Data processing device and data processing method

Abstract

To provide a data processing system that is capable of creating data in which a status concept relationship between terms in a set of patent documents are classified appropriately, and providing the data to users.SOLUTION: A data processing device 1: acquires a patent document group containing a predetermined keyword: acquires a term group, which is a set of terms in the patent document group; creates feature value data containing term pairs, etc.; creates learning data by adding labels representing status concept relationships to the feature values data; trains a prediction model using the learning data; creates a status concept data group using an output of the prediction model when the feature data is input to the trained prediction model; and displays the status concept dataset on a display 1a.SELECTED DRAWING: Figure 16

Description

[Technical field]

[0001] The present invention relates to a data processing apparatus and method for processing data such as a collection of patent documents. [Background technology]

[0002] A conventional data processing device that creates a doughnut chart showing the relationships between technical keywords from a collection of patent documents is described in Non-Patent Document 1. In this method, technical keywords are extracted from a collection of patent documents by a text mining method, and a two-layered doughnut chart is created by performing clustering based on the frequency of appearance of the extracted technical keywords.

[0003] Furthermore, a conventional data processing device that creates a tree diagram showing the relationships between technical keywords from a collection of non-patent documents such as academic documents and technical documents is described in Non-Patent Document 2. In this method, technical keywords are searched for from a collection of non-patent documents, and a tree diagram is created by performing clustering based on the search results. [Prior art documents] [Patent documents]

[0004] [Non-Patent Document 1] "Efficient patent research using text mining and machine learning", [online], [Retrieved May 18, 2023], Internet<URL:https: / / www.msi.co.jp / event / file / muc17_501_2.pdf> [Non-Patent Document 2] "JDreamUsers Day 2022", [online], [Retrieved May 18, 2023], Internet<URL:https: / / jdream3.com / seminar / document / jdream_users_day_2022.pdf> Summary of the Invention [Problem to be solved by the invention]

[0005] The data processing device of Non-Patent Document 1 has the following problems. That is, in the case of technical keywords (hereinafter referred to as "terms") in claims in patent documents, when two terms are compared, one may be a term of a higher concept and the other a term of a lower concept. In contrast, according to the technology of Non-Patent Document 1, when creating a doughnut chart from a collection of patent documents, clustering is performed based on the frequency of appearance without distinguishing between the higher concept and the lower concept relationships (hereinafter referred to as "status concept relationships") between terms. As a result, the status concept relationships between terms are not properly classified in the created doughnut chart, and the user cannot properly grasp the status concept relationships between the terms.

[0006] The present invention has been made to solve this problem, and has as its first object to provide a data processing device and the like that can create data in which the status concept relationships between terms in a collection of patent documents are appropriately classified and provide the data to a user.

[0007] Furthermore, according to the data processing device of Non-Patent Document 2, when creating a tree diagram from a collection of non-patent documents, clustering is performed without distinguishing the status concept relationships between technical keywords (hereinafter referred to as "terms"). As a result, the status concept relationships between terms are not properly classified in the created tree diagram, and the user cannot properly grasp the status concept relationships between terms.

[0008] The present invention has been made to solve this problem, and a second object of the present invention is to provide a data processing device or the like that can create data in which the status concept relationships between terms in a collection of non-patent literature are appropriately classified, and provide the data to a user.

[0009] Furthermore, when a specified substance is extracted as a technical keyword from a collection of patent documents using a text mining technique, as in the above-mentioned Non-Patent Document 1, it is desirable to further extract terms that represent any of the raw materials, parts, and materials of this specified substance (hereinafter referred to as "basic terms") and to create data that appropriately classifies the status conceptual relationships between the basic terms.

[0010] The present invention has been made to solve this problem, and a third object of the present invention is to provide a data processing device and the like that can create data in which the status concept relationships of basic terms for a specified substance are appropriately classified, and provide the data to users. [Means for solving the problem]

[0011] In order to achieve the first object, a data processing device according to the first aspect of the present invention comprises: a trained model storage unit that stores a trained model, which is a classification model trained by a predetermined learning algorithm using training data including a set of A-th term pairs, which are pairs of terms included in an A-th patent document group, which is a set of patent documents, and labels that represent status concept relationships including relations between superordinate and subordinate concepts between the A-th term pairs; a status concept data group creation unit that creates a status concept data group that represents the relationship between superordinate and subordinate concepts between the Bth term pairs of the feature data by using an output of a trained model when feature data including a set of Bth term pairs, which are pairs of terms included in the Bth patent document group, which is a set of patent documents, is input to the trained model; and an output unit that executes an output process for outputting the status concept data group to an output device.

[0012] In order to achieve the first object, a data processing method according to the second aspect of the present invention comprises: a learned model storage step of storing a learned model, which is a classification model learned by a predetermined learning algorithm using learning data including a set of A-th term pairs, which are pairs of terms included in an A-th patent document group, which is a set of patent documents, and labels representing status concept relationships including relations of superordinate and subordinate concepts between the A-th term pairs; a status concept data group creation step of creating a status concept data group representing the relationship between superordinate and subordinate concepts between the Bth term pairs of the feature data by using an output of the trained model when feature data including a set of Bth term pairs, which are pairs of terms included in the Bth patent document group, which is a set of patent documents, is input to the trained model; an output step of executing an output process for outputting the status concept data group to an output device; is executed by a data processing device.

[0013] According to the above data processing device or data processing method, a trained model is stored, which is a classification model trained by a predetermined learning algorithm using training data including a set of A-th term pairs, which are pairs of terms included in the A-th patent document group, which is a collection of patent documents, and labels representing status concept relationships including superordinate and subordinate concept relationships between the A-th term pairs. Then, a status concept data group representing the superordinate and subordinate concept relationships between the B-th term pairs of feature data is created using the output of the trained model when feature data including a set of B-th term pairs, which are pairs of terms included in the B-th patent document group, which is a collection of patent documents, is input to the trained model.

[0014] Here, the feature data includes a set of B-term pairs, which are pairs of terms included in the B-th patent document group. When such feature data is input to the trained model, the output of the trained model appropriately predicts the relationship between the superordinate and subordinate concepts between the B-th term pairs of the feature data. Therefore, by creating a status concept data group using the output of such a prediction model, the status concept data group can be created as data in which the relationship between the superordinate and subordinate concepts between the B-th term pairs is appropriately classified. Furthermore, by executing an output process for outputting the status concept data group created as described above to an output device, the status concept data group can be provided to a user via the output device. Note that the "patent document" in this specification corresponds to a document including claims such as a patent publication, a Japanese Patent Application Laid-Open Publication, and a utility model registration publication. In addition, the "classification model in which learning has been performed" in this specification means a classification model in which model parameters have been learned. Furthermore, the A-th patent document group and the B-th patent document group may be a set of the same patent documents, may be a set of patent documents different from each other, or may be a set of patent documents that are partially the same. Furthermore, in this specification, a "pair of terms" is not limited to a pair of terms themselves, but also means a pair of symbols representing terms, a pair of numbers representing terms, and the like.

[0015] In the first aspect of the present invention, a patent document group acquisition unit for acquiring a B-th patent document group including the predetermined keyword by searching a database of patent documents based on the predetermined keyword; a term group acquiring unit that acquires a term group that is a term having a predetermined rank in descending order of frequency of appearance among terms in the Bth patent document group by data mining the Bth patent document group; a feature amount data creation unit that creates feature amount data from a group of terms; It is preferable that the device further comprises:

[0016] In the second aspect of the present invention, a patent document group acquisition step of acquiring a B-th patent document group including the predetermined keyword by searching a database of patent documents based on the predetermined keyword; a term group acquisition step of acquiring a term group, which is terms in the B-th patent document group in descending order of frequency of appearance up to a predetermined rank, by data mining the B-th patent document group; a feature amount data creation step of creating feature amount data from the term group; is preferably further performed by a data processing device.

[0017] According to the above data processing device or data processing method, a database of patent documents is searched based on a predetermined keyword to obtain a group of patent documents B containing the predetermined keyword, and a group of terms, which are terms in the group of patent documents B in order of frequency of appearance up to a predetermined rank, is obtained by data mining the group of patent documents B. Then, a group of status concept data is created using the output of a trained model when feature data created from such a group of terms is input to the trained model. This makes it possible to create the group of status concept data as data that appropriately classifies the relationship between superordinate and subordinate concepts between pairs of B terms in a collection of patent documents.

[0018] In the first aspect of the present invention, the feature amount data creation section preferably creates the feature amount data so as to further include claim data of the Bth patent document in which the Bth term pair is included.

[0019] In the second aspect of the present invention, in the feature amount data creating step, the feature amount data is preferably created so as to further include claim data of the B-patent document in which the B-term pair is included.

[0020] According to the above data processing device or data processing method, the feature data is created so as to further include data of the claims of Patent Document B that includes the term pair B. This makes it possible to link the terms and claims included in the status concept data group when the status concept data group is created using the output of the trained model.

[0021] In the first aspect of the present invention, it is preferable to further include a training data creation unit that creates training data in which labels are added to feature data by searching a status concept database that defines status concept relations based on the Bth term pair of feature data.

[0022] In the second present invention, it is preferable that the data processing device further executes a learning data creation step of creating learning data, which is data in which labels are added to feature data, by searching a status concept database that defines status concept relations based on the Bth term pair of the feature data.

[0023] According to the above data processing device or data processing method, a status concept database that defines status concept relations based on the Bth term pair of feature data is searched to create learning data that is data in which labels are added to feature data. Therefore, by creating learning data by changing the predetermined keywords and repeatedly learning the classification model using such learning data, the learning effect of the classification model can be further improved.

[0024] In the first aspect of the present invention, the training data creation unit preferably creates the training data as data in which, in addition to the labels, the claim data of Patent Document B containing the term is further added to the feature amount data.

[0025] In the second aspect of the present invention, in the training data creation step, the training data is preferably created as data in which, in addition to the labels, data of the claims of Patent Document B containing the terms is further added to the feature data.

[0026] According to the above data processing device or data processing method, the learning data is created as data in which, in addition to the labels, the data of the claims of Patent Document B, which includes terms, is further added to the feature data. This can further improve the learning effect of the prediction model using the learning data.

[0027] In the first aspect of the present invention, it is preferable that the status concept relationship of the Bth term pair included in the status concept data group is added to the status concept database.

[0028] In the second aspect of the present invention, it is preferable that the status concept relationship of the B-th term pair included in the status concept data group is added to the status concept database.

[0029] According to the above data processing device or data processing method, the status concept relationship of the B-th term pair included in the status concept data group is added to the status concept database. As a result, the status concept database with the increased data volume is used to create learning data, thereby further improving the learning effect of the prediction model using the learning data.

[0030] In the first aspect of the present invention, in the learning data, a label defining a superordinate concept is set to a first predetermined value, and a label defining a subordinate concept is set to a second predetermined value, An average value calculation unit is further provided for calculating an average value of a label representing a relationship between a B-th term pair in an output of the trained model when the feature data is input to the trained model, The output unit preferably executes a data output process for outputting data of the B-th term pair, the label of which has an average value within a predetermined range, to an output device.

[0031] In the second aspect of the present invention, in the learning data, a label defining a superordinate concept is set to a first predetermined value, and a label defining a subordinate concept is set to a second predetermined value, The data processing device further executes an average value calculation step of calculating an average value of a label representing a relationship between the B-th term pair in an output of the trained model when the feature data is input to the trained model; In the output step, a data output process is preferably performed to output data of the B-th term pair whose label average value is within a predetermined range to an output device.

[0032] According to the above data processing device or data processing method, the average value of the labels representing the relationship of the Bth term pair in the output of the trained model when the feature data is input to the trained model is calculated, and a data output process is executed to output data of the Bth term pair in which the average value is within a predetermined range to the output device. Therefore, if the predetermined range is set to a range in which the prediction accuracy is expected to be low, when data of the Bth term pair with low prediction accuracy is output from the output device, the user can confirm the data of the Bth term pair with low prediction accuracy. This allows the user to assign / modify the labels of the term pairs with low prediction accuracy in the training data by himself / herself. Furthermore, when the training of the prediction model is re-executed using the training data to which the labels have been assigned / modified in this way, the prediction accuracy of the prediction model can be improved.

[0033] In the first aspect of the present invention, the status concept data group creation section preferably creates the status concept data group in a state in which the patent document number is linked to each term of the Bth term pair.

[0034] In the second aspect of the present invention, in the status concept data group creating step, it is preferable that the status concept data group is created in a state in which the patent document number is linked to each term of the Bth term pair.

[0035] According to the above data processing device or data processing method, a status concept data group is created with a patent document number linked to each term of the B-th term pair in the status concept data group. This allows a user to further grasp the patent document number corresponding to each term of the B-th term pair in addition to the status concept relationship of the B-th term pair when the status concept data group is output from an output device.

[0036] In the first aspect of the present invention, the status concept data group creating section preferably creates the status concept data group as tree-shaped graphic data in which the terms of the B-th term pair are arranged in accordance with a relationship of superordinate and subordinate concepts.

[0037] In the second aspect of the present invention, in the status concept data group creating step, the status concept data group is preferably created as tree-shaped graphic data in which the terms of the B-th term pair are arranged in accordance with a relationship of superordinate and subordinate concepts.

[0038] According to the above data processing device or data processing method, the status concept data group is created as tree-shaped graphic data in which the terms of the B-th term pair are arranged according to the relationship of superordinate and subordinate concepts. When the status concept data group is output to an output device, the user can easily grasp the relationship of superordinate and subordinate concepts between the B-th term pair via the output device.

[0039] In the first aspect of the present invention, the patent document group acquisition unit further acquires, in addition to the Bth patent document group, a second patent document group which is a set of patent documents including the second predetermined keyword by searching a database of patent documents based on a second predetermined keyword which is similar to the predetermined keyword in at least one of concept, property, and classification; the term group acquisition unit performs data mining on the Bth patent document group and the second patent document group to further acquire a second term group, which is terms in the patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance; the feature data creation unit further creates second feature data including a set of second term pairs, which are pairs of terms included in the second term group, the second term pairs being pairs of terms included in the Bth patent document group and the second patent document group; It is preferable that the status concept data group creation unit creates a second status concept data group representing the relationship between superordinate and subordinate concepts between second term pairs in the second feature data, using the output of the prediction model when the second feature data is input to the prediction model.

[0040] In the second aspect of the present invention, in the patent document group acquisition step, in addition to the Bth patent document group, a database of patent documents is searched based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification, to further acquire a second patent document group, which is a set of patent documents including the second predetermined keyword; In the term group acquisition step, a second term group is further acquired, which is terms in the Bth patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance, by performing data mining on the Bth patent document group and the second patent document group; In the feature data creation step, second feature data is further created, the second feature data including a set of second term pairs, which are pairs of terms included in the patent documents and the second patent document group, among the terms included in the second term group; In the status concept data group creation step, it is preferable that a second status concept data group representing the relationship between superordinate and subordinate concepts between term pairs of the second feature data is created using the output of the predictive model when the second feature data is input into the predictive model.

[0041] According to the above data processing device or data processing method, the second patent document group is obtained as a set of patent documents including the second predetermined keyword by searching a database of patent documents based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. Furthermore, by data mining the B patent document group and the second patent document group, a second term group is further obtained, which is terms in the B patent document group and the second patent document group that appear frequently up to a predetermined rank. Then, the second status concept data group is created using the output of a prediction model when the second feature amount data created from such a second term group is input to the prediction model, so that the second status concept data group can be created to include terms similar to the status concept data group in at least one of concept, property, and classification.

[0042] In order to achieve the second object described above, a data processing device according to a third aspect of the present invention comprises: a non-patent document group acquisition unit that searches a database of non-patent documents based on a predetermined keyword to acquire a non-patent document group that is a set of non-patent documents containing the predetermined keyword; a term group acquiring unit that acquires a term group, which is terms in the non-patent literature in descending order of frequency of appearance up to a predetermined rank, by performing data mining on the non-patent literature; a patent document group acquisition unit that searches a database of patent documents based on the term group to acquire a patent document group, which is a set of patent documents including the term group; a feature data creation unit that creates feature data including a set of term pairs that are pairs of terms included in the patent document group, among the terms included in the term group; a status concept data group creation unit that creates a status concept data group that represents the relationship between superordinate and subordinate concepts between term pairs of the feature amount data by using an output of the prediction model, which is a classification model, when the feature amount data is input into the prediction model; an output unit that executes an output process for outputting the status concept data group to an output device; Equipped with The prediction model is characterized in that it is configured to output label values ​​that represent the relationship between superordinate and subordinate concepts between a pair of terms when feature amount data is input.

[0043] In order to achieve the second object described above, a data processing method according to a fourth aspect of the present invention includes: a non-patent document group acquisition step of acquiring a non-patent document group, which is a set of non-patent documents including the predetermined keyword, by searching a database of non-patent documents based on the predetermined keyword; a term group acquisition step of acquiring a term group, which is terms in the non-patent literature group, in order of frequency of appearance up to a predetermined rank from the terms in the non-patent literature group by data mining the non-patent literature group; a patent document group acquisition step of acquiring a patent document group, which is a set of patent documents including the term group, by searching a database of patent documents based on the term group; a feature data creation step of creating feature data including a set of term pairs, which are pairs of terms included in the patent documents, among the terms included in the term group; a status concept data group creation step of creating a status concept data group representing the relationship between superordinate and subordinate concepts between term pairs of the feature data by using an output of the prediction model, which is a classification model, when the feature data is input into the prediction model; an output step of executing an output process for outputting the status concept data group to an output device; is executed by a data processing device, The prediction model is characterized in that it is configured to output label values ​​that represent the relationship between superordinate and subordinate concepts between a pair of terms when feature amount data is input.

[0044] According to the above data processing device or data processing method, feature data including a set of term pairs, which are pairs of terms included in a group of terms and which are included in a group of patent documents, is created, and a status concept data group that defines the status concept relationship of the term pairs of the feature data is created using the output of the prediction model when the feature data is input to the prediction model. Here, the prediction model is configured to output label values ​​that represent the relationship of superordinate and subordinate concepts between the term pairs when the feature data is input, so that the output of the prediction model appropriately predicts the relationship of superordinate and subordinate concepts between the term pairs of the feature data. Therefore, by creating a status concept data group using the output of such a prediction model, the status concept data group can be created as data in which the relationship of superordinate and subordinate concepts between the term pairs is appropriately classified. Furthermore, an output process for outputting the status concept data group created as described above to an output device is executed, so that the status concept data group can be provided to a user via the output device.

[0045] In addition, by searching a database of non-patent documents, a group of non-patent documents containing a predetermined keyword is obtained, and by data mining the group of non-patent documents, a group of terms, which are terms in the group of non-patent documents that appear in a predetermined order from the most frequently occurring terms, is obtained, and by searching a database of patent documents based on the group of terms, a group of patent documents, which is a set of patent documents containing the group of terms, is obtained. As a result, the status concept data group can be created as data in which the relationship between the superordinate and subordinate concepts between terms in the set of patent documents and the relationship between the superordinate and subordinate concepts between terms in non-patent documents are appropriately classified. As a result, when the status concept data group created in this way is output to an output device, the user can check, via the output device, the status concept data group in which the relationship between the superordinate and subordinate concepts between terms in the set of patent documents and the relationship between the superordinate and subordinate concepts between terms in non-patent documents are appropriately classified.

[0046] In the third aspect of the present invention, a learning data creation step of creating learning data in which a label representing a status concept relationship is added to the feature data by searching a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between the term pairs based on the term pairs of the feature data; a model learning step of learning a predictive model by a predetermined learning algorithm using the learning data; is preferably further performed by a data processing device.

[0047] In the fourth aspect of the present invention, a learning data creation step of creating learning data in which a label representing a status concept relationship is added to the feature data by searching a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between the term pairs based on the term pairs of the feature data; a model learning step of learning a predictive model by a predetermined learning algorithm using the learning data; is preferably further performed by a data processing device.

[0048] According to the above data processing system or data processing method, a prediction model is trained by a predetermined learning algorithm using training data in which labels representing status concept relationships, including relationships between superordinate and subordinate concepts, between term pairs of feature data are added to the feature data. As a result, when the feature data is input to the prediction model, a prediction result that appropriately predicts the superordinate and subordinate concept relationships between the term pairs of the feature data is output from the prediction model. Therefore, by using the output of such a prediction model, a status concept data group that defines the status concept relationships of the term pairs of the feature data can be appropriately created.

[0049] In the third aspect of the present invention, the status concept data group creating section preferably creates the status concept data group as tree-shaped graphic data in which the terms of the term pairs are arranged in accordance with a relationship of superordinate and subordinate concepts.

[0050] In the fourth aspect of the present invention, in the status concept data group creating step, the status concept data group is preferably created as tree-shaped graphic data in which the terms of the term pairs are arranged in accordance with a relationship of superordinate and subordinate concepts.

[0051] According to the above data processing device or data processing method, a status concept data group is created as tree-shaped graphic data in which the terms of a term pair are arranged according to the relationship between superordinate and subordinate concepts. When this status concept data group is output to an output device, a user can easily grasp the relationship between superordinate and subordinate concepts between the term pairs via the output device.

[0052] In the third aspect of the present invention, it is preferable that the status concept data group creation unit creates the status concept data group in a state in which at least one of the data representing the patent document number and the non-patent document is linked to each term of the term pair.

[0053] In the fourth aspect of the present invention, in the status concept data group creation step, it is preferable that the status concept data group is created in a state in which at least one of the patent document number and the data representing the non-patent document is linked to each term of the term pair.

[0054] According to the above data processing device or data processing method, a status concept data group is created in a state where at least one of the data representing the patent document number and the non-patent document is linked to each term of the term pair in the status concept data group. A user can further grasp at least one of the data representing the patent document number and the non-patent document corresponding to each term of the term pair, in addition to the status concept relationship of the term pair, via the output device.

[0055] In the third aspect of the present invention, the non-patent document group acquisition unit further acquires a second non-patent document group, which is a set of non-patent documents including the second predetermined keyword, by searching a database of non-patent documents based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification, in addition to the non-patent document group; the term group acquisition unit performs data mining on the non-patent document group and the second non-patent document group to further acquire a second term group, which is terms in the non-patent document group and the second non-patent document group that appear in a predetermined order in descending order of frequency of appearance; the patent document group acquisition unit searches a database of patent documents based on the second term group to acquire a second patent document group, which is a set of patent documents including the second term group; the feature data creation unit further creates second feature data including a set of second term pairs, which are pairs of terms included in the second term group and are included in the second patent documents; It is preferable that the status concept data group creation unit creates a second status concept data group representing the relationship between superordinate and subordinate concepts between second term pairs in the second feature data, using the output of the prediction model when the second feature data is input to the prediction model.

[0056] In the fourth aspect of the present invention, in the non-patent document group acquisition step, in addition to the non-patent document group, a database of non-patent documents is searched based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification, to further acquire a second non-patent document group, which is a set of non-patent documents including the second predetermined keyword; In the term group acquisition step, a second term group is further acquired, which is terms in the non-patent document group and the second non-patent document group that appear in a predetermined order in descending order of frequency of appearance from the non-patent document group and the second non-patent document group, by performing data mining on the non-patent document group and the second non-patent document group; In the patent document group acquisition step, a database of patent documents is searched based on the second term group to acquire a second patent document group, which is a set of patent documents including the second term group; In the feature data creation step, second feature data including a set of second term pairs, which are pairs of terms included in the second term group and are included in the second patent document group, are further created; In the status concept data group creation step, it is preferable that a second status concept data group representing the relationship between superordinate and subordinate concepts between term pairs of the second feature data is created using the output of the predictive model when the second feature data is input into the predictive model.

[0057] According to the above data processing device or data processing method, the second non-patent document group is obtained as a set of non-patent documents including the second predetermined keyword by searching a database of non-patent documents based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. In addition, by performing data mining on the non-patent document group and the second non-patent document group, a second term group, which is terms in the non-patent document group and the second non-patent document group that appear frequently up to a predetermined rank, is further obtained. Furthermore, the second status concept data group is created using the output of a prediction model when feature data created from such a second term group is input to the prediction model, so that the second status concept data group can be created to include terms similar to the status concept data group in at least one of concept, property, and classification.

[0058] In order to achieve the third object described above, a data processing device according to a fifth aspect of the present invention comprises: a patent document group acquisition unit that searches a database of patent documents based on a predetermined keyword that represents a predetermined substance to acquire a patent document group that is a collection of patent documents that include the predetermined keyword; a term group acquiring unit that acquires a term group, which is terms in the patent documents in descending order of frequency of appearance up to a predetermined rank, by performing data mining on the patent documents; a basic term group acquiring unit that acquires a basic term group, which is a set of basic terms, from a term group by searching a basic term database that defines basic terms that represent any of the raw materials, parts, and materials of a predetermined substance based on the predetermined substance; a feature data creation unit that creates feature data including a set of selected term pairs, each of which is a pair of a basic term included in a basic term group and one of a term included in a term group and a basic term included in the basic term group; a status concept data group creation unit that creates a status concept data group that represents the relationship between superordinate and subordinate concepts between selected term pairs of the feature data by using an output of the prediction model, which is a classification model, when the feature data is input into the prediction model; an output unit that executes an output process for outputting the status concept data group to an output device; Equipped with The prediction model is characterized in that it is configured to output label values ​​that represent the relationship between superordinate and subordinate concepts between selected term pairs when feature data is input.

[0059] In order to achieve the third object described above, a data processing method according to the sixth aspect of the present invention comprises: a patent document group acquisition step of acquiring a patent document group, which is a collection of patent documents including a predetermined keyword by searching a database of patent documents based on a predetermined keyword representing a predetermined substance; a term group acquisition step of acquiring a term group, which is terms in the patent documents in descending order of frequency of appearance up to a predetermined rank, by performing data mining on the patent documents; a basic term group acquisition step of acquiring a basic term group, which is a set of basic terms, from a term group by searching a basic term database that defines basic terms that represent any of the raw materials, parts, and materials of the predetermined substance based on the predetermined substance; a feature data creation step of creating feature data including a set of selected term pairs each of which is a pair of a basic term included in a basic term group and either a term included in a term group or a basic term included in the basic term group; a status concept data group creation step of creating a status concept data group representing the relationship between superordinate and subordinate concepts between selected term pairs of the feature data by using an output of the prediction model, which is a classification model, when the feature data is input into the prediction model; an output step of executing an output process for outputting the status concept data group to an output device; is executed by a data processing device, The prediction model is characterized in that it is configured to output label values ​​that represent the relationship between superordinate and subordinate concepts between selected term pairs when feature data is input.

[0060] According to the above data processing device or data processing method, feature data including a set of selected term pairs, which are pairs of a basic term included in a basic term group and a term included in a term group and one of the basic terms included in the basic term group, is created, and a status concept data group representing the relationship of superordinate and subordinate concepts between the selected term pairs of the feature data is created using the output of the prediction model, which is a classification model, when the feature data is inputted, and an output process is executed to output the status concept data group to an output device. Here, the prediction model is configured to output the value of a label representing the relationship of superordinate and subordinate concepts between the selected term pairs when the feature data is inputted, so that the output of the prediction model appropriately predicts the relationship of superordinate and subordinate concepts between the selected term pairs of the feature data. Therefore, by creating a status concept data group using the output of such a prediction model, the status concept data group can be created as data in which the relationship of superordinate and subordinate concepts between the selected term pairs is appropriately classified. Furthermore, an output process is executed to output the status concept data group created as above to the output device, so that the status concept data group can be provided to a user via the output device.

[0061] In addition, a patent document database is searched based on a predetermined keyword representing a predetermined substance to obtain a patent document group that is a collection of patent documents containing the predetermined keyword, and the patent document group is data mined to obtain a term group that is terms in the patent document group that appear in a predetermined order from the most frequently occurring terms. Furthermore, a basic term database that defines basic terms that represent any of the raw materials, parts, and materials of the predetermined substance is searched based on the predetermined substance to obtain a basic term group that is a collection of basic terms from the term group. This makes it possible to create a status concept data group as data that appropriately classifies the relationships between superordinate and subordinate concepts between basic terms or between basic terms and terms in a collection of patent documents.

[0062] In the fifth aspect of the present invention, a learning data creation unit creates learning data in which a label representing a status concept relationship is added to feature data by searching a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between selected term pairs based on the selected term pairs of feature data; a model learning unit that uses the learning data to learn a predictive model according to a predetermined learning algorithm; It is preferable that the device further comprises:

[0063] In the sixth aspect of the present invention, a learning data creation step of creating learning data in which a label representing a status concept relationship is added to the feature data by searching a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between the selected term pairs based on the selected term pairs of the feature data; a model learning step of learning a predictive model by a predetermined learning algorithm using the learning data; is preferably further performed by a data processing device.

[0064] According to the above data processing system or data processing method, a prediction model is trained by a predetermined learning algorithm using training data in which labels representing status concept relationships, including relationships between superordinate and subordinate concepts, between selected term pairs of feature data are added to the feature data. As a result, when the feature data is input to the prediction model, a prediction result that appropriately predicts the relationship between the superordinate and subordinate concepts between the selected term pairs of the feature data is output from the prediction model. Therefore, by using the output of such a prediction model, a status concept data group that defines the status concept relationships between selected term pairs of the feature data can be appropriately created.

[0065] In the fifth aspect of the present invention, the status concept data group creating section preferably creates the status concept data group as tree-shaped graphic data in which the selected terms of the selected term pairs are arranged in accordance with a relationship of superordinate and subordinate concepts.

[0066] In the sixth aspect of the present invention, in the status concept data group creating step, the status concept data group is preferably created as tree-shaped graphic data in which the selected terms of the selected term pairs are arranged in accordance with the relationship of superordinate and subordinate concepts.

[0067] According to the above data processing device or data processing method, a status concept data group is created as tree-shaped graphic data in which the selected terms of the selected term pairs are arranged in accordance with the relationship between superordinate and subordinate concepts. When the status concept data group is output to an output device, a user can easily grasp the relationship between superordinate and subordinate concepts of the selected terms of the selected term pairs via the output device.

[0068] In the fifth aspect of the present invention, it is preferable that the status concept data group creation unit further uses a product database that defines the relationship between the basic terms of the selected term pairs in the status concept data group and the product information of the basic terms, to create the status concept data group in a state in which the product information is linked to the basic terms.

[0069] In the sixth aspect of the present invention, in the status concept data group creation step, it is preferable that the status concept data group is created with product information linked to the basic terms, further using a product database that defines the relationship between the basic terms of the selected term pairs in the status concept data group and the product information of the basic terms.

[0070] According to the above data processing device or data processing method, a status concept data group is created in a state where the basic term of the selected term pair in the status concept data group is linked with the product information of the basic term. Thus, when the status concept data group is output to an output device, the user can grasp the product information of the basic term via the output device.

[0071] In the fifth aspect of the present invention, The patent document group acquisition unit further acquires a second patent document group, which is a set of patent documents including a second predetermined keyword, by searching a database of patent documents based on a second predetermined keyword that represents a second substance having at least one of properties and classification similar to the predetermined substance, in addition to the patent document group; the term group acquisition unit performs data mining on the patent group and the second patent group to further acquire a second term group, which is terms in the patent group and the second patent group that appear in a predetermined order in descending order of frequency of appearance; the basic term group acquisition unit searches the basic term database based on the second substance to further acquire a second basic term group, which is a set of basic terms for the second substance, from the second term group; the feature data creation unit further creates second feature data including a set of second selected term pairs each of which is a pair of a basic term included in the second basic term group and one of a second term included in the second term group and a basic term included in the second basic term group; It is preferable that the status concept data group creation unit further creates a status concept data group representing the relationship between superordinate and subordinate concepts between the second selected term pairs of the second feature data, using the output of the prediction model when the second feature data is input to the prediction model.

[0072] In the sixth aspect of the present invention, In the patent document group acquisition step, in addition to the patent document group, a database of patent documents is searched based on a second predetermined keyword that represents a second substance similar in at least one of properties and classification to the predetermined substance, thereby further acquiring a second patent document group, which is a set of patent documents including the second predetermined keyword; In the term group acquisition step, a second term group is further acquired, which is terms in the patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance, by performing data mining on the patent document group and the second patent document group; In the basic term group acquisition step, a second basic term group, which is a set of basic terms for the second substance, is further acquired from the second term group by searching a basic term database based on the second substance; In the feature data creation step, second feature data is further created, the second feature data including a set of second selected term pairs each being a pair of a basic term included in the second basic term group and one of a second term included in the second term group and a basic term included in the second basic term group; In the status concept data group creation step, it is preferable that a status concept data group representing the relationship between superordinate and subordinate concepts between the second selected term pairs of the second feature data is further created using the output of the predictive model when the second feature data is input into the predictive model.

[0073] According to the above data processing device or data processing method, the second patent document group is obtained as a set of patent documents including the second predetermined keyword by searching a database of patent documents based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. In addition, a second basic term group, which is basic terms in the non-patent document group and the second non-patent document group in order of appearance frequency up to a predetermined rank, is obtained by data mining the patent document group and the second patent document group. Furthermore, a second status concept data group is created using the output of a prediction model when feature data created from such a second basic term group is input to the prediction model, so that the second status concept data group can be created to include basic terms similar to the status concept data group in at least one of concept, property, and classification. [Brief description of the drawings]

[0074] [Figure 1] 1 is a diagram showing a data processing device according to a first embodiment of the present invention. [Diagram 2] 1 is a block diagram showing a functional configuration of a data processing device according to a first embodiment; [Diagram 3] FIG. 13 is a diagram showing an example of a term group. [Figure 4] FIG. 4 is a diagram illustrating an example of feature amount data. [Diagram 5] FIG. 4 is an explanatory diagram of a procedure for creating feature data. [Figure 6]FIG. 4 is an explanatory diagram of a procedure for creating feature data. [Figure 7] FIG. 4 is an explanatory diagram of a procedure for creating feature data. [Figure 8] FIG. 4 is an explanatory diagram of a procedure for creating feature data. [Figure 9] FIG. 13 is a diagram illustrating an example of learning data. [Figure 10] FIG. 13 is a diagram showing an example of a status concept database. [Figure 11] FIG. 13 is an explanatory diagram of a procedure for creating a status concept data group. [Figure 12] FIG. 13 is an explanatory diagram of a procedure for creating a status concept data group. [Figure 13] FIG. 13 is an explanatory diagram of a procedure for creating a status concept data group. [Figure 14] FIG. 13 is an explanatory diagram of a procedure for creating a status concept data group. [Figure 15A] FIG. 13 is a diagram showing an example of a status concept data group. [Figure 15B] FIG. 13 is a diagram showing a state in which a term in a status concept data group is selected. [Figure 16] FIG. 11 is a diagram showing another example of a status concept data group. [Figure 17] 13 is a flowchart showing a trained model creation process. [Figure 18] 13 is a flowchart showing a status concept data group process. [Figure 19] 13 is a flowchart showing a resetting data process. [Figure 20] 13 is a flowchart showing a second trained model creation process. [Figure 21] 13 is a flowchart showing a second status concept data group process. [Figure 22] FIG. 13 is a diagram showing a modified example of a status concept data group. [Diagram 23] FIG. 11 is a block diagram showing a functional configuration of a data processing device according to a second embodiment. [Figure 24] 13 is a flowchart showing a trained model creation process. [Diagram 25] 13 is a flowchart showing a status concept data group process. [Figure 26] 13 is a flowchart showing a second trained model creation process. [Figure 27] 13 is a flowchart showing a second status concept data group process. [Figure 28] FIG. 13 is a block diagram showing a functional configuration of a data processing device according to a third embodiment. [Figure 29] FIG. 13 is a diagram showing an example of a term group. [Diagram 30] FIG. 13 is a diagram showing an example of a basic term group. [Diagram 31] FIG. 4 is a diagram illustrating an example of feature amount data. [Diagram 32] FIG. 13 is a diagram illustrating an example of learning data. [Diagram 33] FIG. 13 is a diagram showing an example of a status concept database. [Diagram 34] FIG. 13 is a diagram showing an example of a status concept data group. [Diagram 35] FIG. 13 is a diagram showing a display example of a status concept data group. [Diagram 36] FIG. 13 is a diagram showing another display example of a status concept data group. [Figure 37] FIG. 13 is a diagram showing an example of a database of basic term numbers and product names of basic terms. [Figure 38] 13 is a flowchart showing a trained model creation process. [Figure 39] 13 is a flowchart showing a status concept data group process. [Diagram 40] 13 is a flowchart showing a second trained model creation process. [Diagram 41] 13 is a flowchart showing a second status concept data group process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0075] A data processing device according to a first embodiment of the present invention will be described below with reference to the drawings. As shown in Fig. 1, the data processing device 1 according to this embodiment is of a personal computer type, and includes a display 1a, a device body 1b, and an input interface 1c. The display 1a as an output device is of a liquid crystal display type, and displays various data.

[0076] The device main body 1b includes a storage such as an HDD, a processor, and memory (RAM, E2PROM, ROM, etc.) (none of which are shown in the figure). Application software for executing the learning process described below is installed in the storage of this device main body 1b, and a patent document database 20 and a status concept database 21 (see FIG. 2) are also stored therein.

[0077] The databases 20 and 21 will be described in detail later. The input interface 1c is composed of a keyboard, a mouse, and the like for operating the data processing device 1.

[0078] As shown in Figure 2, the data processing device 1 has the functions of a patent document group acquisition unit 11, a term group acquisition unit 12, a feature data creation unit 13, a learning data creation unit 14, a model learning unit 15, a status concept data group creation unit 16, an output unit 17, and an average value calculation unit 18.

[0079] In the patent document group acquisition unit 11, when a specified keyword is input by a user (not shown) operating the input interface 1c, a patent document group is acquired by searching the patent document database 20 in the data processing device 1 based on the specified keyword.

[0080] Here, the patent document database 20 includes many documents including claims, such as patent publications, JP Patent Publications, and utility model registration publications, and the patent document group is acquired as a collection of patent documents in which a predetermined keyword is included in a claim. In the following explanation, an example will be explained in which the predetermined keyword is "polyoxyethylene diglycerin saturated fatty acid ester."

[0081] Next, the term group acquiring unit 12 acquires a term group by data mining the patent documents using morphological analysis. This term group is acquired as a set of terms in the claims of the patent documents in order of frequency of appearance up to a predetermined rank. In the case of this embodiment, as shown in Fig. 3, the terms with the highest frequency of appearance from 1st to 5345th are acquired, and KEYW0 to KEYW5344 are added as the term numbers, respectively.

[0082] Next, the feature data creation unit 13 creates feature data based on the patent document group and the term group. As shown in FIG. 4, this feature data is created so as to include the application number of the patent document, a term pair consisting of term 1 and term 2, and a claim number and a claim sentence including these term pairs. FIG. 4 shows an example in which the feature data includes the term numbers as term 1 and term 2, but the feature data may be configured so that the terms themselves are included as term 1 and term 2 instead of the term numbers. This feature data is created by a program created by the applicant in the following manner. In this embodiment, the term pairs correspond to the A-th term pair and the B-th term pair.

[0083] First, in the data of the patent documents shown in FIG. 5 (only a part of which is shown), terms (words shown in bold) in the term group included in the claims are picked up. In FIG. 5, for ease of understanding, the picked up terms are represented by A to H, resulting in the state shown in FIG. 6. For example, the terms in FIG. 6 corresponds to the term "polyoxyethylene diglycerin saturated fatty acid ester" in Figure 5. corresponds to "polyoxyethylene diglycerin linear saturated fatty acid ester" in Figure 5. In Figures 5 and 6, the claim numbers are shown separately from the claim text for ease of understanding.

[0084] Next, by deleting claims of the patent document that contain only one term from the data set shown in Figure 6, a data set in which two or more terms are included in the claims of the patent document is created as shown in Figure 7.

[0085] Furthermore, as shown in Fig. 8, by picking up term pairs, which are combinations of two or more terms included in a claim, as "Term 1, Term 2," the feature data in Fig. 4 described above is created. For example, in the case of the publication with application number 2000-321321 in Fig. 8, since claim 1 includes three terms A to C, there are six term pairs: "A,B," "B,A," "B,C," "C,B," "A,C," and "C,A."

[0086] The learning data creation unit 14 creates the learning data shown in Fig. 9 from the feature amount data of Fig. 4 described above by referring to the status concept database 21 in the data processing device 1. Here, the status concept database 21 defines status concept relationships that represent the relationship between higher and lower concepts between term 1 and term 2, as shown in Fig. 10.

[0087] In Fig. 10, the label values ​​represent the status concept relationship between the term pairs, and specifically, a label value of "1" represents that term 1 is a superordinate concept of term 2. A label value of "0" represents that term 1 is a subordinate concept of term 2, and a label value of "2" represents that term 1 and term 2 are not in a superordinate or subordinate concept relationship.

[0088] In this embodiment, the value 1 corresponds to the first predetermined value, and the value 0 corresponds to the second predetermined value. The first and second predetermined values ​​are not limited to the above values, and can be set to various values. For example, the first and second predetermined values ​​can be set to values ​​10 and 0, respectively, or the first and second predetermined values ​​can be set to values ​​100 and 0, respectively.

[0089] The learning data creation unit 14 refers to the status concept database 21 and adds a label representing the status concept relationship between the term pairs to the feature amount data, thereby creating the learning data shown in FIG.

[0090] Next, in the model learning unit 15 (model storage unit), learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm (for example, a natural language model of Transformer) using the learning data created as described above. In the following description, a prediction model that has already been learned is referred to as a "trained model". This trained model is stored in the model learning unit 15.

[0091] Next, the status concept data group creation unit 16 creates a status concept data group using the trained model and feature amount data. This status concept data group is tree-shaped image data (see FIG. 15) in which the terms of the above-mentioned term pairs are arranged according to the relationship between superordinate and subordinate concepts, and is created by the method described below.

[0092] That is, in the status concept data group creation unit 16, the status concept relationship between the term pairs is determined based on the label value of the output of the trained model when feature data is input to the trained model, and a status concept data group is created by arranging the terms of the term pairs according to their superordinate or subordinate concept relationships based on the determination result.

[0093] More specifically, when the label value of the trained model output is "1" or "0," the two terms of the term pair are arranged according to the status concept relationship. On the other hand, when the label value of the trained model output is "2," that is, when the two terms of the term pair are not in a hierarchical or subordinate concept relationship, the two terms are not arranged in a hierarchical or subordinate concept relationship.

[0094] For example, in the feature data of Fig. 8 described above, when the data of claim 2 of application number 2000-321321 is input to the trained model, the label of the term pair "A, B" in the output of the trained model will be a value of 1 because term A is the higher concept, and the label of the term pair "B, C" will be a value of 2 because terms B and C are not in a relationship of higher or lower concepts. Furthermore, the label of the term pair "C, A" will be a value of 0 because term A is the higher concept.

[0095] As a result, a status concept data group of the three terms A to C is created in a state in which the subordinate concepts of terms B and C are subordinate to the superior concept of term A, as shown in Fig. 11. As is clear from referring to Figs. 5 and 8 described above, term A in Fig. 11 is "polyoxyethylene diglycerin saturated fatty acid ester (represented as "saturated fatty acid" in Fig. 11)" and term B is "polyoxyethylene diglycerin linear saturated fatty acid ester (represented as "linear" in Fig. 11)". In addition, term C is "polyoxyethylene diglycerin branched saturated fatty acid ester (represented as "branched" in Fig. 11)".

[0096] Furthermore, in the feature data of FIG. 8 described above, when the data of claim 3 of application number 2000-321321 is input to the trained model, the label of the term pair "B, D" in the output of the trained model will be value 1 because term B is the higher concept. As a result, a status concept data group including two terms B and D is created in a state in which term D, which is a lower concept, is subordinate to term B, which is a higher concept, as shown in FIG. 12. As is clear from reference to FIG. 5 and FIG. 8 described above, term D in FIG. 12 is "polyoxyethylene diglycerin myristate ester (written as "myristic acid" in FIG. 12)."

[0097] Furthermore, in the feature quantity data of FIG. 8 described above, when the data of claims 2 and 3 of application number 2017-333333 are input to the trained model, a status concept data group of four terms A, C, E, and F is created by the same principle as above, as shown in FIG. 13. As is clear from reference to FIG. 5 and FIG. 8 described above, term E in FIG. 13 is "polyoxyethylene diglycerin isopalmitic acid ester (written as "isopalmitic acid" in FIG. 13)" and term D is "polyoxyethylene diglycerin isostearic acid ester (written as "isostearic acid" in FIG. 13)".

[0098] Furthermore, in the feature quantity data of FIG. 8 described above, when the data of claims 1 and 2 of application number 2011-123456 are input to the trained model, a status concept data group of four terms B, D, G, and H is created by the same principle as above, as shown in FIG. 14. As is clear from reference to FIG. 5 and FIG. 8 described above, the term G in FIG. 14 is "polyoxyethylene diglycerin monomyristate ester (written as "monomyristic acid" in FIG. 14)" and the term H is "polyoxyethylene diglycerin dimyristate ester (written as "dimyristic acid" in FIG. 14)."

[0099] By continuing to execute the above process, the status concept data group shown in Fig. 15 is finally created. This status concept data group is created with the patent document number linked to each term. In this case, for example, the patent document numbers may be configured so that those for which patent registration gazettes have been issued and the registration period remains, and those for which patent publication gazettes have been issued and there is a possibility of registration, are linked to each term.

[0100] Then, the data of the status concept data group created as described above is output from the output unit 17 to the display 1a (output process). As a result, the status concept data group shown in Fig. 15 is displayed on the display 1a. That is, the term "polyoxyethylene diglycerin saturated fatty acid ester" is set as the highest concept, and the subordinate concepts "polyoxyethylene diglycerin linear saturated fatty acid ester" and "polyoxyethylene diglycerin branched saturated fatty acid ester" are displayed as subordinate concepts.

[0101] Furthermore, the term "polyoxyethylene diglycerol linear saturated fatty acid ester" has subordinate concepts such as "polyoxyethylene diglycerol myristic acid ester" hanging from it, and the term "polyoxyethylene diglycerol branched saturated fatty acid ester" has subordinate concepts such as "polyoxyethylene diglycerol isostearate ester" hanging from it.

[0102] In this state where the status concept data group is displayed on the display 1a, when the user selects any term from the status concept data group via the input interface 1c, the number of the patent document including the term in the claim is displayed. For example, as shown in Fig. 16, when the term "isostearic acid" is selected, the number of the patent document including the term "polyoxyethylene diglycerol isostearate" in the claim is displayed (Patent Application No. 2011-111111, Patent Application No. 2012-222222, ...).

[0103] In addition, the average value calculation unit 18 calculates the average value of the labels of each term pair based on the output of the trained model when the feature data is input to the trained model. In this case, the average value of the label of each term pair is calculated when the label is "0" or "1".

[0104] Then, the output unit 17 outputs data of term pairs whose average label value is within a predetermined range among the average label values ​​of each term pair calculated as above to the display 1a (data output process). This allows the user to grasp the labels of term pairs whose average label value is within the predetermined range in the learning data. This predetermined range is set by the user to a value (e.g., 0.3 to 0.7) that allows the user to determine that the prediction accuracy of the prediction model is low and that the learning data needs to be corrected.

[0105] Therefore, the user can create corrected learning data by operating the input interface 1c to assign "0", "1" or "2" to the labels of term pairs in the learning data whose average label value is within a predetermined range at his / her own discretion. As a result, when the learning of the prediction model is performed again using such corrected learning data, the prediction accuracy of the prediction model can be improved. In this case, the user may add data or correct the labels of the data in the status concept database 21 in addition to or instead of the learning data.

[0106] In the data processing device 1, the above process is repeatedly executed by changing the predetermined keyword, whereby, for example, a status concept data group as shown in FIG. 16 is created.

[0107] The above process may be repeatedly performed by sequentially setting the 1st to Nth (N is an integer) terms with the highest frequency of appearance as the predetermined keywords from among the term group acquired by the term group acquiring unit 12. In this case, the predetermined keywords may be automatically set.

[0108] In addition, in the elements 11 to 17 of the data processing device 1 described above, when a second predetermined keyword (e.g., polyoxyethylene diglycerin stearate ester) that is similar to the predetermined keyword in at least one of concept, property, and classification is input, various processes are executed as described below.

[0109] That is, in the patent document group acquisition unit 11, when a second specified keyword is input by the user operating the input interface 1c, the second patent document group is acquired by searching the patent document database 20 in the data processing device 1 based on the second specified keyword.

[0110] Next, the term group acquiring unit 12 performs data mining on the group of patent documents and the second patent document to acquire a second term group (not shown) similar to the term group in Fig. 3. This second term group is acquired as a set of terms in the claims of the group of patent documents and the second patent document group, in descending order of appearance frequency up to a predetermined rank.

[0111] Furthermore, in the feature amount data creation unit 13, second feature amount data is created (not shown) based on the group of patent documents, the second group of patent documents, and the group of terms, similarly to the feature amount data in FIG. 4 described above.

[0112] In addition, in the learning data creation unit 14, by referring to the status concept database 21 in the data processing device 1, second learning data is created from the above-mentioned second feature data, similar to the learning data of Figure 9 described above (not shown).

[0113] Next, in the model learning unit 15, additional learning of the trained model is performed by the above-mentioned predetermined learning algorithm using the second training data created as described above. Hereinafter, the prediction model for which additional learning has been performed using the second training data is referred to as the "second trained model."

[0114] Furthermore, in the status concept data group creation unit 16, a second status concept data group is created (not shown) using the second learned model and the second feature data, similar to the status concept data group of Figure 15 described above.

[0115] Then, the data of the second status concept data group created as described above is output from the output unit 17 to the display 1a. As a result, the second status concept data group is displayed on the display 1a (not shown). In this case, in addition to the second status concept data group, the status concept data group may be configured to be displayed on the display 1a at the same time.

[0116] Next, various arithmetic processes executed by the data processing device 1 of this embodiment will be described. First, the trained model creation process will be described with reference to Fig. 17. This trained model creation process is for creating a trained model as described below, and is executed when the above-mentioned predetermined keyword is input by the user's operation of the input interface 1c.

[0117] It is assumed that various values ​​calculated in the following description are stored in the storage or memory of the data processing device 1.

[0118] In this trained model creation process, first, a patent document group acquisition process is executed (FIG. 17 / STEP 1). In this acquisition process, as described above, the patent document group is acquired by searching the patent document database 20 based on a predetermined keyword (patent document group acquisition step).

[0119] Next, a term group acquisition process is executed (FIG. 17 / STEP 2). In this acquisition process, as described above, a term group is acquired by data mining a group of patent documents (term group acquisition step).

[0120] Next, a feature data creation process is executed (FIG. 17 / STEP 3). In this creation process, as described above, the feature data shown in FIG. 4 is created based on the patent documents and the terminology group (feature data creation step).

[0121] Following this feature data creation process, a learning data creation process is executed (FIG. 17 / STEP 4). In this creation process, as described above, the learning data shown in FIG. 9 is created from the feature data in FIG. 4 using data present in the status concept database 21 (learning data creation step).

[0122] Next, a model learning process is executed (FIG. 17 / STEP 5). In this model learning process, as described above, learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm using the learning data (model learning step). By executing the above processes, a trained model is created.

[0123] Next, the status concept data group process will be described with reference to Fig. 18. This status concept data group process is for creating a status concept data group and displaying it on the display 1a, as described below, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0124] In this status concept data group process, first, a status concept data group creation process is executed (FIG. 18 / STEP 10). In this creation process, as described above, the status concept data group shown in FIG. 15 is created using the trained model and feature amount data (status concept data group creation step).

[0125] Next, a display process of the status concept data group is executed (FIG. 18 / STEP 11). In this display process, as described above, the status concept data group shown in FIG. 15 is displayed on the display 1a (output step).

[0126] Next, the reset data process will be described with reference to Fig. 19. As described below, this reset data process is for displaying data that requires label resetting on the display 1a, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0127] In this reset data processing, first, an average value calculation process is executed (FIG. 19 / STEP 20). In this average value calculation process, an average value of the labels of the term pairs in the output of the trained model when the feature data is input to the trained model is calculated.

[0128] Next, a process for creating data for resetting is executed (FIG. 19 / STEP 21). In this process, term pairs whose average values ​​of the labels of the term pairs in the output of the trained model are within the above-mentioned predetermined range are selected, and data for resetting is created (not shown) so as to include the values ​​of these term pairs and their labels.

[0129] Next, a display process of the reset data is executed (FIG. 19 / STEP 22). In this display process, the reset data is displayed on the display 1a (output step). This enables the user to check the data of the term pair with low prediction accuracy and to create learning data by correcting the labels of the term pair. Furthermore, when the learning of the predictive model is re-executed using such corrected learning data, the prediction accuracy of the trained model can be improved. In this case, the user may add data or correct the labels of the data in addition to or instead of the learning data.

[0130] Next, the second trained model creation process will be described with reference to Fig. 20. This second trained model creation process is for creating a second trained model as described below, and is executed when the above-mentioned second predetermined keyword is input by the user's operation of the input interface 1c.

[0131] In this second trained model creation process, first, a process of acquiring a second patent document group is executed (FIG. 20 / STEP 31). In this acquisition process, as described above, the second patent document group is acquired by searching the patent document database 20 based on the second predetermined keyword (patent document group acquisition step).

[0132] Next, a process for acquiring a second term group is executed (FIG. 20 / STEP 32). In this process, as described above, the second term group is acquired by data mining the patent document group and the second patent document group (term group acquisition step).

[0133] Next, a process for creating second feature data is executed (FIG. 20 / STEP 33). In this process, as described above, the second feature data is created based on the patent documents, the second patent documents, and the second terminology group (feature data creation step).

[0134] Following this process of creating the second feature amount data, a process of creating second learning data is executed (FIG. 20 / STEP 34). In this process, as described above, the second learning data is created from the second feature amount data by referring to the status concept database 21 (learning data creation step).

[0135] Next, a second model learning process is executed (FIG. 20 / STEP 35). In this second model learning process, as described above, a prediction model is learned by a predetermined learning algorithm using the second learning data (model learning step). By executing the above processes, a second trained model is created.

[0136] Next, the second status concept data group process will be described with reference to Fig. 21. This second status concept data group process is for creating a second status concept data group and displaying it on the display 1a, as described below, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0137] In this second status conceptual data group process, first, a process of creating a second status conceptual data group is executed (FIG. 21 / STEP 40). In this process, as described above, the second status conceptual data group is created using the second trained model and the second feature amount data (second status conceptual data group creation step).

[0138] Next, a display process of the second status concept data group is executed (FIG. 21 / STEP 41). In this display process, as described above, the second status concept data group is displayed on the display 1a (output step). In this case, in addition to the second status concept data group, the status concept data group may be configured to be displayed on the display 1a at the same time.

[0139] As described above, according to the data processing device 1 of the first embodiment, learning data is created as data in which labels representing status concept relationships in the term pairs of feature data are added to the feature data, and learning of a prediction model, which is a classification model, is executed using this learning data. Then, a conceptual data group representing the relationships between superordinate and subordinate concepts of the term pairs of the feature data is created using the output of the trained model when the feature data is input to the trained model.

[0140] Here, the feature data includes term pairs, which are pairs of two terms included in the claims of the patent documents, among the terms included in the term group. When such feature data is input to the trained model, the output of the trained model appropriately predicts the relationship between the superordinate and subordinate concepts of the term pairs in the feature data.

[0141] Therefore, by creating a status concept data group using the output of such a prediction model, the status concept data group can be created as data in which the superordinate and subordinate concept relationships of term pairs are appropriately classified. In addition, by searching a database of patent documents, the patent document group is obtained as a set of patent documents that include a predetermined keyword in their claims. As a result, the status concept data group can be created as data in which the superordinate and subordinate concept relationships between terms in the claims of the set of patent documents are appropriately classified.

[0142] Furthermore, the status concept data group is created as tree-shaped graphic data in which the terms of the term pairs are arranged according to the relationship of superordinate and subordinate concepts, and is displayed on the display 1a. This allows the user to easily understand the relationship of superordinate and subordinate concepts between the terms. Also, when a term of the status concept data group is selected by the user operating the input interface 1c, the number of the patent document including the term in the claim is displayed. This allows the user to further understand the number of the patent document including each term of the term pair, in addition to the relationship of superordinate and subordinate concepts of the term pair.

[0143] In addition, in the process of creating the data for resetting, the data for resetting is created so as to include term pairs and label values ​​whose average value of the labels of the term pairs in the output of the trained model is within a predetermined range, and is displayed on the display 1a. This allows the user to check the data of the term pairs with low prediction accuracy, and to create training data in which the labels of the term pairs are corrected, or to add or correct the contents of the status concept data. Furthermore, when the training of the prediction model is re-run using such corrected training data, the prediction accuracy of the trained model can be improved.

[0144] On the other hand, in the second trained model creation process, a second patent document group is acquired based on a second predetermined keyword, a second term group is acquired by data mining the patent document group and the second patent document group, and second feature data is created based on the patent document group, the second patent document group, and the second term group. Furthermore, second training data is created from the second feature data, and a predictive model is trained using this second training data to create a second trained model, and a second status concept data group is created using the second trained model and the second feature data. Thereby, the second status concept data group can be created so as to include terms similar to the status concept data group in at least one of concepts, properties, and classifications.

[0145] In the first embodiment, the data processing device is an example of a personal computer type data processing device 1, but the data processing device may be configured with multiple personal computers or one or more servers (in this case, the server also includes a cloud server), or may be configured by combining a personal computer and a server.

[0146] In addition, the first embodiment is an example in which a display 1a is used as an output device, but the output device of the present invention is not limited to this, and may be any device capable of outputting a status concept data group from a data processing device. For example, if the data processing device is a cloud server, a terminal device (such as a personal computer or a printer) that receives and outputs the status concept data group output from the cloud server via a communication network may be used as the output device.

[0147] Furthermore, in the first embodiment, a set of patent documents in which a predetermined keyword is included in a claim is acquired as a patent document group, but instead, a set of patent documents in which a predetermined keyword is included in any of the entire sentences, including the claims, in the patent document may be acquired as a patent document group. In that case, the feature data and the learning data may be configured so as not to include claim data.

[0148] On the other hand, the first embodiment is an example in which the patent document number is included in the feature data and the learning data, but instead, the patent document number may be configured not to be included in the feature data and the learning data. In that case, for example, a database linking claims and patent document numbers may be stored in the data processing device, and the patent document number may be displayed when a term in the status concept data group is selected.

[0149] In addition, the first embodiment is an example in which the status concept data group is created as two-dimensional image data as shown in FIG. 15, but the status concept data group may be created as image data represented in three-dimensional format as shown in FIG. 22, or may be created as tabular data.

[0150] Furthermore, when creating the status concept data group, a name matching process may be performed for each term by referring to a name matching dictionary (not shown), and then the status concept data group may be created.

[0151] On the other hand, the first embodiment is an example in which a Transformer natural language model is used as the predetermined learning algorithm, but Random Forest, XGBoost, a support vector machine, a neural network, or the like may also be used as the predetermined learning algorithm.

[0152] In addition, the first embodiment is an example in which a predictive model trained by the model training unit 15 is used as the trained model, but instead, a general-purpose trained model (e.g., BERT: Bidirectional Encoder Representations from Transformers) may be used as is without fine-tuning.

[0153] Furthermore, a trained model created by transfer learning may be used as the trained model. For example, the trained data created by the information processing device 1 may be transmitted to a server, and the server may execute learning of a predictive model such as BERT to create a trained model, and the trained model may be transmitted to the information processing device 1 and stored therein. In that case, in the information processing device 1, each time feature data is created, the feature data may be input to the trained model without executing learning of a predictive model, thereby creating a position concept data group.

[0154] On the other hand, the status concept data group creation unit 16 of the first embodiment may be configured to create data to which a label indicating the relationship of a superordinate concept or subordinate concept between terms in the status concept data group is added each time a status concept data group is created, and to add the data to the status concept database 21 as shown by the two-dot chain line in Fig. 2. When configured in this way, the status concept database with an increased amount of data is used to create learning data, thereby further improving the learning effect of the prediction model using the learning data.

[0155] Also, in the first embodiment, in the information processing device 1, learning data is created and prediction model learning (additional learning) is performed every time feature data is created, but instead, after prediction model learning has been performed M (M is an integer equal to or greater than 1) times, when feature data is created, the feature data is input to the trained model without performing prediction model learning, thereby creating a status concept data group. For example, the second status concept data may be created by inputting the above-mentioned second feature data to a trained model that has been trained once with a predetermined keyword.

[0156] Furthermore, in the first embodiment, the learning data is created by the learning data creating unit 14, but instead, the learning data may be created by the user.

[0157] In addition, in the first embodiment, "polyoxyethylene diglycerin saturated fatty acid ester" is used as the predetermined keyword, but the predetermined keyword of the present invention is not limited to this, and various keywords can be used. For example, silicone oil may be used as the predetermined keyword.

[0158] Furthermore, although the first embodiment is an example in which "polyoxyethylene diglycerol stearate" is used as the second predetermined keyword, the second predetermined keyword of the present invention is not limited to this, and may be any keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. For example, when "polyoxyalkylene polyglyceryl ether" is the predetermined keyword, "polyoxyethylene diglyceryl ether" may be used as the second predetermined keyword.

[0159] In addition, the first embodiment is an example in which a group of patent documents is obtained from the patent document database 20 stored in the data processing device 1, but for example, if the patent document database is stored in a processing device such as a server other than the data processing device 1, the group of patent documents may be configured to be obtained from the patent document database in the processing device by communication between the data processing device 1 and the processing device.

[0160] Furthermore, the first embodiment is an example in which learning data is created using the status concept database 21 stored in the data processing device 1, but for example, if the status concept database is stored in a processing device such as a server other than the data processing device 1, the learning data may be configured to be created using the status concept database in the processing device through communication between the data processing device 1 and the processing device.

[0161] On the other hand, the first embodiment is an example using the status concept database 21 shown in Fig. 10, but the status concept database of the present invention is not limited to this, and may be any database that represents the status concept relationship between a pair of terms. For example, in the status concept database, the symbol ">" may be used to indicate that term 1 is a superordinate concept of term 2, the symbol ">" may be used to indicate that term 1 is a subordinate concept of term 2, and the symbol ">" may be used to indicate that term 1 and term 2 are not in a superordinate or subordinate concept relationship ....

[0162] 10, the status concept database 21 may be configured so that only the values ​​"1" and "0" representing the relationship between a superordinate concept and a subordinate concept, respectively, are set as the labels representing the relationship between term 1 and term 2. In that case, the learning data creation unit 14 may be configured so that, when the term pair of the feature amount data is not included in the status concept database 21, the value "2" is set as the label representing the relationship between term 1 and term 2 in the term pair of the feature amount data.

[0163] 10, the status concept database 21 may be configured so that, in addition to the values ​​"1", "0", and "2" respectively representing the superordinate concept relationship, subordinate concept relationship, and no superordinate or subordinate concept relationship, the value "3" is set as a label representing a coordinate concept as a label representing the relationship between term 1 and term 2. In that case, the learning data creation unit 14 may be configured to set values ​​from "0" to "3" as labels for term pairs in the learning data.

[0164] Next, a data processing device according to a second embodiment of the present invention will be described. The data processing device 1A (see FIG. 23) according to this embodiment is the same personal computer type as the data processing device 1 in FIG. 1, so the following description will focus on the differences. Also, the same reference numerals are used for the same components as those in the first embodiment, and the description thereof will be omitted.

[0165] The data processing device 1A of this embodiment stores a non-patent document database 39, a patent document database 40, and a status concept database 41 (see FIG. 23). The details of these databases 39 to 41 will be described later.

[0166] As shown in Figure 23, the data processing device 1A has the functions of a non-patent document group acquisition unit 30, a term group acquisition unit 31, a patent document group acquisition unit 32, a feature data creation unit 33, a learning data creation unit 34, a model learning unit 35, a status concept data group creation unit 36, and an output unit 37.

[0167] First, in the non-patent document group acquisition unit 30, when a specified keyword is input by the user operating the input interface 1c, a group of non-patent documents is acquired by searching the non-patent document database 39 in the data processing device 1A based on the specified keyword.

[0168] Here, the non-patent literature database 39 contains many documents such as academic literature and technical literature, and the non-patent literature group is acquired as a collection of non-patent literature in which a predetermined keyword is included in text, figures, tables, etc. In the following explanation, an example will be explained in which the predetermined keyword is "polyoxyethylene diglycerin saturated fatty acid ester."

[0169] Next, the term group acquiring unit 31 acquires a term group by performing data mining on the non-patent literature group. This term group is acquired as a set of terms that appear in a predetermined order from the most frequently occurring terms included in the sentences of the non-patent literature group.

[0170] Next, the patent document group acquisition unit 32 acquires a patent document group by searching the patent document database 40 based on the term group. This patent document group is acquired as a collection of patent documents in which the term in the term group is included in the claim.

[0171] Furthermore, in the feature amount data creation unit 33, feature amount data (not shown) is created based on the group of patent documents and the group of terms by the same method as the feature amount data creation method of FIG. 4 described above.

[0172] Moreover, the learning data creation unit 34 creates learning data by referring to a status concept database 41 in the data processing device 1A. Here, the status concept database 41 is configured similarly to the status concept database 21 described above, and the learning data is created similarly to the learning data of FIG.

[0173] Next, in the model learning unit 35, using the learning data created as described above, learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm (for example, a natural language model of Transformer).

[0174] Next, the status concept data group creation unit 36 ​​creates a status concept data group using the trained model and the feature amount data. This status concept data group is not shown, but is created in the same manner as the status concept data group in Fig. 15 described above. That is, the status concept data group is created as image data in which the terms of the term pairs are arranged in accordance with the relationship between superordinate and subordinate concepts.

[0175] Then, the data of the status concept data group created as described above is output from the output unit 37 to the display 1a (output process). As a result, the status concept data group is displayed on the display 1a (not shown).

[0176] When the status concept data group is displayed on the display 1a in this manner, when the user selects any term in the status concept data group via the input interface 1c, at least one of data representing a non-patent document containing the term (the name or number of the non-patent document) and the number of a patent document which includes the term in a claim is displayed, although not shown.

[0177] In addition, in the elements 30 to 37 of the above data processing device 1A, when a second predetermined keyword (e.g., polyoxyethylene diglycerin stearate ester) that is similar to the predetermined keyword in at least one of concept, property, and classification is input, various processes are executed as described below.

[0178] In other words, when a second specified keyword is input by the user operating the input interface 1c, the non-patent document group acquisition unit 30 acquires the second non-patent document group by searching the non-patent document database 39 in the data processing device 1A based on the second specified keyword.

[0179] Next, the term group acquiring unit 31 acquires a second term group by performing data mining on the non-patent document group and the second non-patent document group (not shown). This second term group is acquired as a set of terms that appear in a predetermined order in descending order of frequency of appearance among terms contained in sentences of the non-patent document group and the second non-patent document group.

[0180] Next, the patent document group acquisition unit 32 acquires a second patent document group by searching the patent document database 40 based on the second term group. This second patent document group is acquired as a set of patent documents in which the terms in the second term group are included in the claims.

[0181] Furthermore, in the feature amount data creation section 33, second feature amount data is created based on the second group of patent documents and the second group of terms, similarly to the feature amount data described above (not shown).

[0182] Moreover, the learning data creation unit 34 creates second learning data (not shown) from the second feature amount data by referring to the status concept database 41, in the same way as the learning data described above.

[0183] Next, in the model learning unit 35, learning of a prediction model, which is a classification model, is executed by the above-mentioned predetermined learning algorithm using the second learning data created as described above. As a result, a second trained model is created.

[0184] Furthermore, in the status concept data group creation unit 36, a second status concept data group is created (not shown) using the second trained model and the second feature amount data, similar to the above-mentioned status concept data group.

[0185] Then, the data of the second position concept data group created as described above is output from the output unit 17 to the display 1a (output process). As a result, the second position concept data group is displayed on the display 1a (not shown).

[0186] Next, various arithmetic processes executed by the data processing device 1A of this embodiment will be described. First, the trained model creation process will be described with reference to Fig. 24. This trained model creation process is for creating a trained model as described below, and is executed when the above-mentioned predetermined keyword is input by the user's operation of the input interface 1c.

[0187] In this trained model creation process, first, a process of acquiring a group of non-patent documents is executed (FIG. 24 / STEP 51). In this acquisition process, as described above, a group of non-patent documents is acquired by searching the non-patent document database 39 based on a predetermined keyword (non-patent document group acquisition step).

[0188] Next, a term group acquisition process is executed (FIG. 24 / STEP 52). In this acquisition process, as described above, a term group is acquired by data mining a group of non-patent documents (term group acquisition step).

[0189] Next, a patent document acquisition process is executed (FIG. 24 / STEP 53). In this acquisition process, as described above, a patent document group is acquired by searching the patent document database 40 based on the term group (patent document group acquisition step).

[0190] Next, a process for creating feature data is executed (FIG. 24 / STEP 54). In this process, as described above, feature data is created based on the patent documents and the terminology group (feature data creation step).

[0191] Following this feature data creation process, a learning data creation process is executed (FIG. 24 / STEP 55). In this creation process, as described above, the learning data is created by referring to the status concept database 41 (learning data creation step).

[0192] Next, a model learning process is executed (FIG. 24 / STEP 56). In this model learning process, as described above, learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm using the learning data (model learning step). By executing the above processes, a trained model is created.

[0193] Next, the status concept data group process will be described with reference to Fig. 25. This status concept data group process is for creating a status concept data group and displaying it on the display 1a, as described below, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0194] In this status concept data group process, first, a status concept data group creation process is executed (FIG. 25 / STEP 60). In this creation process, as described above, a status concept data group is created using a trained model and feature amount data (status concept data group creation step).

[0195] Next, a display process of the status concept data group is executed (FIG. 25 / STEP 61). In this display process, although not shown, as described above, the status concept data group is displayed on the display 1a (output step).

[0196] Next, the second trained model creation process will be described with reference to Fig. 26. This second trained model creation process is for creating a second trained model as described below, and is executed when the above-mentioned second predetermined keyword is input by the user's operation of the input interface 1c.

[0197] In this second trained model creation process, first, a process of acquiring a second non-patent document group is executed (FIG. 26 / STEP 71). In this acquisition process, as described above, the second non-patent document group is acquired by searching the non-patent document database 39 based on the second predetermined keyword (non-patent document group acquisition step).

[0198] Next, a process for acquiring a second term group is executed (FIG. 26 / STEP 72). In this process, as described above, the non-patent literature and the second non-patent literature group are data mined to acquire the second term group (term group acquisition step).

[0199] Next, a process for acquiring the second group of patent documents is executed (FIG. 26 / STEP 73). In this process, as described above, the second group of patent documents is acquired by searching the patent document database 40 based on the second group of terms.

[0200] Furthermore, a process for creating second feature data is executed (FIG. 26 / STEP 74). In this process, as described above, the second feature data is created based on the second group of patent documents and the second group of terms (feature data creation step).

[0201] Following this process of creating the second feature amount data, a process of creating the second learning data is executed (FIG. 26 / STEP 75). In this process, as described above, the second learning data is created from the second feature amount data by referring to the status concept database 41 (learning data creation step).

[0202] Next, the second model learning process is executed (FIG. 26 / STEP 76). In this second model learning process, as described above, the second learning data is used to learn a prediction model by a predetermined learning algorithm (model learning step). By executing the above processes, a second trained model is created.

[0203] Next, the second status concept data group process will be described with reference to Fig. 27. This second status concept data group process is for creating a second status concept data group and displaying it on the display 1a, as described below, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0204] In this second status conceptual data group process, first, a process of creating a second status conceptual data group is executed (FIG. 27 / STEP 80). In this process, as described above, the second status conceptual data group is created using the second trained model and the second feature amount data (second status conceptual data group creation step).

[0205] Next, a display process of the second status concept data group is executed (FIG. 27 / STEP 81). In this display process, as described above, the second status concept data group (not shown) is displayed on the display 1a (output step). In this case, in addition to the second status concept data group, the status concept data group may be configured to be displayed on the display 1a at the same time.

[0206] As described above, according to the data processing device 1A of the second embodiment, learning of a prediction model, which is a classification model, is performed using learning data, which is data in which labels representing status concept relationships in term pairs of feature data are added to feature data. Then, a status concept data group representing the relationships between superordinate and subordinate concepts between term pairs of feature data is created using the output of the trained model when the feature data is input to the trained model.

[0207] Here, the feature data includes term pairs, which are pairs of terms in a term group present in the claims of the patent documents. When such feature data is input to a trained model, the output of the trained model appropriately predicts the relationship between superordinate and subordinate concepts between the term pairs in the feature data. Therefore, by creating a status concept data group using the output of such a prediction model, the status concept data group can be created as data in which the relationship between superordinate and subordinate concepts between the term pairs is appropriately classified.

[0208] In addition, by searching a database of non-patent documents, a group of non-patent documents is obtained as a set of non-patent documents containing a predetermined keyword, by data mining the group of non-patent documents, a group of terms that are terms in the group of non-patent documents that appear in a predetermined order from the most frequently occurring terms is obtained, and by searching a database of patent documents based on the group of terms, a group of patent documents is obtained that is a set of patent documents in which the group of terms is included in claims. This makes it possible to create a group of status concept data as data that appropriately classifies the superordinate and subordinate concept relationships between terms in the claims of the set of patent documents and the superordinate and subordinate concept relationships between terms in non-patent documents.

[0209] Furthermore, the status concept data group is created as tree-shaped graphic data in which the terms of the term pairs are arranged according to the relationship of superordinate and subordinate concepts, and is displayed on the display 1a. This allows the user to easily grasp the relationship of superordinate and subordinate concepts between the terms. Furthermore, when a term of the status concept data group is selected by the user operating the input interface 1c, data representing a non-patent document containing the term (the name or number of the non-patent document) and / or the number of a patent document containing the term in a claim are displayed. This allows the user to further grasp, in addition to the status concept relationship of the term pair, data representing a non-patent document containing each term of the term pair and / or the number of a patent document in which each term of the term pair is included in a claim.

[0210] Furthermore, in the second trained model creation process, a second group of non-patent documents is acquired based on a second predetermined keyword, a second group of terms is acquired by data mining the group of non-patent documents and the second group of non-patent documents, and a second group of patent documents is acquired by searching the patent document database 40 based on the second group of terms. Furthermore, second feature amount data is created based on the second group of patent documents and the second group of terms, and second training data is created from the second feature amount data by referring to the status concept database 41.

[0211] Then, a second trained model is created by using this second training data to train a prediction model, and a second status concept data group is created using the second trained model and the second feature amount data. This allows the second status concept data group to be created so as to include terms similar to the status concept data group in at least one of concepts, properties, and classifications.

[0212] In the second embodiment, the data processing device is an example of a personal computer type data processing device 1A, but the data processing device may be configured with multiple personal computers or one or more servers (in this case, the server also includes a cloud server), or may be configured by combining a personal computer and a server.

[0213] In addition, the second embodiment is an example in which a display 1a is used as an output device, but the output device of the present invention is not limited to this, and may be any device capable of outputting a status concept data group from a data processing device. For example, if the data processing device is a cloud server, a terminal device (such as a personal computer or a printer) that receives and outputs the status concept data group output from the cloud server via a communication network may be used as the output device.

[0214] Furthermore, while the second embodiment is an example in which a set of patent documents in which a term group is included in a claim is acquired as a set of patent documents, instead, a set of patent documents in which a term group is included in any of the entire sentences, including the claims, in a patent document may be acquired as a set of patent documents. In that case, the feature data and the learning data may be configured so as not to include claim data.

[0215] In addition, the second embodiment is an example in which the status concept data group is two-dimensional image data similar to that of Figure 15, but the status concept data group may be created as three-dimensional image data similar to that of Figure 22 described above, or may be created as tabular data.

[0216] Furthermore, when creating the status concept data group, a name matching process may be performed for each term by referring to a name matching dictionary (not shown), and then the status concept data group may be created.

[0217] On the other hand, the second embodiment is an example in which a Transformer natural language model is used as the predetermined learning algorithm, but Random Forest, XGBoost, a support vector machine, a neural network, or the like may also be used as the predetermined learning algorithm.

[0218] In addition, the second embodiment is an example in which a predictive model trained by the model training unit 35 is used as the trained model, but instead, a general-purpose trained model (e.g., BERT: Bidirectional Encoder Representations from Transformers) may be used as is without fine-tuning.

[0219] Furthermore, a trained model created by transfer learning may be used as the trained model. For example, the trained data created by the information processing device 1A may be transmitted to a server, and the server may execute learning of a predictive model such as BERT to create a trained model, and the trained model may be transmitted to the information processing device 1A. In that case, in the information processing device 1A, each time feature data is created, the feature data may be input to the trained model without executing learning of a predictive model, thereby creating a status concept data group.

[0220] On the other hand, the status concept data group creation unit 36 ​​of the second embodiment may be configured to create data to which a label indicating the relationship of a superordinate concept or subordinate concept between terms in the status concept data group is added each time a status concept data group is created, and to add the data to the status concept database 41 as shown by the two-dot chain line in Fig. 23. When configured in this way, the status concept database with an increased amount of data is used to create learning data, thereby further improving the learning effect of the prediction model using the learning data.

[0221] Also, in the second embodiment, in the information processing device 1A, learning data is created and prediction model learning (additional learning) is performed every time feature data is created, but instead, after prediction model learning has been performed M (M is an integer equal to or greater than 1) times, when feature data is created, the feature data is input to the trained model without performing prediction model learning, thereby creating a status concept data group. For example, the second status concept data may be created by inputting the above-mentioned second feature data into a trained model that has been trained once with a predetermined keyword.

[0222] Furthermore, the second embodiment is an example in which the learning data is created by the learning data creation unit 34, but instead of this, a configuration may be adopted in which the user creates the learning data.

[0223] In the second embodiment, "polyoxyethylene diglycerin saturated fatty acid ester" is used as the predetermined keyword, but the predetermined keyword of the present invention is not limited to this, and various keywords can be used. For example, silicone oil may be used as the predetermined keyword.

[0224] Furthermore, although the second embodiment is an example in which "polyoxyethylene diglycerol stearate" is used as the second predetermined keyword, the second predetermined keyword of the present invention is not limited to this, and may be any keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. For example, when "polyoxyalkylene polyglyceryl ether" is the predetermined keyword, "polyoxyethylene diglyceryl ether" may be used as the second predetermined keyword.

[0225] Furthermore, the second embodiment is an example in which a group of non-patent documents is acquired from the non-patent document database 39 stored in the data processing device 1A. However, for example, if the non-patent document database is stored in a processing device such as a server other than the data processing device 1A, the group of non-patent documents may be configured to be acquired from the non-patent document database in the processing device by communication between the data processing device 1A and the processing device.

[0226] Furthermore, the second embodiment is an example in which a group of patent documents is obtained from a patent document database 40 stored in the data processing device 1A. However, for example, if the patent document database is stored in a processing device such as a server other than the data processing device 1A, the group of patent documents may be configured to be obtained from the patent document database in the processing device through communication between the data processing device 1A and the processing device.

[0227] On the other hand, the second embodiment is an example in which learning data is created using a status concept database 41 stored in the data processing device 1A, but for example, if the status concept database is stored in a processing device such as a server other than the data processing device 1A, the learning data may be configured to be created using the status concept database in the processing device through communication between the data processing device 1A and the processing device.

[0228] Meanwhile, the second embodiment is an example using a status concept database 41 similar to the status concept database 21 described above, but the status concept database of the present invention is not limited to this, and may be anything that represents the status concept relationship between a term pair. For example, in the status concept database, the symbol ">" may be used to indicate that term 1 is a superordinate concept of term 2, the symbol ">" may be used to indicate that term 1 is a subordinate concept of term 2, and the symbol ">" may be used to indicate that term 1 and term 2 are not in a superordinate or subordinate concept relationship ....

[0229] Also, the status concept database 41 may be configured so that only the values ​​"1" and "0" representing the relationship between a superordinate concept and a subordinate concept, respectively, are set as labels representing the relationship between term 1 and term 2. In that case, the learning data creation unit 34 may be configured so that, when the term pair of the feature data is not included in the status concept database 41, the value "2" is set as the label representing the relationship between term 1 and term 2 in the term pair of the feature data.

[0230] Furthermore, in the status concept database 41, in addition to the values ​​"1", "0", and "2" respectively representing the superordinate concept relationship, subordinate concept relationship, and no superordinate or subordinate concept relationship, the value "3" may be set as a label representing a coordinate concept as a label representing the relationship between term 1 and term 2. In that case, the learning data creation unit 34 may be configured to set values ​​from "0" to "3" as labels for term pairs in the learning data.

[0231] Next, a data processing device according to a third embodiment of the present invention will be described. The data processing device 1B (see FIG. 28) of this embodiment is the same personal computer type as the data processing device 1 of FIG. 1, so the following description will focus on the differences. Also, the same reference numerals are used for the same components as those of the first embodiment, and the description thereof will be omitted.

[0232] The data processing device 1B of this embodiment stores a patent document database 59, a basic terminology database 60, and a status concept database 61 (see FIG. 28). The details of these databases 59 to 61 will be described later.

[0233] As shown in Figure 28, the data processing device 1B has the functions of a patent document group acquisition unit 50, a term group acquisition unit 51, a basic term group acquisition unit 52, a feature data creation unit 53, a learning data creation unit 54, a model learning unit 55, a status concept data group creation unit 56, and an output unit 57.

[0234] First, in the patent document group acquisition unit 50, when a specified keyword is input by the user operating the input interface 1c, a patent document group is acquired by searching the patent document database 59 in the data processing device 1B based on the specified keyword.

[0235] Here, the patent document database 59 includes many documents including claims, such as patent publications, JP Patent Publications, and utility model registration publications, and the patent document group is acquired as a collection of patent documents in which a predetermined keyword is included in a claim. In the following explanation, an example will be explained in which the predetermined keyword is "silicone oil."

[0236] Next, the term group acquiring unit 51 performs data mining on the group of patent documents to acquire a term group as shown in Fig. 29. This term group is acquired as a set of terms included in the claims of the group of patent documents, in descending order of frequency of appearance up to a predetermined rank.

[0237] Next, the basic term group acquiring unit 52 acquires the basic term group shown in Fig. 30 from the term group by referring to the basic term database 60. Here, the basic term database 60 contains basic terms that represent the raw materials of various substances including silicone oil, and the basic term group is acquired by extracting the basic terms contained in the basic term database 60 from the term group.

[0238] Furthermore, the feature data creation unit 53 creates feature data shown in Fig. 31 from the basic term group and the term group. As shown in Fig. 31, the feature data is created to include a selected term pair consisting of selected term 1 and selected term 2. These selected term pairs correspond to a combination of one basic term extracted from the basic term group and another basic term extracted from the basic term group or one term extracted from the term group, as selected term 1 and selected term 2. Note that in the data in Fig. 31, specific indications of application numbers and claim data are omitted, and this also applies to Fig. 32 described later.

[0239] Moreover, the learning data creation unit 54 creates learning data shown in Fig. 32 from the feature amount data shown in Fig. 31 using a status concept database 61 in the data processing device 1B. Here, the status concept database 61 defines status concept relationships such as relationships between superordinate and subordinate concepts between selected term pairs as shown in Fig. 33.

[0240] 33, the label values ​​represent the status concept relationship between the selected term pairs, and specifically, a label value of "1" represents that selected term 1 is a superordinate concept of selected term 2. A label value of "0" represents that selected term 1 is a subordinate concept of selected term 2, and that selected term 1 and selected term 2 are not in a superordinate or subordinate concept relationship.

[0241] The learning data creation unit 54 refers to this status concept database 61 and adds labels to the feature amount data in FIG. 31, thereby creating the learning data in FIG.

[0242] Next, in the model learning unit 55, using the learning data created as described above, learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm (for example, a natural language model of Transformer).

[0243] Next, the status concept data group creation unit 56 creates a status concept data group using the trained model and the feature amount data. This status concept data group is tree-shaped image data in which the selected terms of the selected term pairs described above are arranged according to the relationship between superordinate and subordinate concepts, as shown in Fig. 34, and is created by the method described below.

[0244] That is, in the status concept data group creation unit 56, the status concept relationship between the selected term pairs is determined based on the label values ​​of the output of the trained model when feature data is input into the trained model, and a status concept data group is created by arranging the selected term pairs based on the determination results.

[0245] For example, if the feature data selected term 1 and selected term 2 are "silicone oil" and "cyclomethicone," the output of the trained model will be "1," and a status concept data group will be created with the subordinate selected term "cyclomethicone" hanging from the superior concept selected term "silicone oil," as shown in Figure 34.

[0246] Furthermore, when selected term 1 and selected term 2 are "cyclomethicone" and "cyclopentasiloxane," the output of the trained model is "1," and a status concept data group is created with the subordinate concept selected term "cyclopentasiloxane" hanging from the superior concept selected term "cyclomethicone," as shown in Figure 34. By continuing to execute the above process, the status concept data group shown in Figure 34 is finally created.

[0247] Then, the data of the status concept data group created as described above is output from the output unit 57 to the display 1a (output process). As a result, the status concept data group is displayed on the display 1a in the state shown in FIG.

[0248] In this state where the status concept data group is displayed on the display 1a, when the user selects any basic term among the selected terms in the status concept data group via the input interface 1c, the basic term number of the basic term is displayed. For example, as shown in Figure 35, when the basic term "dimethicone" is selected, three basic term numbers B002, B003, and B004 are displayed.

[0249] These basic term numbers B002 to B004 are stored in a database linked to the product names of the basic terms, as shown in the database of Fig. 37. This allows the user to understand the product names "Silicone XXXXX", "Silicone KKKKK", and "Silicone SSSSS" of the basic term "Dimethicone" by referring to Fig. 37.

[0250] In this case, when the basic term numbers B002 to B004 are displayed on the display 1a and one of the basic term numbers B002 to B004 is selected, the product name described in Fig. 37 may be displayed in correspondence with the selected basic term number, or the database in Fig. 37 may be displayed on the display 1a in a separate screen. Also, when one of the basic terms in the status concept data group is selected, the number of the patent document containing the basic term may be displayed on the display 1a instead of or in addition to the basic term number.

[0251] Also, as shown in Fig. 36, when the basic term "cyclomethicone" is clicked, the basic term numbers (B001, B003) of the subordinate basic term "cyclopentasiloxane" are displayed, and the basic term number (B009) of the subordinate basic term "cyclohexasiloxane" is displayed. This allows the user to understand the product names of the basic terms "cyclopentasiloxane" and "cyclohexasiloxane" by referring to the database in Fig. 37 described above.

[0252] In addition, in the elements 50 to 57 of the data processing device 1B described above, when a second predetermined keyword (e.g., phenyl trimethicone) that is similar to the predetermined keyword in at least one of concept, property, and classification is input, various processes are executed as described below.

[0253] That is, when a second specified keyword is input by the user operating the input interface 1c, the patent document group acquisition unit 50 acquires the second patent document group by searching the patent document database 59 in the data processing device 1B based on the second specified keyword.

[0254] Next, the term group acquiring unit 51 acquires a second term group by performing data mining on the group of patent documents and the second patent document (not shown). This second term group is acquired as a set of terms in the claims of the group of patent documents and the second group of patent documents, in descending order of appearance frequency up to a predetermined rank.

[0255] Next, the basic term group acquiring unit 52 refers to the basic term database 60 and extracts basic terms contained in the basic term database 60 from the second term group, thereby acquiring a second basic term group (not shown).

[0256] Furthermore, in the feature amount data creating section 53, second feature amount data is created based on the second term group and the second basic term group, similar to the feature amount data in FIG. 31 described above (not shown).

[0257] In addition, in the learning data creation unit 54, by referring to the above-mentioned status concept database 61, second learning data is created from the above-mentioned second feature data, similar to the learning data of Figure 32 described above (not shown).

[0258] Next, in the model learning unit 55, learning of a prediction model, which is a classification model, is executed by the above-mentioned predetermined learning algorithm using the second learning data created as described above. Hereinafter, the prediction model that has been learned using the second learning data is referred to as a "second trained model."

[0259] Furthermore, in the status concept data group creation unit 56, a second status concept data group is created (not shown) using the second learned model and the second feature data, similar to the status concept data group of Figure 34 described above.

[0260] Then, the data of the second status concept data group created as described above is output from the output unit 57 to the display 1a (output process). As a result, the second status concept data group is displayed on the display 1a (not shown).

[0261] Next, various arithmetic processes executed by the data processing device 1B of this embodiment will be described. First, the trained model creation process will be described with reference to Fig. 38. This trained model creation process is for creating a trained model as described below, and is executed when the above-mentioned predetermined keyword is input by the user's operation of the input interface 1c.

[0262] In this trained model creation process, first, a patent document group acquisition process is executed (FIG. 38 / STEP 91). In this acquisition process, as described above, a patent document group is acquired by searching the patent document database 59 based on a predetermined keyword (patent document group acquisition step).

[0263] Next, a term group acquisition process is executed (FIG. 38 / STEP 92). In this acquisition process, as described above, a term group is acquired by data mining a group of patent documents (term group acquisition step).

[0264] Next, a basic term group acquisition process is executed (FIG. 38 / STEP 93). In this acquisition process, as described above, the basic term group is acquired from the term group by referring to the basic term database 60 (basic term group acquisition step).

[0265] Furthermore, a process for creating feature data is executed (FIG. 38 / STEP 94). In this process, as described above, feature data is created from the term group and the basic term group (feature data creating step).

[0266] Following this feature data creation process, a learning data creation process is executed (FIG. 38 / STEP 95). In this creation process, as described above, the learning data is created by referring to the status concept database 61 (learning data creation step).

[0267] Next, a model learning process is executed (FIG. 38 / STEP 96). In this model learning process, as described above, learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm using the learning data (model learning step). By executing the above processes, a trained model is created.

[0268] Next, the status concept data group process will be described with reference to Fig. 39. This status concept data group process is for creating a status concept data group and displaying it on the display 1a, as described below, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0269] In this status concept data group process, first, a status concept data group creation process is executed (FIG. 39 / STEP 100). In this creation process, as described above, a status concept data group is created using a trained model and feature amount data (status concept data group creation step).

[0270] Next, a display process of the status concept data group is executed (FIG. 39 / STEP 101). In this display process, although not shown, as described above, the status concept data group is displayed on the display 1a (output step).

[0271] Next, the second trained model creation process will be described with reference to Fig. 40. This second trained model creation process is for creating the above-mentioned second trained model, and is executed when the above-mentioned second predetermined keyword is input by the user's operation of the input interface 1c.

[0272] In this second trained model creation process, first, a process of acquiring a second patent document group is executed (FIG. 40 / STEP 111). In this acquisition process, as described above, the second patent document group is acquired by searching the patent document database 59 based on the second predetermined keyword (patent document group acquisition step).

[0273] Next, a process for acquiring a second term group is executed (FIG. 40 / STEP 112). In this process, as described above, the patent document and the second patent document group are data mined to acquire the second term group (term group acquisition step).

[0274] Furthermore, a process of acquiring a second basic term group is executed (FIG. 40 / STEP 113). In this process, as described above, the second basic term group is acquired by referring to the basic term database 60 and extracting basic terms contained in the basic term database 60 from the second term group.

[0275] Next, a process for creating second feature data is executed (FIG. 40 / STEP 114). In this process, as described above, the second feature data is created based on the second term group and the second basic term group (feature data creation step).

[0276] Following this process of creating the second feature amount data, a process of creating the second learning data is executed (FIG. 40 / STEP 115). In this process, as described above, the second learning data is created from the second feature amount data by referring to the status concept database 61 (learning data creation step).

[0277] Next, a second model learning process is executed (FIG. 40 / STEP 116). In this second model learning process, as described above, learning of a prediction model, which is a classification model, is executed by a predetermined learning algorithm using the second learning data (model learning step). By executing the above processes, a second trained model is created.

[0278] Next, the second status concept data group process will be described with reference to Fig. 41. This second status concept data group process is for creating a second status concept data group and displaying it on the display 1a, as described below, and is executed in response to a predetermined operation of the input interface 1c by the user.

[0279] In this second status conceptual data group process, first, a process of creating a second status conceptual data group is executed (FIG. 41 / STEP 120). In this process, as described above, the second status conceptual data group is created using the second trained model and the second feature amount data (second status conceptual data group creation step).

[0280] Next, the display process of the second status concept data group is executed (FIG. 41 / STEP 121). In this display process, as described above, the second status concept data group (not shown) is displayed on the display 1a (output step).

[0281] As described above, according to the data processing device 1B of the third embodiment, learning of a prediction model, which is a classification model, is performed using learning data, which is data in which labels representing status concept relationships in selected term pairs of feature data are added to feature data. Then, a status concept data group that defines the status concept relationships of selected term pairs of feature data is created using the output of the trained model when the feature data is input to the trained model.

[0282] Here, the feature data includes selected term pairs, which are pairs of a basic term included in the basic term group and another basic term included in the basic term group or a term included in the term group. When such feature data is input to a trained model, the output of the trained model appropriately predicts the status concept relationship of the selected term pair of the feature data. Therefore, by creating a status concept data group using the output of such a prediction model, the status concept data group can be created as data in which the status concept relationship of the selected term pair is appropriately classified.

[0283] Furthermore, the status concept data group is created as tree-shaped graphic data in which the selected terms of the selected term pairs are arranged according to the relationship between superordinate and subordinate concepts, and is displayed on the display 1a. This allows the user to easily understand the status concept relationship between the selected terms. Also, when the user operates the input interface 1c to click on a basic term in the selected terms of the status concept data group, a basic term number linked to the product name of the basic term is displayed. This allows the user to further understand the product name of the basic term in addition to the status concept relationship between the selected terms.

[0284] On the other hand, in the second trained model creation process, a second patent document group is acquired based on a second predetermined keyword, a second term group is acquired by data mining the patent document group and the second patent document group, and a second basic term group is acquired from the second term group by referring to the basic term database 60. Furthermore, second feature amount data is created based on the second term group and the second basic term group, and second training data is created from the second feature amount data by referring to the status concept database 61. Then, a second trained model is created by executing training of a prediction model using this second training data, and a second status concept data group is created using the second trained model and the second feature amount data. Thereby, the second status concept data group can be created so as to include terms similar to the status concept data group in at least one of concepts, properties, and classifications.

[0285] In the third embodiment, the data processing device is an example of a personal computer type data processing device 1B, but the data processing device may be configured with multiple personal computers or one or more servers (in this case, the server also includes a cloud server), or may be configured by combining a personal computer and a server.

[0286] In addition, the third embodiment is an example in which a display 1a is used as an output device, but the output device of the present invention is not limited to this, and may be any device capable of outputting a status concept data group from a data processing device. For example, if the data processing device is a cloud server, a terminal device (such as a personal computer or a printer) that receives and outputs the status concept data group output from the cloud server via a communication network may be used as the output device.

[0287] Furthermore, while the third embodiment is an example in which a set of patent documents in which a term group is included in a claim is acquired as a set of patent documents, instead, a set of patent documents in which a term group is included in any of the entire sentences, including the claims, in a patent document may be acquired as a set of patent documents. In that case, the feature data and the learning data may be configured so as not to include claim data.

[0288] Furthermore, the third embodiment is an example in which the status concept data group is two-dimensional image data as shown in Figure 34, but the status concept data group may be created as three-dimensional image data similar to that shown in Figure 22 described above, or may be created as tabular data.

[0289] Furthermore, when creating the status concept data group, a name matching process may be performed for each term by referring to a name matching dictionary (not shown), and then the status concept data group may be created.

[0290] In addition, in the third embodiment, a term representing a raw material of a predetermined substance is used as a basic term, but the basic term of the present invention is not limited to this and may be any term representing a raw material, part, or material of a predetermined substance. For example, a metal material may be used as a basic term.

[0291] Furthermore, the third embodiment is an example in which the basic term product name is used as the basic term product information, but instead of this, the basic term product number may be used as the basic term product information.

[0292] On the other hand, the third embodiment is an example in which a natural language model of Transformer is used as the predetermined learning algorithm, but Random Forest, XGBoost, a support vector machine, a neural network, or the like may also be used as the predetermined learning algorithm.

[0293] In addition, the third embodiment is an example in which a predictive model trained by the model training unit 55 is used as the trained model, but instead of this, a general-purpose trained model (e.g., BERT: Bidirectional Encoder Representations from Transformers) may be used as is without fine-tuning.

[0294] Furthermore, a trained model created by transfer learning may be used as the trained model. For example, the trained data created by the information processing device 1B may be transmitted to a server, and the server may execute learning of a predictive model such as BERT to create a trained model, and the trained model may be transmitted to the information processing device 1B. In that case, in the information processing device 1B, each time feature data is created, the feature data may be input to the trained model without executing learning of a predictive model, thereby creating a status concept data group.

[0295] On the other hand, each time a status concept data group is created in the status concept data group creation unit 56 of the third embodiment, data may be added to the basic term database 60 and the status concept database 61, as described below. That is, when a selected term in the status concept data group is not included in the basic term database 60, the selected term may be configured to be added to the basic term database 60. When configured in this way, the amount of data in the status concept data group can be increased by creating feature amount data using the basic term database 60 with an increased amount of data.

[0296] Also, data may be created to which a label indicating the relationship of a superordinate concept or subordinate concept between selected terms in the status concept data group is added, and if that data is not included in the status concept database 61, it may be configured to add it to the status concept database 61 as shown by the two-dot chain line in Fig. 28. When configured in this way, the learning effect of the prediction model using the learning data can be further improved by creating learning data using the status concept database with an increased amount of data.

[0297] In addition, in the third embodiment, in the information processing device 1B, learning data is created and prediction model learning (additional learning) is performed every time feature data is created, but instead, after prediction model learning has been performed M (M is an integer equal to or greater than 1) times, when feature data is created, the feature data is input to the trained model without performing prediction model learning, thereby creating a status concept data group. For example, the second status concept data may be created by inputting the above-mentioned second feature data to a trained model that has been trained once with a predetermined keyword.

[0298] Furthermore, the third embodiment is an example in which the learning data is created by the learning data creation unit 54, but instead of this, a configuration may be adopted in which the user creates the learning data.

[0299] In addition, in the third embodiment, "silicone oil" is used as the predetermined keyword, but the predetermined keyword of the present invention is not limited to this, and various keywords can be used. For example, rose extract may be used as the predetermined keyword.

[0300] Furthermore, although the third embodiment is an example in which "phenyl trimethicone" is used as the second predetermined keyword, the second predetermined keyword of the present invention is not limited to this, and may be any keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. For example, when "surfactant" is the predetermined keyword, "viscosity modifier" may be used as the second predetermined keyword.

[0301] Furthermore, the third embodiment is an example in which a group of patent documents is acquired from a patent document database 59 stored in the data processing device 1B. However, for example, if the patent document database is stored in a processing device such as a server other than the data processing device 1B, the group of patent documents may be configured to be acquired from the patent document database in the processing device by communication between the data processing device 1B and the processing device.

[0302] Furthermore, although the third embodiment is an example in which a group of basic terms is obtained from a basic term database 60 stored in the data processing device 1B, for example, if the basic term database is stored in a processing device such as a server other than the data processing device 1B, the group of basic terms may be configured to be obtained from the basic term database in the processing device through communication between the data processing device 1B and the processing device.

[0303] On the other hand, the third embodiment is an example in which learning data is created using the status concept database 61 stored in the data processing device 1B, but for example, if the status concept database is stored in a processing device such as a server other than the data processing device 1B, the learning data may be configured to be created using the status concept database in the processing device through communication between the data processing device 1B and the processing device.

[0304] On the other hand, the third embodiment is an example using the status concept database 61 shown in Fig. 33, but the status concept database of the present invention is not limited to this, and may be any database that represents the status concept relationship between selected term pairs. For example, in the status concept database, the symbol ">" may be used to indicate that selected term 1 is a superordinate concept of selected term 2, the symbol ">" may be used to indicate that selected term 1 is a subordinate concept of term 2, and the symbol ">" may be used to indicate that selected term 1 and term 2 are not in a superordinate or subordinate concept relationship ....

[0305] Also, the status concept database 61 may be configured so that only the values ​​"1" and "0" representing the relationship between a superordinate concept and a subordinate concept are set as labels representing the relationship between selected term 1 and selected term 2. In that case, the learning data creation unit 54 may be configured so that when the selected term pair of the feature data is not included in the status concept database 61, the value "2" is set as the label representing the relationship between selected term 1 and selected term 2 in the selected term pair of the feature data.

[0306] Furthermore, in the status concept database 61, in addition to the values ​​"1," "0," and "2" respectively representing the above-mentioned superordinate concept relationship, subordinate concept relationship, and no superordinate or subordinate concept relationship, the value "3" may be set as a label representing a coordinate concept as a label representing the relationship between selected term 1 and selected term 2. In that case, the learning data creation unit 34 may be configured to set values ​​from "0" to "3" as labels for selected term pairs in the learning data. [Explanation of symbols]

[0307] 1 Data Processing Device 1a Display (output device) 11 Patent document acquisition department 12 Terminology acquisition unit 13 Feature data creation section 14. Learning Data Creation Department 15 Model learning unit (model memory unit) 16. Status Concept Data Group Creation Department 17 Output section 18 Average value calculation section 20 Patent Document Database 21 Status Concept Database 1A Data Processing Unit 30 Non-Patent Document Collection Section 31 Terminology acquisition unit 32 Patent document acquisition department 33 Feature Data Creation Department 34 Learning Data Creation Department 35 Model Learning Department 36 Status Concept Data Group Creation Department 37 Output section 39 Non-Patent Document Databases 40 Patent Document Database 41 Status Concept Database 1B Data processing device 50 Patent document acquisition department 51 Terminology acquisition unit 52 Basic vocabulary acquisition unit 53 Feature Data Creation Department 54 Learning Data Creation Department 55 Model Learning Department 56 Status Concept Data Group Creation Department 57 Output section 59 Patent Document Database 60 Basic Terminology Database 61 Status Concept Database

Claims

1. a trained model storage unit that stores a trained model, which is a classification model trained by a predetermined learning algorithm using training data including a set of A-th term pairs, which are pairs of terms included in an A-th patent document group, which is a set of patent documents, and labels that represent status concept relationships, including relationships between superordinate and subordinate concepts, between the A-th term pairs; a status concept data group creation unit that creates a status concept data group that represents the relationship between the superordinate concept and the subordinate concept between the B-term pairs of the feature data by using an output of the trained model when feature data including a set of B-term pairs, which are pairs of terms included in a B-th patent document group, which is a set of patent documents, is input to the trained model; an output unit that executes an output process for outputting the status concept data group to an output device; A data processing device comprising:

2. 2. The data processing device according to claim 1, a patent document group acquisition unit that searches a database of the patent documents based on a predetermined keyword to acquire the B patent document group including the predetermined keyword; a term group acquiring unit that acquires a term group that is the terms in the B-th patent document group in descending order of frequency of appearance up to a predetermined rank by performing data mining on the B-th patent document group; a feature data creation unit that creates the feature data from the term group; The data processing device further comprising:

3. 3. The data processing device according to claim 2, The data processing apparatus, wherein the feature amount data creation unit creates the feature amount data so as to further include claim data of the patent document B in which the term pair B is included.

4. 3. The data processing device according to claim 2, The data processing device further comprises a learning data creation unit that creates the learning data, which is data in which the label is added to the feature data, by searching a status concept database that defines the status concept relationship based on the B term pair of the feature data.

5. 5. The data processing device according to claim 4, The data processing device is characterized in that the learning data creation unit creates the learning data as data in which, in addition to the label, data of a claim of the B patent document containing the term is further added to the feature data.

6. 5. The data processing device according to claim 4, The data processing apparatus is characterized in that the status concept relationship between the B-term pair included in the status concept data group is added to the status concept database.

7. 5. The data processing device according to claim 4, In the learning data, the label defining the superordinate concept is set to a first predetermined value, and the label defining the subordinate concept is set to a second predetermined value, An average value calculation unit calculates an average value of a label representing the status concept relationship between the B-term pair in an output of the trained model when the feature data is input to the trained model, The data processing apparatus, wherein the output unit executes a data output process for outputting data of the B term pair, the average value of the label of which is within a predetermined range, to the output device.

8. 8. The data processing device according to claim 1, The data processing device according to claim 1, wherein the status concept data group creating section creates the status concept data group in a state in which the patent document number is linked to each term of the B term pair.

9. 8. The data processing device according to claim 1, A data processing device characterized in that the status concept data group creation unit creates the status concept data group as tree-shaped graphic data in which the terms of the B term pair are arranged in accordance with the relationship between the superordinate concept and the subordinate concept.

10. 3. The data processing device according to claim 2, The patent document group acquisition unit further acquires a second patent document group, which is a set of patent documents including a second predetermined keyword, by searching the database of patent documents based on a second predetermined keyword that is similar to the second predetermined keyword in at least one of concept, property, and classification in addition to the B patent document group; the term group acquisition unit performs data mining on the B patent document group and the second patent document group to further acquire a second term group, which is terms in the B patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance; the feature data creation unit further creates second feature data including a set of second term pairs, which are pairs of the terms included in the second term group, the terms being pairs of the terms included in the B patent document group and the second patent document group; The data processing device is characterized in that the status concept data group creation unit creates a second status concept data group representing the relationship between the superordinate concept and the subordinate concept of the second term pair of the second feature data using the output of the prediction model when the second feature data is input into the prediction model.

11. a learned model storage step of storing a learned model, which is a classification model learned by a predetermined learning algorithm using learning data including a set of A-th term pairs, which are pairs of terms included in an A-th patent document group, which is a set of patent documents, and labels representing status concept relationships including relations between superordinate and subordinate concepts between the A-th term pairs; a status concept data group creation step of creating a status concept data group representing the relationship between the superordinate concept and the subordinate concept between the B-term pairs of the feature data by using an output of the trained model when feature data including a set of B-term pairs, which are pairs of terms included in a B-th patent document group, which is a set of patent documents, is input to the trained model; an output step of executing an output process for outputting the status concept data group to an output device; 4. A data processing method comprising:

12. 12. The data processing method according to claim 11, a patent document group acquisition step of searching a database of the patent documents based on a predetermined keyword to acquire the patent document group B including the predetermined keyword; a term group acquisition step of acquiring a term group that is the terms in the B-th patent document group in descending order of frequency of appearance up to a predetermined rank by performing data mining on the B-th patent document group; a feature data creation step of creating the feature data from the term group; a data processing method further performed by the data processing device.

13. 13. The data processing method according to claim 12, A data processing method, characterized in that in the feature data creation step, the feature data is created so as to further include data on claims of the patent documents in which the B-term pair is included.

14. 13. The data processing method according to claim 12, a learning data creation step of creating learning data in which the label is added to the feature data by searching a status concept database that defines the status concept relationship based on the B-term pair of the feature data; a data processing method further performed by the data processing device.

15. 15. The data processing method according to claim 14, A data processing method characterized in that, in the training data creation step, the training data is created as data in which, in addition to the label, data of the claims of the B patent document containing the term is further added to the feature data.

16. 15. The data processing method according to claim 14, A data processing method, comprising the steps of: adding the status concept relationship between the B-term pair contained in the status concept data group to the status concept database.

17. 15. The data processing method according to claim 14, In the learning data, the label defining the superordinate concept is set to a first predetermined value, and the label defining the subordinate concept is set to a second predetermined value, The data processing device further executes an average value calculation step of calculating an average value of a label representing the status concept relationship between the B-term pair in an output of the trained model when the feature data is input to the trained model, A data processing method, wherein in the output step, a data output process is executed for outputting data of the B term pair, the average value of the label of which is within a predetermined range, to the output device.

18. A data processing method according to any one of claims 11 to 17, A data processing method characterized in that in the status concept data group creation step, the status concept data group is created with the patent document number linked to each term of the B term pair.

19. A data processing method according to any one of claims 11 to 17, A data processing method characterized in that in the status concept data group creation step, the status concept data group is created as tree-shaped graphical data in which the terms of the B term pair are arranged in accordance with the relationship between the superordinate concept and the subordinate concept.

20. 13. The data processing method according to claim 12, In the patent document group acquisition step, in addition to the B patent document group, a second patent document group is further acquired, which is a set of patent documents including a second predetermined keyword, by searching the database of the patent documents based on a second predetermined keyword that is similar to the second predetermined keyword in at least one of concept, property, and classification. In the term group acquisition step, a second term group is further acquired by performing data mining on the B patent document group and the second patent document group, the second term group being terms in the B patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance; In the feature data creation step, second feature data is further created, the second feature data including a set of second term pairs, which are pairs of the terms included in the second term group, the terms being pairs of the terms included in the B patent document group and the second patent document group; A data processing method characterized in that in the status concept data group creation step, a second status concept data group representing the relationship between the superordinate concepts and the subordinate concepts between the term pairs of the second feature data is created using the output of the prediction model when the second feature data is input into the prediction model.

21. a non-patent document group acquisition unit that searches a database of non-patent documents based on a predetermined keyword to acquire a non-patent document group that is a set of non-patent documents containing the predetermined keyword; a term group acquiring unit that acquires a term group, which is terms in the non-patent document group, in order of frequency of appearance up to a predetermined rank from the terms in the non-patent document group by performing data mining on the non-patent document group; a patent document group acquisition unit that searches a database of patent documents based on the term group to acquire a patent document group that is a set of patent documents including the term group; a feature data creation unit that creates feature data including a set of term pairs that are pairs of the terms included in the term group and that are included in the patent documents; a status concept data group creation unit that creates a status concept data group that represents a relationship between superordinate and subordinate concepts between the term pairs of the feature amount data by using an output of a prediction model that is a classification model when the feature amount data is input into the prediction model; an output unit that executes an output process for outputting the status concept data group to an output device; Equipped with The data processing device according to claim 1, wherein the prediction model is configured to output label values ​​that represent the relationship between the superordinate concept and the subordinate concept between the term pairs when the feature data is input.

22. 22. A data processing apparatus according to claim 21, a learning data creation unit that searches a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between the term pairs based on the term pairs of the feature amount data, and creates learning data in which a label that represents the status concept relationship is added to the feature amount data; a model learning unit that uses the learning data to learn the prediction model according to a predetermined learning algorithm; The data processing device further comprising:

23. 23. A data processing apparatus according to claim 21 or 22, A data processing device characterized in that the status concept data group creation unit creates the status concept data group as tree-shaped graphic data in which the terms of the term pairs are arranged in accordance with the relationship between the superordinate concept and the subordinate concept.

24. 23. A data processing apparatus according to claim 21 or 22, The data processing device is characterized in that the status concept data group creation unit creates the status concept data group in a state in which at least one of the patent document number and the data representing the non-patent document is linked to each term of the term pair.

25. 23. A data processing apparatus according to claim 21 or 22, The non-patent document group acquisition unit further acquires a second non-patent document group, which is a set of non-patent documents including a second predetermined keyword, by searching the database of non-patent documents based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification, in addition to the non-patent document group; the term group acquisition unit performs data mining on the non-patent document group and the second non-patent document group to further acquire a second term group, which is terms in the non-patent document group and the second non-patent document group that appear in a predetermined order in descending order of frequency of appearance; the patent document group acquisition unit searches the database of patent documents based on the second term group to acquire a second patent document group, which is a set of patent documents including the second term group; the feature data creation unit further creates second feature data including a set of second term pairs, which are pairs of the terms included in the second term group and are included in the second patent document group; The data processing device is characterized in that the status concept data group creation unit creates a second status concept data group representing the relationship between the superordinate concept and the subordinate concept between the second term pair of the second feature data using the output of the prediction model when the second feature data is input into the prediction model.

26. a non-patent document group acquisition step of searching a database of non-patent documents based on a predetermined keyword to acquire a non-patent document group, which is a set of non-patent documents including the predetermined keyword; a term group acquisition step of acquiring a term group, which is terms in the non-patent literature group, in order of frequency of appearance up to a predetermined rank from the terms in the non-patent literature group by data mining the non-patent literature group; a patent document group acquisition step of searching a database of patent documents based on the term group to acquire a patent document group that is a set of patent documents including the term group; a feature data creation step of creating feature data including a set of term pairs that are pairs of the terms included in the term group, the terms being included in the patent document group; a status concept data group creation step of creating a status concept data group representing a relationship between superordinate and subordinate concepts between the term pairs of the feature amount data by using an output of a prediction model, which is a classification model, when the feature amount data is input into the prediction model; an output step of executing an output process for outputting the status concept data group to an output device; is executed by a data processing device, The data processing method is characterized in that the predictive model is configured to output label values ​​that represent the relationship between the superordinate concept and the subordinate concept between the term pairs when the feature data is input.

27. 27. The data processing method according to claim 26, a learning data creation step of searching a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between the term pairs based on the term pairs of the feature amount data, thereby creating learning data in which a label representing the status concept relationship is added to the feature amount data; a model learning step of learning the prediction model by a predetermined learning algorithm using the learning data; a data processing method further performed by the data processing device.

28. 28. A data processing method according to claim 26 or 27, A data processing method characterized in that in the status concept data group creation step, the status concept data group is created as tree-shaped graphical data in which the terms of the term pairs are arranged in accordance with the relationship between the superordinate concept and the subordinate concept.

29. 28. A data processing method according to claim 26 or 27, A data processing method characterized in that in the status concept data group creation step, the status concept data group is created in a state in which at least one of the patent document number and the data representing the non-patent document is linked to each term of the term pair.

30. 28. A data processing method according to claim 26 or 27, In the non-patent document group acquisition step, in addition to the non-patent document group, a second non-patent document group, which is a set of non-patent documents including a second predetermined keyword, is further acquired by searching the non-patent document database based on a second predetermined keyword that is similar to the predetermined keyword in at least one of concept, property, and classification. In the term group acquisition step, a second term group is further acquired, which is terms in the non-patent document group and the second non-patent document group that appear in a predetermined order in descending order of frequency of appearance from among the terms in the non-patent document group and the second non-patent document group, by performing data mining on the non-patent document group and the second non-patent document group; In the patent document group acquisition step, a second patent document group, which is a set of patent documents including the second term group, is acquired by searching the database of patent documents based on the second term group; In the feature data creation step, second feature data is further created, the second feature data including a set of second term pairs, which are pairs of the terms included in the second term group and are included in the second patent document group; A data processing method characterized in that in the status concept data group creation step, a second status concept data group representing the relationship between the superordinate concepts and the subordinate concepts between the term pairs of the second feature data is created using the output of the prediction model when the second feature data is input into the prediction model.

31. a patent document group acquisition unit that searches a database of patent documents based on a predetermined keyword that represents a predetermined substance to acquire a patent document group that is a collection of patent documents that include the predetermined keyword; a term group acquiring unit that acquires a term group, which is terms in the group of patent documents, in order of frequency of appearance up to a predetermined rank from the terms in the group of patent documents by data mining the group of patent documents; a basic term group acquiring unit that acquires a basic term group, which is a set of the basic terms, from the term group by searching a basic term database that defines basic terms that represent any of a raw material, a part, and a material quality of the predetermined substance based on the predetermined substance; a feature data creation unit that creates feature data including the basic term included in the basic term group and a set of selected term pairs that are pairs of the term included in the term group and one of the basic terms included in the basic term group; a status concept data group creation unit that creates a status concept data group representing a relationship between superordinate and subordinate concepts between the selected term pairs of the feature amount data by using an output of a prediction model, which is a classification model, when the feature amount data is input into the prediction model; an output unit that executes an output process for outputting the status concept data group to an output device; Equipped with The data processing device according to claim 1, wherein the prediction model is configured to output label values ​​that represent the relationship between the superordinate concept and the subordinate concept between the selected term pairs when the feature data is input.

32. 32. A data processing apparatus according to claim 31, a learning data creation unit that searches a status concept database that defines a status concept relationship including a relationship between a superordinate concept and a subordinate concept between the selected term pairs based on the selected term pairs of the feature amount data, and creates learning data in which a label that represents the status concept relationship is added to the feature amount data; a model learning unit that uses the learning data to learn the prediction model according to a predetermined learning algorithm; The data processing device further comprising:

33. 33. A data processing apparatus according to claim 31 or 32, A data processing system characterized in that the status concept data group creation unit creates the status concept data group as tree-shaped graphical data in which the selected terms of the selected term pairs are arranged in accordance with the relationship between the superordinate concept and the subordinate concept.

34. 33. A data processing apparatus according to claim 31 or 32, A data processing system characterized in that the status concept data group creation unit further uses a product database that defines the relationship between the basic terms of the selected term pairs in the status concept data group and the product information of the basic terms to create the status concept data group with the product information linked to the basic terms.

35. 33. A data processing apparatus according to claim 31 or 32, The patent document group acquisition unit further acquires a second patent document group, which is a set of patent documents including a second predetermined keyword, by searching the database of patent documents based on a second predetermined keyword that represents a second substance having at least one of properties and classification similar to the predetermined substance, in addition to the patent document group; the term group acquisition unit performs data mining on the patent document group and the second patent document group to further acquire a second term group, which is terms in the patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance; the basic term group acquisition unit further acquires a second basic term group, which is a set of the basic terms of the second substance, from the second term group by searching the basic term database based on the second substance; the feature data creation unit further creates second feature data including a set of second selected term pairs each of which is a pair of the basic term included in the second basic term group and one of the second term included in the second term group and the basic term included in the second basic term group; The data processing system is characterized in that the status concept data group creation unit further creates the status concept data group representing the relationship between the superordinate concept and the subordinate concept between the second selected term pair of the second feature data using the output of the prediction model when the second feature data is input into the prediction model.

36. a patent document group acquisition step of searching a database of patent documents based on a predetermined keyword representing a predetermined substance to acquire a patent document group that is a collection of patent documents including the predetermined keyword; a term group acquisition step of acquiring a term group, which is terms in the patent document group, in order of frequency of appearance up to a predetermined rank, from among the terms in the patent document group; a basic term group acquisition step of acquiring a basic term group, which is a set of basic terms, from the term group by searching a basic term database that defines basic terms that represent any of raw materials, parts, and materials of the predetermined substance based on the predetermined substance; a feature data creating step of creating feature data including the basic term included in the basic term group and a set of selected term pairs each being a pair of the term included in the term group and one of the basic terms included in the basic term group; a status concept data group creation step of creating a status concept data group representing a relationship between superordinate and subordinate concepts between the selected term pairs of the feature amount data by using an output of a prediction model, which is a classification model, when the feature amount data is input into the prediction model; an output step of executing an output process for outputting the status concept data group to an output device; is executed by a data processing device, The data processing method is characterized in that the prediction model is configured to output label values ​​that represent the relationship between the superordinate concept and the subordinate concept between the selected term pairs when the feature data is input.

37. 37. The data processing method according to claim 36, a learning data creation step of searching a status concept database that defines status concept relationships including relationships between superordinate and subordinate concepts between the selected term pairs based on the selected term pairs of the feature amount data, thereby creating learning data in which a label representing the status concept relationship is added to the feature amount data; a model learning step of learning the prediction model by a predetermined learning algorithm using the learning data; a data processing method further performed by the data processing device.

38. 38. A data processing method according to claim 36 or 37, A data processing method characterized in that in the status concept data group creation step, the status concept data group is created as tree-shaped graphical data in which the selected terms of the selected term pairs are arranged in accordance with the relationship between the superordinate concept and the subordinate concept.

39. 38. A data processing method according to claim 36 or 37, A data processing method characterized in that in the status concept data group creation step, the status concept data group is created with the product information linked to the basic terms, further using a product database that defines the relationship between the basic terms of the selected term pairs in the status concept data group and the product information of the basic terms.

40. 38. A data processing method according to claim 36 or 37, In the patent document group acquisition step, in addition to the patent document group, a second patent document group is further acquired, which is a set of patent documents including a second predetermined keyword, by searching the patent document database based on a second predetermined keyword that represents a second substance having at least one of properties and classification similar to the predetermined substance; In the term group acquisition step, a second term group is further acquired by performing data mining on the patent document group and the second patent document group, the second term group being terms in the patent document group and the second patent document group that appear in a predetermined order in descending order of frequency of appearance; In the basic term group acquisition step, a second basic term group, which is a set of the basic terms of the second substance, is further acquired from the second term group by searching the basic term database based on the second substance; In the feature data creating step, second feature data is further created, the second feature data including a set of second selected term pairs each being a pair of the basic term included in the second basic term group and one of the second term included in the second term group and the basic term included in the second basic term group; A data processing method characterized in that in the status concept data group creation step, a status concept data group representing the relationship between the superordinate concept and the subordinate concept between the second selected term pair of the second feature data is further created using the output of the prediction model when the second feature data is input into the prediction model.

Images