyses.

[0106] It is also noted that, according to another aspect of the present invention, different effects described herein may be obtained alternatively or in addition.

[0107] 1 is a diagram illustrating a system according to a first embodiment of the present invention; 2 is a diagram illustrating a flow of a method according to the first embodiment of the present invention; 3 is an example of a first request to a first generative AI model according to the first embodiment of the present invention; 4 is a diagram illustrating a flow of a method according to a seventh embodiment of the present invention; and 5 is an example of an eighth request to an eighth generative AI model according to the seventh embodiment of the present invention.

[0108] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0109] (First Embodiment) FIG. 1 shows a system according to a first embodiment of the present invention. To analyze technical literature, the device 100 communicates with a user terminal 110 used by a user and a platform 120 that provides a generative AI model via an IP network such as the Internet. The device 100 also communicates with a technical literature database 130, such as a patent database, to search for technical literature as needed. While the AI ​​model is described as being provided by the platform 130 that can communicate with the device 100, an application for providing the AI ​​model can also be executed on the device 100, allowing the device 100 to provide the AI ​​model. Although the technical literature database 130 is described as a database separate from the device 100, it may also be a database included in the device 100.

[0110] The device 100 includes a communication unit 101 such as a communication interface, a processing unit 102 such as a processor or CPU, and a storage unit 103 including a storage device or storage medium such as a memory or hard disk. The device 100 may be configured by executing a program for each process or operation in the processing unit 102. The device 100 may include one or more devices, computers, or servers. The program may include one or more programs, and may be recorded on a computer-readable storage medium to form a non-transitory program product. The program may be stored in the storage unit 103 or in a storage device or storage medium 104 accessible from the device 100 via an IP network, and instructions included in the program may be executed by at least one processor in the processing unit 102. Data described below as being stored in the storage unit 103 may also be stored in the storage device or storage medium 104, and vice versa. For convenience, the storage device or storage medium 104 is illustrated in FIG. 1 as a database.

[0111] First, a designation of p (p>1) technical documents D = {D1, D2, ..., Dp} to be analyzed is received from the user terminal 110 (S201). While not exhaustive, p is any positive integer greater than 1, such as 2, 3, 4, 5, 10, 25, 50, 75, 100, 250, 500, or 1000. The designation can be performed, for example, by uploading technical document D using an upload screen provided by the device 100 or by uploading a list for identifying technical document D. Alternatively, the designation can be performed by selecting a technical document to be analyzed from an uploaded set of technical documents or a list for identifying them. Alternatively, technical document D can be indirectly designated by a search query for accessing the technical document database 130 and searching for a set of patent documents. In this case, for example, a list for identifying technical document D can be obtained in response to a search request including the search query using an API. The device 100 may not be a separate device from the user terminal 110, and each process or operation may be performed in the user terminal 110. In this case, the designation is input to the user terminal 110, and is read and acquired by the device 100. In either case, it is sufficient to determine the document to be analyzed.

[0112] Also, if the specification is made by uploading a list, it is conceivable that the p features F={F1, F2..., Fp} to be used in the next step will be included in the specification. Technical document D may include documents of, for example, 50 KB or more, or even 100 KB or more, 250 KB or more, 300 KB or more, 500 KB or more, 1000 KB or more, 1500 KB or more, or 2000 KB or more. Also, all documents included in technical document D may be documents of 50 KB or more, or even 100 KB or more, 250 KB or more, 300 KB or more, or 500 KB or more.

[0113] If each document is a patent document, such a list may include, for example, the patent application number, patent number, and / or URL where the contents of each patent document can be accessed, and may further include the filing date, applicant, title of the invention, and / or claims. If the list includes feature F, one useful example is to define each feature as one or more sentences summarizing the respective problem of patent document D. For clarity, in this specification, "sentence" does not necessarily have a subject and a predicate as components, nor is it a single term.

[0114] If each document is a technical document other than a patent document, it may include at least one of document information such as bibliographic information and a URL that allows access to the contents of each technical document, and may also include at least one of the publication date or posting date, author, title, and abstract.

[0115] Next, the device 100 requests (also referred to as the "first request") the generative AI model (also referred to as the "first AI model") to generate q classifications C = {C1, C2 ..., Cq} that can classify the p technical documents D based on p features F = {F1, F2 ..., Fp} associated with each of the p technical documents D (S202).

[0116] FIG. 3 shows an example of a portion of code for a first request to the first AI model according to the first embodiment of the present invention. The example in FIG. 3 is a portion of code written in the programming language Python. By executing this code, the OpenAI API can be called and the first AI model provided on the platform 120 can generate classification C. Here, the variable “engine” represents the type of generative AI model to be used, the variable “temperature” represents the output diversity, and the variable “max_tokens” represents the upper limit on the number of tokens to be output. “Token” is the unit for dividing the input described in the variable “prompt.” The OpenAI API is an example, and other APIs may be used. Furthermore, the programming language Python is an example, and other languages ​​may be used. The code for making the first request is stored in the storage unit 103, and the device 100 acquires this code and executes it, including instructions obtained by setting values ​​of the feature F and other variables included in the code.

[0117] In the example of Figure 3, the generative AI model is requested to generalize the p features F set in the variable "features" as an engineer in the field set in the variable "specifiedField" and classify them into q classifications C, which are less than p. In addition to or instead of changing the description of the variable "prompt," the conditions (also referred to as "first conditions") imposed on the first generative AI model when generating classification C can be specified by setting values ​​for the variables "specifiedField" and "specifiedConditions." This allows the generated classification C to be varied. Changing the description of the variable "prompt" requires rewriting the code stored in the memory unit 103. However, in addition to setting the variable "features," receiving the specification of the first condition or a part of it from the user terminal 110 simultaneously with or before or after the specification of technical document D and setting it as a variable in the code can make the executed code more likely to produce output useful to the user. The variable "specifiedConditions" may be set to one or more expressions to be used for at least one of the generated classifications C. The first condition may be set as a variable in the code as a whole, or a variable other than that included in the first condition, such as the variable "specifiedField", may be included in the description of the variable "prompt", and the variable may be set according to the specification received from the user terminal 110.

[0118] The first condition imposed on the first AI model may include the range of the value of q, q being less than p, or a range or value where q is greater than or equal to a lower limit of 2 or more and less than or equal to an upper limit of 10 or less. Fixing the value of q, i.e., the number of classifications to be classified, may restrict the generated classification C due to human preconceptions, so it may be acceptable to fix it. However, if the goal is to obtain classifications that are difficult for humans to identify, it is preferable not to fix it. That is, the number of classifications constituting classification C may be specified as the first condition as q being greater than 1 and less than p, or the number of generated classifications may satisfy this condition. On the other hand, including one or more classifications to be included in classification C as conditions ensures that the entire classification C is generated based on that assumption, thereby working to produce results that are beneficial to the user. Including a range of the value of q in the first condition is not required, as it produces results closer to the user's expectations without fixing the value of q, but it is easier and more beneficial for the user. If the technical document is a patent document, the first condition may include specifying the use of knowledge in the technical field to which patent document D belongs. Furthermore, since the invention according to this embodiment can handle situations where a given patent classification cannot fully analyze the entire document, it is possible to assume that this will be the case without explicitly stating it. Therefore, the first condition may include the requirement that the generated classification C does not include the patent classification. Furthermore, if the first condition includes the specification that q classifications C contain expressions meaning "others," technical documents that are difficult to classify will be classified as "others," thereby improving the efficiency of noise removal. By excluding technical documents that fall under the noise category in this way and then generating classifications again, accuracy can be improved. Furthermore, if p technical documents are to be broadly classified, it may be preferable to set the value of q to 2 or 3.

[0119] Because the output of the generative AI model is not necessarily the same each time, the first condition may be set to determine each of the categories that make up category C based on multiple output results. Alternatively, the first request may be made multiple times to the first generative AI model, and the device 100 may determine each of the categories that make up category C based on multiple responses to those requests.

[0120] In the example of Fig. 3, the device 100 receives the generated classification C (S203). In this way, the generated classification C or a part thereof may be received, or the necessary data may be transmitted to the device 100 at a stage where subsequent processing is completed in the platform 120. In the following, classification C is associated with technical document D, but the association is assumed to be performed separately. The fact that the user can obtain the generated classification C in itself significantly reduces the amount of work.

[0121] The device 100 requests (also referred to as a "second request") the AI ​​model (also referred to as a "second AI model") to associate one or more classifications representing technical documents Di from among q classifications C as C'i with the technical documents, for i between 1 and p (S204). As with the first request, the second request can be made by preparing code in advance and having the device 100 execute the code containing instructions obtained by setting the necessary variables.

[0122] While Figure 2 distinguishes between a first AI model and a second AI model, they may be considered the same generative AI model. In this specification, "AI model" refers to a machine learning model trained to predict an output for an input, and "generative AI model" refers to a large-scale language model (LLM) trained using text data to generate an output not included in the input. LLMs that apply the Transformer architecture are particularly preferred as generative AI models, but the name of the architecture is expected to change as technology advances. Therefore, in this specification, "Transformer architecture" encompasses architectures that utilize one or more features of the Transformer architecture or improvements thereof. In this specification, whether "generative AI models" are the same is determined by whether the type of generative AI model specified by the user is the same. In the example of Figure 3, if the value of the variable "engine" is the same, they are considered to be the same generative AI model. If the first AI model and the second AI model are not the same, they may be provided on the same platform 120. In the example of Figure 2, the second request is made after the first request, but if the first and second AI models are the same generative AI model or are generative AI models provided on the same platform, these requests may be made with a single API call. Also, the first request for the first model may be divided into multiple requests and realized with multiple API calls, and the same applies to the second model.

[0123] The device 100 then receives and stores the classifications C′i associated with the technical documents Di for i between 1 and p (S205). For example, if the specification of the technical document D is a list upload, a new column may be created in the list to store the classifications C′i associated with the technical documents Di for each i.

[0124] Thereafter, the device 100 may transmit at least a portion of the p classifications C' = {C'1, C'2 ... , C'p} to the user terminal 110 (S206). Transmission of at least a portion of the classifications C' to the user terminal 110 may be by transmitting viewing screen display information for displaying a viewing screen on the user terminal 110 of classification information including at least a portion of the p classifications C' or document information identifying one or more documents to which at least a portion of the classifications C' is assigned. The viewing screen display information may be transmitted, for example, as an HTML file, and loaded by the web browser of the user terminal 110 to display the viewing screen on the display screen of the user terminal 110. Furthermore, if classification information including at least a portion of the generated q classifications C or document information identifying one or more documents to which at least a portion of the classifications C is assigned is transmitted to the user terminal 110 instead of the p classifications C' associated with technical document D, the classification information may be viewable on the viewing screen.

[0125] Thus, according to the first embodiment of the present invention, q classifications C capable of classifying p specified technical documents D are generated in response to a request to the generative AI model based on p features F associated with each of the p specified technical documents D, and from the generated classifications C, those representing each technical document are stored in association with the technical document, thereby reducing the labor-intensive task of classifying technical documents and enabling a variety of analyses.

[0126] This is particularly useful when the technical documents are patent documents, since various analyses are possible even without a specific patent classification. Furthermore, by using each of Feature F as a summary or approximate summary of the problems described in each patent document in Patent Document D, it becomes easy to visualize what problems have been attempted to solve in past research and development, highlighting problems that have not yet been adequately solved and need to be solved. This not only enables analysis of the past but also analysis for the future, bringing new value to patent document analysis. Obtaining classifications that are difficult for humans to discover can lead to the identification of new problems, so being able to specify the number of classifications generated, variables representing the diversity of the output, and other conditions is particularly useful. Even for technical documents other than patent documents, it becomes possible to visualize what subjects have focused on in past research and development. Here, "subject" is used as a term that includes "problem."

[0127] If each of Feature F is made into a summary or an approximate summary of at least one of the claims described in each patent document included in Patent Document D, it becomes easy to visualize what inventions have been made in research and development up to now. Even in the case of technical documents other than patent documents, by making each of Feature F into a summary of the part including the result or conclusion of the technical document, it becomes possible to visualize what results have been obtained in research and development up to now.

[0128] Second Embodiment In the first embodiment, an example was given in which feature F was included in a list and provided to device 100. However, before requesting the first generative AI model to generate classification C, a request (also referred to as a "third request") may be made to the generative AI model (also referred to as a "third AI model") to generate feature F by expressing at least a portion of technical document Di in a length less than or equal to a predetermined length, such as 125, 100, 75, 50, or 25 English words, or 250, 200, 150, 100, or 50 Japanese characters, for i between 1 and p. Furthermore, feature F may be expressed in a length greater than or equal to a predetermined length, such as 10, 25, or 50 English words, or 20, 50, or 100 Japanese characters, so as to fully reflect the content of technical document D. Feature F may be expressed in languages ​​other than English, such as French or German, or in languages ​​other than Japanese, such as Chinese or Korean. The expressed features Fi may include one or more sentences longer than a single term. Furthermore, a combination of the above-mentioned upper limit and these lower limits may be included as a third condition in the third requirement, but the expressed features Fi may also satisfy the above-mentioned length condition without being explicitly included in the third condition. By using a generative AI model, it is possible to properly grasp the content of technical documents Di using expressions that are not included in the input to the generative AI model and are not described in the technical documents Di, rather than simply extracting words, terms, or expressions described in the technical documents Di.

[0129] The third AI model may be the same as or provided on the same platform or device as at least one of the first AI model and the second AI model. The third request for the third model may be divided into multiple requests and implemented by multiple API calls. Similarly to the first request, the third request can be implemented by preparing code in advance and having the device 100 execute the code containing instructions created by setting the necessary variables. A user may wish to modify any of the generated p features F. In such a case, a first request may be made to generate q classifications C = {C1, C2..., Cq} capable of classifying the modified feature F' obtained by modifying feature F', or more generally, by modifying at least one of the p features F'. Here, the modified F'i includes one or more sentences longer than a single term. The modifications may be manual modifications input by the user, modifications made by the device 100 according to predetermined rules, or other modifications.

[0130] For example, it is conceivable to summarize the subject of technical document Di to a predetermined length or less. The subject of technical document Di may be within the first half or even the first third of the document's overall length. Alternatively, the subject may be from the beginning or nearby of the document, provided that the number of tokens does not exceed the upper limit of the number of tokens that can be included in the third requirement. This is because the subject of a document is often found near the beginning. The summary obtained in this way can be said to be at least an approximate summary of the subject of technical document Di.

[0131] If the technical document is a patent document, at least a portion of the technical document Di to be abstracted may be the portion of the patent document up to or including the Summary of the Invention section, the Brief Description of the Drawings section, or the Detailed Description section. Furthermore, at least a portion of the technical document Di to be abstracted may be the portion including one or more claims of the patent document. Since the names of each section may vary depending on the country or application, the portion to be abstracted may be acquired while taking such variations into account. Because the abstract contained in the patent document itself is unlikely to adequately describe the contents of the document, generating features Fi using only the abstract contained in the patent document itself is not included in this embodiment. In other words, at least a portion of the patent document Di to be abstracted may include parts other than the abstract contained in the patent document Di itself.

[0132] The third requirement may include specifying a third condition to be imposed on the third AI model when generating the p features F. For example, the third condition may be to extract a subject or problem described in at least a portion of the abstract. By explicitly specifying "extracting a problem" in this way, the features F generated by the third AI model can be closer to the properties desired by the user. More generally, the third condition may be used to specify properties to be extracted from at least a portion of the abstract. While the generation of features F has been described with the intention of generating a classification C based on the features F, the ability to generate and acquire one or more sentences that adequately grasp the content of technical documents, particularly the problems of the invention described in patent documents, has been difficult to achieve with prior art. Therefore, the second embodiment of the present invention has technical significance in itself. While some patent documents may have structured sections describing the problems, such as a "Problem to be Solved by the Invention" section, it is often difficult to understand the problems of the invention in the document based solely on the description in such sections. Furthermore, many patent documents, such as U.S. patent documents, lack such structure.

[0133] The third condition may also include specifying that the features F be in a predetermined format. For example, the predetermined format may be a format beginning with "Problem to be Solved" or a synonymous expression if at least a portion of the subject matter of the abstract is at least a portion preceding the Summary of the Invention, Brief Description of the Drawings, or Detailed Description. As another example, the predetermined format may be a format beginning with "The Invention" or "Solution to be Solved" or a synonymous expression if at least a portion of the subject matter of the abstract includes one or more claims. The predetermined format may also be two sentences. By generating an abstract for each of the p features F in the same predetermined format, it becomes much easier to identify the nature of the inventions described in each technical document, even if there are a large number of technical documents D.

[0134] As described in the following embodiments and examples, the device 100 may request the AI ​​model to assign one of q predetermined classifications C = {C1, C2..., Cq} to each of the features F'i obtained by correcting the features F generated by the third request or to the corrected features F' obtained by correcting at least one of them. In particular, if the request includes a specification that the q classifications C contain expressions meaning "other," technical documents that are difficult to classify will be classified as "other," thereby improving the efficiency of noise removal. Furthermore, if the value of q is set to 2 or 3, especially 2, the p technical documents D are broadly classified, and less relevant documents can be efficiently removed from the list. By removing technical documents that fall under the "noise" category in this way and then performing the classification generation described in the first embodiment, accuracy can be improved. Additionally, if the AI ​​model is a generative AI model, the classification request may include a specification that the reason for assigning the classification to at least one of the q classifications C is described. The AI ​​model may be provided on the same platform or device as at least one of the first to third AI models. The classification request may be split into multiple requests and implemented by multiple calls to the API, or, as with the first request, this can be achieved by preparing code in advance and having device 100 execute the code containing instructions that are created by setting the necessary variables.

[0135] (Third Embodiment) The viewing screen described in the first embodiment may allow a user using the user terminal 110 to input feedback for one or more classifications included in the p classifications C'. Examples of feedback include a positive or negative selection or evaluation of the one or more classifications, a suggested or specified modification of the one or more classifications, a suggested or specified new classification by the user, and a suggested or specified abstraction or concretization of the p classifications C'. The positive or negative evaluation may be a multi-level evaluation, for example, of three or more levels. Furthermore, feedback may be input for one or more classifications included in the generated q classifications C, rather than for the p classifications C' associated with technical document D. The matters described in this specification with respect to feedback for the p classifications C' are equally applicable to feedback for the q classifications C.

[0136] After receiving the feedback, the device 100 may request (also referred to as a "fourth request") the generative AI model (also referred to as a "fourth AI model") to generate r classifications C"={C1, C2..., Cr} capable of classifying p technical documents D or their features F or modified features F' in response to the feedback. r is, for example, less than p. The fourth AI model may be the same as or provided on the same platform or device as at least one of the first to third AI models. The fourth request to the fourth model may be divided into multiple requests and realized by multiple calls to the API. Similarly to the first request, the fourth request can be made by preparing code in advance and having the device 100 execute the code containing instructions obtained by setting the necessary variables.

[0137] Also, after receiving the feedback, the device 100 may make a second request again using r classifications C''={C1, C2..., Cr} that correspond to at least some of the p classifications C determined in response to the feedback.

[0138] (Fourth Embodiment) The viewing screen described in the first embodiment can display analysis results using at least a portion of the p classes C' or the q classes C as a first axis among multiple analysis axes. A second axis different from the first axis among the multiple analysis axes may be an axis representing changes over time. Furthermore, a second axis different from the first axis among the multiple analysis axes may be an axis representing patent classifications such as IPC, CPC, USPC, and FI. Representing the analysis results as a map using the generated classes C and existing patent classifications allows a user to determine which patent classification symbols in classes C are highly relevant to the user and which are less relevant. This information is useful, for example, when a user creates a new search query for a patent database.

[0139] Fifth Embodiment The device 100 may receive a designation of s (s > 1) technical documents D = {D1, D2, ..., Ds} in addition to the p technical documents D, and may request (also referred to as a "fifth request") the AI ​​model (also referred to as a "fifth AI model") to associate a classification representing the technical document D as C'i from among the generated q classifications C for i between 1 and s. In this case, the second request may or may not be made. The designation of the second set of s technical documents D may also be received before generating the q classifications C. If classifications are regenerated in response to feedback to obtain r classifications C'' or if r classifications C'' corresponding to at least some of the q classifications C have been determined, these classifications may be used. The fifth AI model may be the same as or provided on the same platform as at least one of the first to fourth AI models. The fifth request to the fifth model may be divided into multiple requests and implemented by multiple API calls. Furthermore, the fifth request, like the first request, can be made by preparing code in advance and having the device 100 execute the code containing instructions obtained by setting the necessary variables.

[0140] The device 100 may request the platform 120 to perform a more direct search rather than explicitly associating q classifications C with s technical documents D. That is, the device 100 may receive a request from the user terminal 110 to determine, from among the s technical documents D, technical documents related to at least some of the q classifications C sent to the user terminal 110 or a classification corresponding to the q classifications C, a classification specified by the user, and transmit a request (also referred to as a "sixth request") to an AI model (also referred to as a "sixth AI model") to make the determination. The device 100 may receive and store the determination result. The sixth AI model may be the same as or provided on the same platform or device as at least one of the first to fifth AI models. The sixth request for the sixth model may be divided into multiple requests and implemented by multiple API calls. Similarly to the first request, the sixth request can be realized by preparing code in advance and having the device 100 execute the code containing instructions obtained by setting the necessary variables.

[0141] (Sixth Embodiment) In the first embodiment, the case where Patent Document D is indirectly specified using a search expression was mentioned. However, the device 100 may receive a description describing a technology from the user terminal 110 and request (also referred to as a "seventh request") a generating AI model (also referred to as a "seventh AI model") to generate one or more search expressions based on the description. The seventh AI model may be the same as or provided on the same platform or device as at least one of the first to sixth AI models. The seventh request to the seventh model may be divided into multiple requests and realized by multiple calls to the API. Furthermore, like the first request, the seventh request can be realized by preparing code in advance and having the device 100 execute the code containing instructions obtained by setting the necessary variables.

[0142] The device 100 may request the technical literature database 130 for the number of documents obtained by searching the technical literature database 130 using one or more generated search expressions, and may use the number of documents to evaluate the search expressions. The request to the technical literature database 130 may be made using a search expression that combines all or two or more of the generated search expressions. By making the sixth request multiple times until the number of documents satisfies a predetermined condition, the complex process of creating a search expression can be significantly automated or labor-saving. For example, the condition may be 50 or more and more but less than 100 or less. More generally, the condition may be 50 or more and more than a lower limit of 50 or more and less than an upper limit of 100 or less. Even more generally, the condition may be 1 or more and more than a lower limit and less than an upper limit of 50 or less and less than 100 or less.

[0143] At least a portion of the generated one or more search expressions may be transmitted from the device 100 to the user terminal 110 so that a user using the user terminal 110 can input feedback. This transmission may be by transmitting viewing screen display information for displaying a viewing screen of at least a portion of the generated one or more search expressions in the user terminal 110. The viewing screen display information may be transmitted, for example, as an HTML file, and read by the web browser of the user terminal 110 to display the viewing screen on the display screen of the user terminal 110.

[0144] Examples of feedback include a positive or negative selection or evaluation of the displayed search formula, a suggested or specified modification to the displayed search formula, a suggestion or specification of a new search formula by the user, a suggested or specified abstraction or specification of the displayed search formula, etc. Positive or negative evaluations may be multi-level evaluations, for example, three or more levels. When a suggestion or specification of a modification to a search formula is made, it can be determined whether the generated one or more search formulas or the modified one or more search formulas satisfy the above-mentioned predetermined conditions.

[0145] It is determined whether any of the search expressions satisfies a predetermined condition, and if the determination result is affirmative, the device 100 can obtain data identifying p (p>1) technical documents D={D1, D2, ..., Dp} determined by searching the technical document database 130 using the search expression. For example, the data may be obtained by receiving a list for identifying technical document D from the technical document database 130. If the determination result is negative, the seventh request can be made again.

[0146] Furthermore, the device 100 may request the seventh generative AI model to generate one or more search queries using at least some of the patent classifications, or the top patent classifications, assigned to each of the technical documents D, up to a predetermined rank. If the search query used to identify the patent document D includes one or more patent classifications, the comprehensiveness of the patent classifications can be increased by additionally or alternatively using classifications other than the one or more patent classifications.

[0147] The seventh AI model may be provided by executing an application or program on the user terminal 110. In this case, the specification of technical document D, which has been described as being transmitted from the user terminal 110 to the device 100, may be acquired by the application on the user terminal 110. By doing so, for example, if the technology to be analyzed is undisclosed technology, it is not necessary to transmit a description of the technology to the device 100 or the platform 120, and there is no risk of confidentiality being compromised.

[0148] Focusing on the user terminal 110, in this embodiment, the user terminal 110 provides a description describing the technology to be analyzed to an application running on the user terminal 110, and the application generates one or more search expressions based on the description using a sixth AI model, which is a generative AI model.

[0149] If necessary, in the embodiments described above or below, it is also possible that the generative AI model is provided by an application running on the user terminal 110, as in this embodiment.

[0150] The generation of a search formula and the identification of technical document D based on it described in this embodiment automates or reduces the labor required to form a population of technical documents to be analyzed, and it should be noted that this has technical significance in itself even when separated from the subsequent analysis.

[0151] Seventh Embodiment The identification of p technical documents D by searching the technical document database 130 described in the sixth embodiment can be applied to a search of prior art documents that may be conducted when filing a patent application.

[0152] First, the device 100 receives a description describing a technology from the user terminal 110 (S401). Then, the device 100 makes a seventh request to the seventh AI model to generate one or more search expressions based on the description (S402). The seventh generative AI model generates one or more search expressions (S403) and transmits them to the device 100 (S404).

[0153] Next, the device 100 makes a search request to the technical literature database 130 using at least one of the received one or more search expressions (S405), and obtains data identifying p technical literatures D in response (S406).

[0154] Then, the device 100 makes a third request to the third AI model to generate features F by summarizing at least a portion of the technical documents D for i between 1 and p (S407). The third AI model generates the features F (S408) and transmits them to the device 100 (S409).

[0155] The device may further request (also referred to as the "eighth request") the AI ​​model (hereinafter also referred to as the "eighth AI model") to assign one of q classifications C = {C1, C2..., Cq} representing the relevance of a feature Fi associated with technical document Di to the description, for i between 1 and p (also referred to as the "eighth request"). The eighth AI model assigns one of q classifications C = {C1, C2..., Cq} to feature Fi as classification Ci' (S411) and transmits it to the device 100 (S412). The eighth AI model may be provided on the same platform or device as at least one of the first through seventh AI models. The eighth request to the eighth model may be divided into multiple requests and realized by multiple API calls. Similarly to the first request, the eighth request can be made by preparing code in advance and having the device 100 execute the code containing instructions obtained by setting the necessary variables.

[0156] Here, it is preferable that q of the q classifications C is 2 or 3. This is because a two- or three-level evaluation is often sufficient, as it is only necessary to evaluate the relevance to the description representing the technology received from the user terminal 110. When q is 2, for example, classification C may be {Y,N}, where "Y" indicates a relatively high relevance and "N" indicates a relatively low relevance. If the eighth AI model is a generative AI model, the eighth request may include a specification that describes the reason for assigning at least one of the q classifications C. For example, the eighth AI model may be requested to describe the reason why "Y" in classification C = {Y,N} is relatively high relevance. It is also possible not to explicitly assign classification "N," which indicates a relatively low relevance.

[0157] While Figure 4 shows an example in which classification is repeatedly requested for each feature F, classification may be performed for each feature included in feature F with a single API call. Also, in the example of Figure 4, the generation of feature Fi and classification for feature Fi are realized with separate requests, but if the third AI model and the eighth AI model are the same generative AI model or are generative AI models provided on the same platform, these requests may be made with a single API call.

[0158] The device 100 then receives and stores the classifications Ci' assigned to the technical documents Di for i between 1 and p, and transmits at least a portion of the classifications Ci' to the user terminal as needed (S413). This transmission may be transmission of viewing screen display information for displaying a viewing screen on the user terminal 110 containing classification information including at least a portion of the p classifications C' or document information identifying one or more documents to which the at least a portion of the classifications C' is assigned. The viewing screen display information may be transmitted, for example, as an HTML file, and may be read by the web browser of the user terminal 110 to display the viewing screen on the display screen of the user terminal 110.

[0159] The browsing screen may include only one or more technical documents that have been assigned one of the q classifications C among the p technical documents D. For example, only documents that have been assigned classification "Y" may be presented to the user. Furthermore, the browsing screen may include a search formula that defines the p technical documents D. In a prior art document search, the search formula used to form the population is of great importance in terms of reliability. Not only can the search results be obtained that are related to the technology being searched, but the search formula used for that purpose can also be known.

[0160] Since the results of the prior art literature search as described above are often described in an invention proposal or invention notification form that is prepared by a company to determine whether or not to apply for a patent for a created invention, the viewing screen may be configured to include a document in the form of an invention notification form or invention proposal. More specifically, the document may include a description received from the user terminal 110, a part thereof, or a description corresponding thereto, and may include literature information identifying one or more technical literatures to which any one of the q classifications C among the p technical literatures D is assigned, one or more features associated with the one or more technical literatures, and a search formula defining the p technical literatures D, or any combination thereof.

[0161] FIG. 5 shows an example of a portion of code for an eighth request to an eighth AI model according to the seventh embodiment of the present invention. The example in FIG. 8 is a portion of code written in the programming language Python. By executing this code, the OpenAI API can be called and the eighth AI model provided on the platform 120 can be assigned a classification. Here, the variable "model" represents the type of generative AI model to be used. The OpenAI API is an example, and other APIs may be used. Furthermore, the programming language Python is an example, and other languages ​​may be used. The code for making the eighth request is stored in the storage unit 103, and the device 100 acquires this code and executes code including instructions obtained by setting values ​​to variables included in the code.

[0162] In the example of Figure 5, the generative AI model sets the description of the technology received from the user terminal 110 as the value of the variable "description", and sets the summary of the problem associated with each of the p technical documents D as the value of the variable "problem" for each document. By repeatedly executing this code p times, classification is assigned to the p technical documents D. Also, in the example of Figure 5, the value of q is 2, which indicates that if classification "Y" is assigned, the reason for this should be included in the response.

[0163] By changing the variable "problem" to the variable "claim" and setting the abstracts of the claims associated with p technical documents D for each document, the code can be modified appropriately based on the desired analytical perspective to enable classification from the perspective of the claims. For example, modifying the code to assign a classification indicating relatively high relevance when the technology described in the description provided from the user terminal 110 is likely to use the invention defined by the claim abstract can lead to more efficient infringement prevention investigations rather than the pre-filing investigations described above. Furthermore, by setting the value of the variable "problem" for each document as the value of p technical documents D and modifying the code to assign a classification indicating relatively high relevance when the technology described in the description provided from the user terminal 110 is likely to be a solvable problem, it is possible to discover potential customers to whom the company's technology can be provided.

[0164] Eighth Embodiment As described in the first embodiment, the device 100 can request (hereinafter also referred to as a "ninth request") an AI model (hereinafter also referred to as a "ninth AI model") to generate one or more new problems to be solved based on at least a portion of the generated features F, where the features F are summaries or approximate summaries of the subjects described in each of the p technical documents D. The new problems generated include one or more sentences longer than a single term. The ninth AI model may be provided on the same platform or device as at least one of the first to eighth AI models. The ninth request to the ninth model may be divided into multiple requests and realized by multiple calls to an API. Similarly to the first request, the ninth request can be realized by preparing code in advance and having the device 100 execute the code containing instructions created by setting the necessary variables.

[0165] At least some of the p features F used in the ninth requirement can be assigned any of the q classifications C assigned to the p features F. In this way, by limiting one or more problems that serve as the context for generating a new problem, it is possible to generate a new problem that can be used as a practical reference and encourage the creation of a new invention.

[0166] (Ninth Embodiment) The viewing screen described in the first, third, sixth, and seventh embodiments can display document information identifying one or more patent documents to which at least some of the q classifications C have been assigned, and the document information may include drawings from each patent document. For example, the drawings to be displayed can be extracted and selected based on their relevance to the claims of each patent document. Specifically, by inputting a claim or an abstract of a patent document from a database in which one or more drawings from each patent document are embedded, highly relevant drawings can be output.

[0167] Example 1: A population of 50 U.S. patent documents containing the term "metaverse" in the claims was searched. A spreadsheet containing the application number, filing date, publication number, applicant or patent owner, and invention title of each of the 50 patent documents was created. For each of the 50 patent documents listed in the spreadsheet, the text was read from the OCR-processed PDF file of the publication up to the page containing the detailed description section or the corresponding page, and a request was made to GPT-4™, a generative AI model provided by OpenAI as an API, to generate a summary of the problem described in the text. This reading and request were performed by executing code written in the programming language Python. The generated problem summary was added to the spreadsheet. The resulting spreadsheet is shown below.

[0168]

[0169]

[0170]

[0171]

[0172]

[0173]

[0174]

[0175]

[0176]

[0177]

[0178] The first 20 of the 50 patent documents included in the spreadsheet were selected as technical document D to be analyzed, and when the generating AI model was GPT-4, a classification capable of classifying the 20 technical documents D was generated using codes corresponding to those shown in Figure 3. A request was also made to associate a classification representing each technical document with the technical document. As a result, "Metaverse Enhancements" was associated with documents 2, 10, 14, 15, and 16, and "Data Organization and Security" was associated with documents 4, 5, 6, 7, 8, and 11. It was possible to distinguish between documents of different natures without reading the 20 patent documents, making it possible to analyze these documents.

[0179] Furthermore, the last 20 of the 50 patent documents included in the spreadsheet were selected as technical document D to be analyzed, and a request was sent to GPT-4 in the same way. Documents 31, 32, 43, and 50 were associated with "Secure authentication and object verification," and documents 35, 38, 39, 46, and 48 were associated with "Avatar behavior and interaction." Judging from the problem summaries in the spreadsheet, documents 39 and 48 may not necessarily be classified appropriately, but the others are classified as relevant to the problem summaries, making it possible to analyze the documents without reading the 20 patent documents.

[0180] As feedback for the classification generated in this way, we specified classification C={metaverse enhancements, data organization, authentication and verification, avatar behavior, others} and sent a request to GPT-4 to determine the classification representing each of patent documents 11 to 40 included in the spreadsheet above. The results are shown in the table below.

[0181]

[0182] For example, although there is room for understanding that documents 11 and 38 are associated with classifications that are not necessarily appropriate, it is inevitable that a certain degree of noise will occur in analyses such as patent research, and the classifications in the table above are useful analytical results that can be used in practice and allow one to grasp the nature of each document without reading any of the patent documents or using existing patent classifications.

[0183] Example 2 The 50 task summaries generated in Example 1 were classified as "Y" or "N" using the code shown in Figure 5. The results are shown in the table below. The variable "description" was set to "detection of authenticity of metaverse objects."

[0184]

[0185] The reasons for the classification "Y" are listed in the table below.

[0186]

[0187]

[0188] As shown in the table, documents that are directly or indirectly related to the authenticity of objects in the metaverse are given a classification of “Y,” making it possible to quickly grasp the contents of a large number of technical documents.

[0189] Note that in the above description, in one example, q classifications C are generated, and among them, the classification representing each of the p technical documents D is associated with the technical document Di as Ci'. However, it is also possible to generate q classifications C = {C1, C2..., Cq} (1 < q < p) that can classify the p technical documents D or their features F or modified features F', associate them with each technical document, and receive and store the classification associated with the technical document Di as C'i.

[0190] In the above embodiments, it should be noted that in this specification, it is assumed that additional information can be considered if there is no description of "only" such as "based only on", "responding only to", or "only in the case of". Also, as an example, it should be noted that the description "do b in the case of a" does not necessarily mean "always do b in the case of a" or "do b immediately after a" except when explicitly stated. Also, the description "each a constituting A" does not necessarily mean that A is composed of a plurality of components, and includes the case where the component is singular.

[0191] It should also be noted that the above-described embodiments of the present invention are included in the disclosure of this specification insofar as they can be arbitrarily combined within a range that does not conflict with each other.

[0192] Also, just in case, even if there is an aspect of performing an operation different from the operations described in this specification in any method, program, terminal, device, server, or system (hereinafter referred to as "method, etc."), each aspect of the present invention is directed to the same operation as any of the operations described in this specification, and it is added that the existence of an operation different from the operations described in this specification does not exclude the method, etc. from the scope of each aspect of the present invention.

[0193] Furthermore, while the above description refers to multiple AI models, for example, in the second embodiment, a request is made to a third AI model before a request is made to a first AI model. In such a case, for ease of readability, the third AI model may be referred to as "AI model k" and the first AI model as "AI model l." If a request is made to a second AI model after a request is made to the first AI model, the second AI model may be referred to as "AI model m," and other AI models may be referred to as appropriate.

[0194] 100 Device 101 Communication unit 102 Processing unit 103 Storage unit 104 Database 110 User terminal 120 Platform 130 Technical literature database

Claims

1. A method for analyzing patent documents, comprising: obtaining a specification of p (p > 1) patent documents D = {D1, D2, …, Dp} to be analyzed; for i from 1 to p, using at least a part of the patent document Di as input, creating an instruction to generate a feature Fi, which is a feature Fi that is a problem of the invention described in the patent document Di, in a predetermined format; performing a request k including the instruction for an AI model (k), wherein the AI model (k) is a generation AI model; storing the generated feature Fi in association with the patent document Di, wherein the feature Fi includes one or more sentences longer than a single term and not included in the input and including.

2. The method according to claim 1, further comprising: performing a request l to request an AI model (l) to assign any one of q classifications C = {C1, C2, …, Cq} to each of the p features F = {F1, F2, …, Fp} associated with each of the p patent documents D or at least any one of the modified features F' obtained by modifying at least any one of them; for i from 1 to p, receiving and storing the classification assigned to the patent document Di and including.

3. The method according to claim 2, further comprising: obtaining a description representing the technology, wherein the q classifications C represent the relevance of each of the features F or the modified features F' to the description.

4. The method according to claim 2 or 3, wherein the request l includes specifying to describe the reason for assigning the classification for at least any one of the q classifications C.

5. The method according to claim 2, wherein the step of performing the request l includes requesting the AI model (l) to generate q (1 < q < p) classifications C = {C1, C2, …, Cq} capable of classifying the p features F = {F1, F2, …, Fp} associated with each of the p patent documents D or at least any one of the modified features F' obtained by modifying at least any one of them.

6. The method according to claim 2, wherein the step of performing the request l includes The method includes a step of requesting the AI ​​model (l) to generate, in association with each technical document, q classifications C = {C1, C2 ..., Cq} capable of classifying p features F = {F1, F2 ..., Fp} associated with each of the p patent documents D or modified features F' in which at least any of them has been modified.

7. 7. The method according to claim 5 or 6, The requirement l includes a specification of a first condition to be imposed on the first generative AI model in generating the q classifications C.

8. 4. The method according to claim 2 or 3, The method further includes a step of transmitting, to the user terminal, viewing screen display information for displaying a viewing screen for document information identifying one or more patent documents to which at least a portion of the p classifications C has been assigned.

9. 4. A method according to any one of claims 1 to 3, comprising: The predetermined format is a format beginning with "the subject is" or an expression having the same meaning.

10. 4. A method according to any one of claims 1 to 3, comprising: The specification is a search query for a technical literature database.

11. 2. The method of claim 1 , The method further includes making a request l to an AI model (l) to generate one or more new problems to be solved based at least in part on the generated p features F, wherein the AI ​​model (l) is a generative AI model, and each of the one or more new problems includes one or more sentences longer than a single term.

12. 1. A program for causing a computer to execute a method for analyzing patent documents, the method comprising: A step of acquiring designations of p (p>1) patent documents D={D1, D2, ..., Dp} to be analyzed; For i between 1 and p, - generating, using as input at least a part of a patent document Di, instructions for generating in a predetermined format a feature Fi which is the subject of the invention described in said patent document Di; making a request k including the instruction to an AI model (k), the AI ​​model (k) being a generative AI model; storing the generated features F in association with the patent documents D i , the features F including one or more sentences longer than a single term not included in the input; Includes.

13. 13. The program of claim 12, wherein the method further comprises: A step of making a request l to an AI model (l) to assign one of q classifications C = {C1, C2 ..., Cq} to each of p features F = {F1, F2 ..., Fp} associated with each of the p patent documents D or modified features F' in which at least one of them has been modified; receiving and storing classifications assigned to patent documents Di for i between 1 and p; Includes.

14. 1. An apparatus for carrying out a method for analyzing patent documents, comprising: Obtain the specification of p (p>1) patent documents D={D1, D2, ..., Dp} to be analyzed, For i between 1 and p, using at least a part of a patent document Di as an input, create an instruction to generate a feature Fi, which is a problem to be solved by the invention described in the patent document Di, in a predetermined format, and make a request k including the instruction to an AI model (k) that is a generating AI model; For i between 1 and p, the generated features Fi, which include one or more sentences longer than a single term that are not included in the input, are stored in association with patent documents Di.

15. 15. The apparatus of claim 14, A request l is made to an AI model (l) to assign one of q classifications C = {C1, C2 ..., Cq} to each of p features F = {F1, F2 ..., Fp} associated with each of the p patent documents D or modified features F' in which at least one of them has been modified; It is further configured to receive and store, for i between 1 and p, classifications assigned to the patent documents Di.