JP2018200621A - Patent requirement propriety prediction device and patent requirement propriety prediction program - Google Patents

Abstract

To effectively reduce a preparation burden of an application document by predicting propriety of patent requirements about contents adapted to examination practice for an invention to be surveyed.SOLUTION: A patent requirement propriety prediction server comprises patent requirement propriety prediction processing means which has a novelty prediction processing part retrieving public bulletin data using a retrieval document and generating novelty prediction data according to the retrieval result, and a progress prediction processing part generating progress prediction data. The progress prediction processing part comprises a main reference invention retrieval part retrieving a main reference invention, and a document classification part classifying document vectors. The main reference invention retrieval part is configured to set most similar hits literature as main reference literature in which the main reference invention is disclosed, and the document classification part is configured to classify input invention motion vectors into one of whether it conforms to the requirement of the progress and output requirement propriety document vectors according to the classification result by machine learning.SELECTED DRAWING: Figure 4

Description

  The present invention relates to a patent requirement suitability prediction apparatus and a patent requirement suitability prediction program for predicting whether or not a search subject invention that is a subject of prior art search conforms to patent requirements such as novelty and inventive step.

  Conventionally, prediction is performed in various scenes such as prediction of power demand and stock price, prediction of purchase of goods, and prediction of future price of real estate, and many apparatuses and methods have been proposed. A technology for predicting patentability (also referred to as patentability or patentability) relating to an applied invention has also been proposed with respect to a right-giving operation for applying a patent for an invention (see, for example, Patent Document 1). Patent Document 1 describes the following patentability prediction apparatus. This device obtains a patent application (notified application) that has been notified of examination results from the patent database, detects the amount of information about the claims of the already-notified application, and the number of similar prior applications, and The target regression analysis is executed, and the predicted value of the patentability of the application for which the examination result is not notified is calculated according to the registration prediction formula calculated from these.

  Conventionally, in addition to prediction of patentability, devices and methods for evaluating the patentability and quality of an invention, the value of patent applications and patent rights have also been proposed (for example, Patent Documents 2, 3, 4, and 4). 5 and 6).

JP 2009-238074 A JP2015-207194A JP 2000-181966 A JP 2000-132606 A Japanese Patent Laying-Open No. 2015-187883 JP 2007-108803 A Japanese Patent No. 6123143

  As described above, according to the prior art, it is possible to predict the patentability and evaluate the patent right for a patent application.

  However, in the above-described conventional technology, for example, the patentability prediction apparatus described in Patent Document 1, the patentability prediction is calculated based on information such as the amount of information about the claims of a previously notified application and the number of similar prior applications. This is done according to the registration preview formula. This prediction was made based on a statistical correlation between the breadth of claims, the density of technical fields, and patentability, and was not based on patent law or patent / utility model examination standards. Therefore, the patentability prediction apparatus described in Patent Document 1 has a problem that there is a high possibility that a prediction result in accordance with the patent practice cannot be obtained.

  By the way, when applying for a patent, it is essential to conduct a prior art search on the invention to be applied. Conventionally, full-text search using a keyword and concept search using a natural sentence are known as prior art search techniques, and a plurality of search services utilizing these are provided.

  However, in the prior art, only prior art documents related to the search subject invention that is the subject of the prior art search are picked up, and the prior art disclosed in the prior art literature is the patent requirement of the search subject invention. It was not clarified whether it was a technology that hinders the preparation.

  Therefore, in order to clarify the difference between the invention to be filed and the prior art found in the prior art search, it has been difficult to reduce the time and effort for that purpose by the conventional prior art search.

  On the other hand, the examination of patent requirements is conducted by JPO examiners in accordance with the Patent Law and Patent / Utility Model Examination Standard (Patent Act Article 47). It is essential to understand patent law and patent / utility model examination standards. In particular, in order to prepare application documents so as to satisfy the requirements (inventive step) of Article 29, Paragraph 2 of the Patent Law, A sufficient understanding of patent / utility model examination standards and more experienced experience and skills are required. Therefore, it is very time consuming and time consuming to prepare the application documents so as to satisfy the patent requirements.

  In this regard, since the patentability of the invention is predicted by the prior art described in Patent Document 1, it may be possible to reduce the burden required for preparing the application documents by utilizing this.

  However, in the prior art described in Patent Document 1, there is a possibility that the prediction result may not be suitable for examination practice. Therefore, even if such a prediction result is used, there is a possibility that the reduction of the preparation burden of the application documents may not be effective. is there. Conventionally, there is an idea of making a prediction suitable for examination practice (see Patent Document 7). However, in this conventional technique, the document with the highest similarity is regarded as the document disclosing the main cited invention, and this is not possible. Since the property (for example, the case where the document with the next highest similarity becomes the main cited invention) is not considered, there is a possibility that the accuracy of prediction cannot be improved.

  Therefore, in order to reduce the preparatory burden on application documents that have patent requirements, the prediction of propriety of patent requirements is made with respect to the subject invention to be examined in conformity with examination practice, and the accuracy of the prediction is increased. It was hoped that

  The present invention has been made to solve the above-mentioned problems, and predictions regarding the suitability of patent requirements with contents that conform to examination practice are made for the subject invention to be investigated so that the reduction of the preparation burden on application documents is effective. In addition, an object of the present invention is to provide a patent requirement suitability prediction apparatus and a patent requirement suitability prediction program in which the accuracy of prediction is increased.

  In order to solve the above-mentioned problems, the present invention searches for published gazette data, which is electronic data of a published patent gazette, using a search document used for searching a search target invention that is a subject of prior art search, and the search result According to the novelty prediction processing unit for generating novelty prediction data indicating whether or not the requirement for novelty of the subject invention is appropriate, and the inventive step prediction for generating the inventive step prediction data indicating whether or not the requirement for inventive step is appropriate A patent requirement propriety prediction processing means having a processing unit and a prediction result file generating unit that generates a prediction result file indicating the patent requirement suitability of the invention to be investigated using the novelty prediction data and the inventive step prediction data; The inventive step predictive processing unit includes a main citation invention search unit that searches for a main citation invention closest to the search target invention among the prior art inventions specified by the publication gazette data, and a document vector The main citation invention search unit uses the search document to generate main search document data for searching the main citation invention, and uses the main search document data as a target for publication gazette data. The result of the concept search was to extract a plurality of similar documents including the most similar document with the highest similarity in descending order of similarity, and the full-text search term generated from the search document was used. If there is a similar hit document that matches multiple similar documents in the hit document that was hit in the full-text search for the public gazette data, the most similar hit document with the highest similarity is selected from the similar hit documents. The document classification unit uses the inventive movement vector that is input by machine learning using a plurality of training data including a learning document vector and a teacher vector as an inventive step. It is constructed so as to classify whether or not it conforms to the requirement and output a requirement conformity document vector according to the classification result. It is characterized by a patent requirement suitability prediction device that is a vector corresponding to a difference from a citation candidate vector corresponding to a similar hit document.

  Further, the present invention searches the public gazette data, which is the electronic data of the published patent gazette, using the search document used for the search of the search target invention that is the subject of the prior art search, and searches according to the search result. A novelty prediction processing unit for generating novelty prediction data indicating whether or not the requirements for novelty of the subject invention are appropriate, an inventive step prediction processing unit for generating inventiveness prediction data indicating whether or not the requirements for inventive step are appropriate, and A patent result propriety prediction processing unit having a prediction result file generation unit for generating a prediction result file indicating the propriety of the patent requirement of the invention to be searched using the novelty prediction data and the inventive step prediction data, and the inventive step prediction processing unit Among the prior art inventions specified by the publication data, the main citation invention search unit that searches for the main citation invention closest to the search target invention, and the document classification that classifies the document vectors The main citation invention search unit generates main search document data for searching the main citation invention using the search document, and performs a concept search for the public gazette data using the main search document data. As a result of the concept search, a plurality of similar documents including the most similar document with the highest similarity in descending order of similarity are extracted, and each similar document is set as a main cited document in which the main cited invention is disclosed, The document classification unit is a class that has a high possibility that the input invention movement vector does not meet the inventive step requirement by machine learning using a plurality of training data including a learning document vector and a teacher vector. It is constructed so as to classify into one of the classes that meet the requirements of inventive step and output the requirement conformity document vector according to the classification result, and the invention movement vector is investigated according to the search document A plurality of vectors corresponding to differences between the invention vector and each citation candidate vector corresponding to each similar document, and the patent requirement propriety prediction processing means responds to the plurality of requirement conformity document vectors output from the document classification unit. Thus, there is provided a patent requirement conformity prediction apparatus that further includes a non-conformance rate calculating unit that calculates a non-conformance rate indicating a possibility that the invention to be investigated does not conform to the inventive step requirement.

  In the case of the above-mentioned patent requirement conformity prediction apparatus, the document classification unit uses training data as a combination of a learning document vector as a first learning document vector and a teacher vector as a vector indicating that there is a reason for rejection of novelty and inventive step, The combination of the document vector is the second learning document vector and the teacher vector is a vector indicating that there is a reason for refusal of inventive step and there is no reason for refusal of novelty, and the learning document vector is the third learning document vector and the teacher vector is inventive. Is used in combination with a vector indicating that there is no reason for refusal, and the first learning document vector is an application for which a notice of reason for refusal has been issued among already published applications, The reason for refusal of a novelty / inventive step rejection application was pointed out by citing the same document and the reason for refusal without novelty and inventive step was pointed out A first moving document vector corresponding to a difference between a document vector corresponding to a request and a first cited document vector corresponding to a first main cited publication which is a document cited for the reason for refusal The second learning document vector is an application for which a notice of reason for refusal has been issued in the published application, and the reason for refusal of novelty is not pointed out in the notice of reason for refusal, but the reason for refusal of inventive step is pointed out The document vector corresponding to the claim in which the reason for refusal of the inventive stepped rejection application was pointed out, and the second corresponding to the second main cited publication cited as the main publication for the reason for refusal A second moving document vector corresponding to a difference from the cited document vector, and the third learning document vector is a non-rejection application in which a patent decision is issued without a notice of reason for refusal being issued in the published application or Issuing reason for rejection A document vector according to claim 1 of a non-inventive refusal application for which the reason for refusal of refusal was not pointed out in the notice of reasons for refusal, and an application without refusal or an application without inventive step refusal 3. The patent requirement conformity prediction according to claim 2, which is a third moving document vector obtained from a difference from a non-cited document vector corresponding to the most similar document for learning that has the highest similarity as a result of the concept search. It can be a device.

  Furthermore, the present invention is a patent requirement suitability prediction program for causing a computer to function as a patent requirement suitability prediction device, using the search document used for searching the search subject invention that is the subject of prior art search. A novelty prediction processing unit that searches public gazette data, which is electronic data of the published patent gazette, and generates novelty prediction data indicating whether the novelty requirement of the invention to be investigated is appropriate according to the search result; An inventive step prediction processing unit for generating inventive step prediction data indicating whether or not the requirements of the subject invention are necessary, and a prediction result file indicating the suitability of patent requirements of the subject invention using the novelty prediction data and the inventive step prediction data And a prediction result file generation unit that generates a patent requirement conformity prediction processing unit. Among the prior art inventions, the invention includes a main citation invention search unit that searches for a main citation invention that is closest to the search target invention, and a document classification unit that classifies document vectors. To generate main search document data for searching the main cited invention, perform a concept search on the public gazette data using the main search document data, and as a result of the concept search, the similarity in descending order of similarity Extract multiple similar documents including the most similar document with the highest and match multiple similar documents to hit documents hit by full-text search using public text data using full-text search terms generated from search documents If there is a similar hit document, the most similar hit document with the highest similarity among the similar hit documents is set as the main cited document in which the main cited invention is disclosed, and the document classification unit Requirement conformity document according to the classification result by classifying the input invention movement vector as one that meets the inventive step requirement by machine learning using a plurality of training data including teacher and teacher vectors A patent requirement propriety prediction program that is constructed so as to output a vector, and whose invention movement vector is a vector according to a difference between a search target invention vector according to a search document and a citation candidate vector according to the most similar hit document I will provide a.

  As described in detail above, according to the present invention, prediction regarding the propriety of patent requirements with contents suitable for examination practice is performed for the subject invention, so that the preparation burden of application documents can be reduced. Can be obtained, and a patent requirement conformity prediction apparatus and a patent requirement conformity prediction program are obtained.

1 is a system configuration diagram of a patent requirement suitability prediction system including a patent requirement suitability prediction server according to a first embodiment of the present invention. It is a block diagram which mainly shows the internal structure of a patent requirement suitability prediction server. It is a block diagram which mainly shows the internal structure of a user terminal device. It is a functional block diagram which shows the main structures of the patent requirement suitability prediction server which concerns on the 1st Embodiment of this invention. It is a functional block diagram which shows the main structures of the patent requirement suitability prediction process part which concerns on the 1st Embodiment of this invention. It is a functional block diagram which shows the main structures of an input vector production | generation part. (A) is a figure which shows an example of the record layout of a full text search file, (b) is a figure which shows an example of the record layout of a concept search file. It is a figure which shows an example of the record layout of an sst table. It is a figure which shows an example of the record layout of a prediction result memory | storage part. It is a figure which shows an example of the network structure of a machine learning part. It is a flowchart which shows an example of the operation | movement procedure of a patent requirement suitability prediction process. It is a flowchart which shows an example of the operation | movement procedure of a patent requirement suitability prediction routine. It is a flowchart which shows an example of the operation | movement procedure of a novelty and expansion prior application prediction routine. It is a flowchart which shows an example of the operation | movement procedure of an expansion prior application prediction routine. It is a flowchart which shows an example of the operation | movement procedure of an inventive step prediction routine. It is a flowchart which shows an example of main citation invention search processing. It is a flowchart which shows an example of a concept search post-process. It is a flowchart which shows an example of a subcitation invention search process. It is the figure which showed typically the relationship between the patent application used as examination object, and several patent publications. It is a figure which shows an example of a patent requirement suitability prediction list. It is a functional block diagram which shows the main structures of the patent requirement suitability prediction server which concerns on the 2nd Embodiment of this invention. It is a functional block diagram which shows the main structures of the patent requirement suitability prediction process part which concerns on the 2nd Embodiment of this invention. It is a functional block diagram which similarly shows the main structures of an input vector production | generation part. It is a figure which similarly shows an example of the record layout of a prediction result memory | storage part. It is a figure which similarly shows an example of a patent requirement suitability prediction list. It is a functional block diagram which shows the main structures of the patent requirement suitability prediction process part which concerns on a modification. It is a figure which shows an example of the layout of a search screen with a prediction. It is a flowchart which shows an example of the post-concept search process in the patent requirement propriety prediction server which concerns on the 2nd Embodiment of this invention.

  Embodiments of the present invention will be described below with reference to the drawings. In addition, the same code | symbol is used for the same element and the overlapping description is abbreviate | omitted.

First embodiment
(Overall structure of patent requirement conformity prediction system)
First, the configuration of the patent requirement suitability prediction system 1 including the patent requirement suitability prediction server 10 according to the first embodiment of the present invention will be described.

  FIG. 1 is a system configuration diagram of a patent requirement suitability prediction system 1. As shown in FIG. 1, the patent requirement suitability prediction system 1 has a patent requirement suitability prediction server 10 and a plurality of user terminal devices 30 (in FIG. 1, fixed terminal devices 30A, 30B, and 30C) operated by the user. These are connected to each other via the Internet N1.

  The patent requirement suitability prediction server 10 performs data processing according to the patent requirement suitability prediction program. The patent requirement suitability prediction server 10 provides a patent requirement (in this embodiment, novelty (Patent Act Article 29, Paragraph 1, Item 3), an expanded prior application (patent) for an invention subject to prior art search by a user. Predict whether or not it conforms to Article 29-2) and inventive step (Patent Act Article 29 (2))). The user terminal device 30 receives or transmits data with the patent requirement suitability prediction server 10.

  In the patent requirement suitability prediction system 1, the patent requirement suitability prediction server 10 performs machine learning using a plurality of training data based on examination results to determine whether or not the reason for refusal of inventive step violation is likely to be found for the subject invention. Appropriateness of patent requirements is predicted by using an artificial intelligence program. The patent requirement suitability prediction server 10 predicts the suitability of patent requirements for the invention to be investigated with the contents suitable for examination practice, so that the preparation burden of application documents can be effectively reduced.

(Configuration of Patent Requirements Compliance Predicting Server 10)
Next, the configuration of the patent requirement suitability prediction server 10 will be described with reference to FIG. FIG. 2 is a block diagram centering on the internal configuration of the patent requirement suitability prediction server 10. The patent requirement propriety prediction server 10 is a server that is operated by a specialized business operator that provides a service related to the prediction of patent requirement propriety of the subject invention.

  The patent requirement conformity prediction server 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, and a RAM (Random Access Memory) 13. The CPU 11 operates in accordance with a program stored in the ROM 12 and inputs input data obtained by operation input of the keyboard 19 and the mouse 20 via the KBC (Key board controller) 17 via the main bus 19A. Input / output of signals to / from other components is performed to control the operation of the patent requirement conformity prediction server 10 as a whole. The CPU 11 operates as a search document data generation unit 101, an sst table generation unit 102, a patent requirement appropriateness prediction processing unit 103, a target publication extraction unit 104, and a prediction result editing processing unit 105, which will be described later, in accordance with a patent requirement appropriateness prediction program described later. Perform the action. The ROM 12 stores a control program executed by the CPU 11 such as a patent requirement suitability prediction program and permanent data. The RAM 13 stores data and programs used when the CPU 11 operates.

  In addition, the patent requirement compliance prediction server 10 includes a hard disk drive (HDD) 14, a communication control unit 15, a communication processing unit 16, and a video controller 18.

  The hard disk device 14 is formed with various storage units or DBs (database) shown in FIG. 4 and other storage units or DBs necessary for executing the patent requirement suitability prediction program. The hard disk device 14 includes a search document data storage unit 153, an sst table 154, a target publication storage unit 155, a prediction result file storage unit 156, a full-text search file 157, and a concept search file 158 (not shown in FIG. 4). Z). Each storage unit or DB will be described later.

  The communication control unit 15 operates in accordance with an instruction from the CPU 11 to control connection and disconnection of a line for performing communication with the user terminal device 30 and a patent office server (not shown). The communication processing unit 16 operates according to an instruction from the communication control unit 15 and executes data transmission / reception performed via the Internet N1.

  The video controller 18 controls image display on a display device (not shown) to display a screen used for various settings.

  The various storage units or DBs of the hard disk device 14 will be described as follows. The search document data storage unit 153 stores search document data generated by the search document data generation unit 101 (electronic data generated from a search document described later). The public gazette DB 150 stores the electronic data of the published patent gazette as public gazette data. As the public gazette DB 150, FIG. 4 assumes a database of a patent information platform (J-PlatPat) operated by the industrial property information / training hall or a database storing electronic data downloaded therefrom. The latter database can be stored in a server (not shown) or may be stored in the HDD 14.

  Here, the search document is a document used for searching the search target invention, and is composed of text data used for grasping the search target invention. The search document and search document data will be described later in detail. The search document is transmitted, for example, from the user terminal device 30 to the patent requirement suitability prediction server 10 by encrypted communication (for example, encrypted communication using SSL).

  The sst table 154 stores a sub search term (sub search term, a keyword for searching a sub-cited invention) generated from search document data. As shown in FIG. 8, the sst table 154 has a search term area 154a, a complete code area 154b, and a document number area 154c, and one record is generated by these.

The search term area 154a is a portion in which a sub search term is stored, and has a fixed area 154d and a variable area 154e. In the fixed area 154d, words that are provided for all sub-search terms and are used fixedly in the full-text search are set. In the case of the present embodiment, data (name data K1) input from the name area 181a of the search screen with prediction 181 (see FIG. 27), which will be described later, and data (purpose / task data K2) input from the purpose / task area 181c. Is set in the fixed area 154d. In the variable area 154e, words that are variably used in the full text search are set. In the case of this embodiment, a word corresponding to data (solution means data K3) input from the solution area 181d of the search screen 181 with prediction is set in the variable area 154e. Since the solution data K3 often includes a plurality of words which are constituent elements of the invention, they are divided into words ( _n from K3 ₁ to K3 _n , in FIG. 8, K3 ₁ to K3 ₅ All combinations including one or more words are set in the variable area 154e. The variable area 154e has a word count area 154f. In the word number area 154f, data corresponding to the number of words set in the variable area 154e such as “1” and “2” is set.

A complete code is set in the complete code area 154b. The complete code is a code “P0” indicating that the word set in the variable area 154e is composed of all of the words of the solution means data K3, or a common code is stored in the combination record for becoming all. ing. For example, in FIG. 8, the complete code is “P11” because the variable area 154e is a record including K3 _{1 and} a record including K3 ₂ to K3 ₄ and includes the words of both records. All words of the data K3 (in FIG. 8, five types from K3 ₁ to K3 ₅ ) are included. The document number area 154c stores the document number of the document hit by the full-text search using the search term in the search term area 154a.

  The target publication storage unit 155 stores public gazette data (search target publication data) to be searched for the main citation invention search and the sub-citation invention search. The prediction result storage unit 156 stores a prediction result file described later as shown in FIG.

  The search document data generation unit 101 generates search document data from the search document, and stores the search document data in the search document data storage unit 153. The search document is input by the user from the search screen with prediction 181 shown in FIG.

  The search screen with prediction 181 is provided with a name area 181a, a purpose / issue area 181c, and a solution means area 181d, and data input from these forms a search document. In the name area 181a, the name of the invention to be investigated is input. Since the name pull-down bar 181b is provided in the name area 181a, this is selected and the name of the invention is input. In the name pull-down bar 181b, a word that is easy to use as the name of the invention (a word used at the end of the name of the invention, also referred to as a name word) is registered in advance, and the user can select it with the user terminal device 30. As the name word, for example, an apparatus, a method, a system, a device, a circuit, a module, a member, a tool, a material, an element, a program, a board, and others are registered. The purpose / problem area 181c is input with the purpose or problem of the invention to be investigated. Means for solving the object or problem of the invention to be investigated is input to the solution means area 181d.

  The search document data generation unit 101 generates search document data by performing morphological analysis and removing unnecessary words using the input search document. Since unnecessary words are removed, the total number of index words used in the search can be reduced, so that the search efficiency is improved. In general, natural languages can be broadly divided into content words and function words. A content word is a word that is meaningful by itself, such as a verb or a noun. A function word is a word that represents a relationship between words, and corresponds to a particle or an auxiliary verb. Since the part-of-speech of each word is known by performing the morphological analysis, the function word can be removed as an unnecessary word based on the result.

  The sst table generation unit 102 generates the sst table 154 having the above-described configuration using the search document data. The sst table generation unit 102 sets words corresponding to the data input from the name area 181a and the data input from the purpose / task area 181c in the fixed area 154d, and corresponds to the data input from the solution means area 181d. The words are subdivided into words and set in the variable area 154e. The complete code is set in the complete code area 154b with reference to the variable area 154e.

  Next, the patent requirement suitability prediction processing unit 103 will be described. As shown in FIG. 5, the patent requirement suitability prediction processing unit 103 includes a novelty / expansion prior application prediction processing unit 125 and an inventive step prediction processing unit 126. The novelty / expanded prior application prediction processing unit 125 uses the search document data stored in the search document data storage unit 153 as a full-text search term to perform a full-text search on the search target publication data in the target publication storage unit 155. In accordance with the result, the novelty / expansion prior application prediction data Nd is output to the prediction result file generation unit 127, and the document number of the hit document is set in the full-text search file 157. The function and operation procedure of the novelty / expansion earlier application prediction processing unit 125 will be described in detail later.

  The inventive step prediction processing unit 126 includes a cited invention search unit 131, an input vector generation unit 132, and a machine learning unit 133. The cited invention search unit 131 has a main cited invention search unit that performs a main cited invention search and a subcited invention search unit that performs a subcited invention search, which will be described later. The cited invention search unit 131 outputs the inventive step prediction data Vd1 to the prediction result file generation unit 127 and inputs the search document data ied and the concept search data Vd2 according to the results of the main citation invention search and the sub-citation invention search. The data is output to the vector generation unit 132. The concept search data Vd2 includes the most similar hit document (described later) determined based on the comparison result between the result of full text search by the novelty / expanded earlier application prediction processing unit 125 and the result of concept search by the main cited invention search unit. The publication data of the most similar document is included (see FIG. 17). The function and operation procedure of the cited invention search unit 131 will be described in detail later.

  As shown in FIG. 6, the input vector generation unit 132 includes a search target invention vector generation unit 132a, a citation candidate vector generation unit 132b, and a movement vector generation unit 132c.

  The search object invention vector generation unit 132a inputs the search document data ied, extracts the feature word, performs weighting according to each word, and generates a search object invention vector EV according to the search document. The citation candidate vector generation unit 132b inputs the public gazette data of the main cited document (the most similar hit document or the most similar document) included in the concept search data Vd2, extracts the feature word, and performs weighting according to each word. Then, a document vector (citation candidate vector) RfV corresponding to the main cited document is generated. The movement vector generation unit 132c calculates a difference between the search target invention vector EV and the citation candidate vector RfV, and generates an invention movement vector V3 corresponding to the difference between the two document vectors.

  As will be described in detail later, the most similar document is a document that has the highest similarity in the concept search, and the most similar hit document is a document that has a similarity greater than or equal to a certain value and that has been hit by a full-text search. For this reason, it is highly probable that any document is cited as a disclosed document of the main cited invention in the examination of the inventive step of the subject invention. In the latter case, there is a high possibility that it will be grounds for lack of novelty requirements. Guessed. Therefore, if the citation candidate vector RfV is obtained using the most similar hit document or the most similar document as a citation candidate and the difference between the citation candidate vector RfV and the search object invention vector EV is calculated to obtain the invention movement vector V3, the most similar to the survey object invention. A document vector (corresponding to the invention movement vector V3) corresponding to the difference from the invention disclosed in the hit document or the most similar document is generated.

  The machine learning unit 133 is a document classification unit according to the embodiment of the present invention, and the machine movement vector V3 is obtained by machine learning (supervised learning) using the following training data (also referred to as a learning pattern). The class is classified into a class that conforms to the requirement of inventive step and a class that does not conform (class that has no reason for refusal), and is configured to output an output signal (requirement conformity document vector) V4 according to the classification result. . In the embodiment of the present invention, the learning pattern can be an HL pattern described below.

  In the HL pattern, the learning document vector is the first learning document vector, and the teacher vector is a vector indicating that there is a reason for rejection of the inventive step (for example, only the dimension corresponding to the correct class is “1” and the others are “0”). Vector), the learning document vector is the second learning document vector, and the teacher vector is a vector indicating that there is no reason for rejection of the inventive step (for example, only a dimension different from the above is “1” and the others are “0”). Is a combination pattern with "vector".

  The first learning document vector is an application for which the first notice of reasons for refusal (1st action) has been issued as a result of examination by the Patent Office in a published application, and the reason for refusal of inventive step violation ( Claims (at the time of issuance of the notice of reasons for refusal) that pointed out the reasons for refusal of the application (reasonable refusal application) for which the reason for refusal that the requirement of Article 29 (2) of the Patent Act was not satisfied was pointed out Is a first moving document vector corresponding to a difference between a document vector corresponding to the document vector and a document vector (cited document vector) corresponding to the cited document 1 (main cited document cited as the main publication) at that time. .

  The second learning document vector is the result of examination in the published application that the patent action is issued without the first action being issued (the application without refusal) or the first action is issued but the reason for refusal is advanced Document vectors according to claim 1 (at the time of the notice of reasons for refusal) of applications for which no reason for refusal of sex violation has been pointed out (applications without inventive refusal), and those applications without refusal or inventive refusal As a result of the concept search for the application, the second moving document vector corresponding to the difference from the document vector (non-cited document vector) corresponding to the document (the most similar document for learning) that has the highest similarity is used. is there.

  The machine learning unit 133 classifies the inventive movement vector V3 into either a class with or without a reason for refusal of inventive step by repeatedly performing learning with the learning pattern of the HL pattern as described above, and classifies the class into the classified class. A corresponding requirement conformity document vector V4 is output. The former corresponds to the case where the reason for refusal of inventive step violation is likely to be issued and the latter is low when the invention to be investigated is filed.

  The machine learning unit 133 may classify the input invention movement vector V3 into a class having a reason for refusal of inventive step and a class having no reason for refusal of inventive step, and output a requirement conformity document vector V4 according to the classification result. A learning algorithm called a support vector machine (SVM) can be applied to the unit 133. According to the support vector machine (SVM), the decision boundary can be obtained so that the distance (margin) from the decision boundary is maximized.

  In addition, a neural network modeled on the brain neural network can be applied to the information processing of the machine learning unit 133. Neural networks include hierarchical neural networks and interconnected neural networks. For example, a perceptron of a hierarchical neural network can be applied as a learning algorithm of the machine learning unit 133.

  The perceptron is composed of a three-layer hierarchical network called S layer, A layer, and R layer (not shown), and there is only one-way coupling from S layer to A layer and from A layer to R layer. . Given the learning pattern of the HL pattern described above, when the output vector when the first learning document vector or the second learning document vector is input is different from each teacher vector, according to the error. The learning ends when the weight of the connection is corrected and the error between the output vector and the teacher vector becomes a certain value or less.

  However, the perceptron may not stop the algorithm when the learning pattern cannot be linearly separated. Therefore, in order for the machine learning unit 133 to acquire a non-linear decision boundary by learning, there are many application examples in a hierarchical neural network, and a BP (back propagation) network that can learn a non-linear discriminant plane with few misidentifications. Is preferably applied.

  As shown in FIG. 10, the BP network has an input layer and an output layer, and an intermediate layer therebetween. A learning algorithm called an error back propagation algorithm makes it possible to learn the weights of all the connections between units. ing. In the error back-propagation algorithm, the input signal is transmitted to the input layer, the intermediate layer, and the output layer, while the error signal is transmitted in the reverse direction, whereby weight adjustment is performed.

Then, when the learning pattern x _p (x ₀ , x ₁ ... X _n ) is input to the BP network shown in FIG. 10, the j (0 ≦ j ≦ n) -th unit in a certain hierarchy A weighted signal is input from the unit one layer before having a connection with the unit j. Therefore, if the signal from the i-th unit (0 ≦ i ≦ n) of the previous layer is t _ip and the weight is w _ij , the input to the unit j is given by Equation 1, and the output of the unit j is Suppose f is a threshold function and Equation 2 is obtained.

Formula 1

Formula 2

Error D _p for learning pattern x _p is the output of unit k of output layer, since is defined by the sum of squares of the difference between the teacher signal b _kp, so equation 3 below. The error D _p 3GS sum for all learning patterns sought D of formula 4, as its D is minimized, is adjusted connection weights between units is learning in machine learning unit 133 is carried out. In this case, each time an individual learning pattern is input, the weight is adjusted by Expression 5. w _ij is a weight before update, w ′ _ij is a weight after update, and ρ is a learning coefficient. This is called stochastic steepest descent. Note that the sigmoid function shown in Equation 6 is used as the unit input / output function.

Formula 3

Formula 4

Formula 5

Equation 6

(Configuration of the user terminal device 30)
As shown in FIG. 1, the user terminal device 30 includes a connection environment to the Internet N <b> 1 and can communicate with the patent requirement suitability prediction server 10. The user terminal device 30 is assumed to be a stationary (or portable notebook type) personal computer, but may be a tablet-type terminal device.

  As illustrated in FIG. 3, the user terminal device 30 includes a CPU 31, a ROM 32, a RAM 33, a data storage unit 34, and a liquid crystal display unit 35. In addition, the user terminal device 30 includes a voice conversion processing unit 36, a communication control unit 37, a communication processing unit 38a, a wireless communication unit 38b, a speaker 39, and a microphone 40.

  The CPU 31 operates according to a program stored in the ROM 32 and controls operation of the entire user terminal device 30. The ROM 32 stores a program executed by the CPU 31, for example, a communication control program for performing data communication. The RAM 33 stores data necessary for the CPU 31 to execute the program.

  Various data are stored in the data storage unit 34. The liquid crystal display unit 35 includes an LCD (Liquid Crystal Display) and its driving unit, and is an image display unit that displays images of characters, figures, symbols, and the like. The voice conversion processing unit 36 decompresses the voice data and outputs it to the speaker 39, while converting and compressing the analog voice signal input from the microphone 40 into digital voice data and inputs the digital voice data to the communication processing unit 38a. The communication control unit 37 operates in response to an instruction from the CPU 31 and controls connection and disconnection of a line for performing data communication. The communication processing unit 38a operates according to an instruction from the communication control unit 37, and executes transmission / reception of data performed via the Internet N1. The wireless communication unit 38b is a wireless communication unit that performs wireless data transmission / reception under the control of the communication control unit 37. The speaker 39 is an audio output means for outputting audio, and the microphone 40 inputs audio such as the user's conversation content and converts it into an electrical signal.

(Operations of the patent requirement conformity prediction system)
Next, referring to FIGS. 11 to 18 together with FIG. 4, the operation content of the patent requirement suitability prediction process by the patent requirement suitability prediction server 10 will be described.

  Here, FIG. 4 is a functional block diagram showing a main configuration of the patent requirement suitability prediction server 10 for realizing the patent requirement suitability prediction process. In the patent requirement suitability prediction server 10, the CPU 11 accesses the search document data while accessing various files and DBs stored in the publication gazette DB 150, the search document data storage unit 153, the sst table 154, etc. according to the patent requirement suitability prediction program. The generation unit 101, the sst table generation unit 102, the patent requirement suitability prediction processing unit 103, the target publication extraction unit 104, and the prediction result editing processing unit 105 are operated. Thereby, a patent requirement suitability prediction process is executed. The patent requirement suitability prediction program is a program for causing the patent requirement suitability prediction server 10 to function as the search document data generation unit 101, the sst table generation unit 102, the patent requirement suitability prediction processing unit 103, the prediction result editing processing unit 105, and the like. It is.

  When the patent requirement suitability prediction server 10 performs the patent requirement suitability prediction process, the CPU 11 performs an operation according to the flowchart shown in FIG. 11 according to the patent requirement suitability prediction program. FIG. 11 is a flowchart showing an example of the operation procedure of the patent requirement suitability prediction process of the CPU 11 according to the patent requirement suitability prediction program. In FIG. 11, FIG. 12, etc., “S” is an abbreviation for step.

  When starting the operation according to the patent requirement suitability prediction program, the CPU 11 proceeds to step 1 and performs user authentication processing. Here, the user authentication process is performed by confirming the user ID and password input by the user using the user terminal device 30. Next, the CPU 11 proceeds to step 2 to perform a point balance confirmation process. In the point balance confirmation process, the CPU 11 confirms whether or not the user's point balance is equal to or greater than a certain value. If the point balance is insufficient, the patent requirement suitability prediction process is terminated, or a message notifying that the point balance is insufficient is transmitted. Do.

  When the process proceeds to step 3, the CPU 11 operates the search document data generation unit 101 to display the search screen with prediction 181 on the user terminal device 30 until the search document is input using the search screen with prediction 181. stand by. When a search document is input, the CPU 11 proceeds to step 4 to generate search document data from the input search document. In step 5, the CPU 11 generates the search document data as a search document data storage unit 153. Remember me.

  Next, the CPU 11 advances the processing to step 6 to determine whether or not the prediction end condition is satisfied. Here, the CPU 11 proceeds to step 8 when the prediction end condition is satisfied, but proceeds to step 7 otherwise. When proceeding to step 7, the CPU 11 executes a patent requirement suitability prediction routine which will be described later. When proceeding to step 8, the CPU 11 executes a termination process and terminates the patent requirement conformity prediction process for the user. By doing so, the patent requirement suitability prediction process is automatically and continuously executed until the prediction end condition is satisfied.

  When the CPU 11 proceeds to step 7, it executes a patent requirement suitability prediction routine according to the flowchart shown in FIG.

(Patent requirement compliance routine)
When the CPU 11 starts the patent requirement suitability prediction routine, the CPU 11 proceeds to step 11 to perform the operation as the target publication extraction unit 104 to perform publication extraction processing. In this case, the CPU 11 extracts public gazette data on the basis of the machine date (extracts public gazette data whose publication date is earlier than the machine date), and uses the extracted data as search target gazette data. Remember me. Next, the CPU 11 advances the process to step 12 to perform the operation as the sst table generation unit 102 and executes the sst table generation process. The data generated here is stored in the sst table 154.

  Subsequently, the CPU 11 proceeds to step 13 to execute a novelty / expansion prior application prediction routine to be described later, and then proceeds to step 14 to execute the inventive step prediction routine. Thereafter, the CPU 11 proceeds to step 15 where the CPU 11 performs an operation as the prediction result editing processing unit 105, and edits and outputs a prediction result list L1 described later. Thereafter, the point consumption process of step 16 is executed, and the point balance is reduced according to the search document data for which the propriety of patent requirement is predicted. Thereafter, the patent requirement suitability prediction routine ends.

(Novelty / expanded application prediction routine)
When the CPU 11 proceeds to step 13, the CPU 11 operates as the novelty / expansion prior application prediction processing unit 125 described above, and follows the flowcharts shown in FIGS. Run the routine.

  In this case, when starting the novelty / expansion prior application prediction routine, the CPU 11 advances the process to step 21, and uses the search document data stored in the search document data storage unit 153 as the full text search term, A full-text search is performed on 155 search target publication data (this full-text search is also referred to as a main invention full-text search), and the number of hit documents is set in the document MAX.

  Subsequently, the process proceeds to step 22 to determine whether or not there is a document hit in step 21 (whether or not the document MAX is 1 or more). If there is a hit document, the process proceeds to step 23; The novelty / expansion prior application prediction routine ends.

  In step 23, "1" is added to the document counter, and in the following step 24, it is determined whether or not the document counter is equal to or smaller than document MAX. If this is true, the process proceeds to step 25, and so on. If not, the process proceeds to step 28. In step 25, whether or not the application publication date (hit document publication date) of the hit document is smaller than the machine date (in this embodiment, it is treated as the application date (target application date) of the search document data). It is determined whether or not the document publication date <the subject application date is satisfied). If this is satisfied, the process proceeds to step 26; otherwise, the process proceeds to step 27. In step 26, novelty / expansion prior application prediction data Nd is generated so as to include a novelty flag “N1” indicating no novelty. Thereafter, the process returns to step 23 and the same processing as described above is repeated.

  In step 27, an enlargement earlier application prediction routine, which will be described later, is executed. When the process proceeds to step 28, novelty / expanded application prediction data Nd is output. When the document number of the document hit by the CPU 11 is stored in the full-text search file 157 in the subsequent step 29, the novelty / expansion prior application prediction routine is thereafter terminated. In the full text search file 157, as shown in FIG. 7A, a serial number (ref-no) and a document number are stored.

  On the other hand, the CPU 11 executes an enlargement-first application prediction routine according to the flowchart shown in FIG. When the extended earlier application prediction routine starts, the process proceeds to step 41, and whether or not the filing date of the hit document (document filing date) is earlier than the target filing date (reference filing date <subject filing date is satisfied) If this is true, the process proceeds to step 42; otherwise, the enlargement earlier application prediction routine is terminated. In step 42, it is determined whether or not the inventor is the same for the research subject invention and the hit document. If this is not true, the process proceeds to step 43. End the routine. In step 43, it is determined whether or not the applicant is the same for the invention to be searched and the hit document. If this is not the case, the process proceeds to step 44. End the routine. Then, when the CPU 11 proceeds the process to step 44, the CPU 11 includes an enlarged first application flag “F1” indicating that the requirement of the enlarged first application (required in Article 29-2 of the Patent Law) is not satisfied. The novelty / expansion prior application prediction data Nd is generated. Thereafter, the enlargement destination application prediction routine ends.

  When the novelty / expansion prior application prediction routine is completed as described above, the process proceeds from step 13 to step 14 in FIG. 12, and the CPU 11 performs an operation as the inventive step prediction processing unit 126 to execute the inventive step prediction routine. The CPU 11 executes an inventive step prediction routine according to the flowcharts shown in FIGS.

(Inventive step prediction routine)
When the CPU 11 starts the inventive step prediction routine, the CPU 11 advances the process to step 71 shown in FIG. 15 and executes the main cited invention search process. The CPU 11 proceeds to the subsequent step 72 to determine whether or not there is a main citation invention in the main citation invention search process (whether or not a main citation document to be described later is set). The process proceeds to 73, and if there is no main cited invention, the process proceeds to step 76. In subsequent step 73, post-concept search post-processing described later is executed, and it is determined in subsequent step 74 whether or not there is a sub-cited invention (whether or not a sub-cited document described later is set). If there is a secondary citation invention, the process proceeds to step 75. If there is no secondary citation invention, the process proceeds to step 77.

  When the processing proceeds to step 75, the CPU 11 sets “VX” to the search flag (search flag) for the corresponding search document data, and sets “VY” to the search flag (search flag) in step 77. Further, when the CPU 11 proceeds with the process to step 76, the CPU 11 generates the inventive step prediction data Vd1 so as to include the set search flag, and outputs it to the prediction result file generation unit 127. Further, the CPU 11 outputs the search document data ied and the concept search data Vd2 corresponding to the search result to the input vector generation unit 132.

  In the inventive step prediction routine, “VX” or “VY” is set in the search flag when the main cited invention is found. If the main cited invention is found, it is likely that it will be judged that the requirement for inventive step is not satisfied, so whether or not the reason for refusal of violating the inventive step is found depends greatly on whether or not the main cited invention is found. Is done. By including such a search flag, the inventive step prediction data Vd1 indicates the suitability of the inventive step requirement. After step 76 is executed, the inventive step prediction routine ends.

  Then, the CPU 11 executes main citation invention search processing according to the flowchart shown in FIG. The main citation invention search process is a process for searching for the main citation invention closest to the search target invention.

  When the CPU 11 starts the main citation invention search process, the CPU 11 proceeds to step 82, reads the search document data from the search document data storage unit 153, and searches the main search document data (when the main citation invention is searched by concept search). Document data).

  In the following step 83, the CPU 11 performs a concept search for the search target publication data stored in the target publication storage unit 155 using the search process of the main cited invention, that is, the main search document data as the input document. In this conceptual search, a vector (document vector) corresponding to each document is generated by extracting and weighting the main search document data and each document to be searched, and the inner product of each vector is obtained. Thus, the similarity is calculated. In the next step 84, it is determined from the result of the concept search in step 83 whether or not the largest similarity is a certain value or more. If it is more than a certain value, the process proceeds to step 85; The main cited invention search process ends.

  In step 85, a plurality of documents having a similarity equal to or greater than a certain value including the most similar documents are stored in the concept search file 158 together with each similarity. After step 85 is executed, the main cited invention search process ends. The concept search file 158 is a file that stores the document numbers of documents whose similarity found in the concept search is equal to or higher than a certain value together with the respective similarities. As shown in FIG. The document number is stored.

  The post-concept search process is executed according to the flowchart shown in FIG. In the post-concept search processing, the result of the main invention full-text search and the result of the concept search are compared, and according to the result, the document (main cited document) and the sub-cited invention are disclosed. A document (sub-cited document) is determined.

  When the CPU 11 starts the post-concept search processing, the CPU 11 proceeds to step 91 and refers to the full-text search file 157. In the following step 92, it is determined whether or not there is a document hit in the main invention full-text search depending on whether or not the document number is stored in the full-text search file 157. If there is a hit document, the process proceeds to step 93. If not, the process proceeds to step 96. In step 93, the CPU 11 refers to the concept search file 158. In the following step 94, the document number stored in the full-text search file 157 and the document number stored in the concept search file 158 are compared, and a plurality of documents found by the concept search are found in the documents hit in the main invention full-text search. It is determined whether there is a document that matches the similar document (the document that matches in this case is also referred to as a similar hit document). If there is a similar hit document, the process proceeds to step 95; otherwise, the process proceeds to step 98. In step 95, a document having the highest similarity among the similar hit documents (that is, a document having the highest similarity among the similar hit documents and also referred to as “the most similar hit document”) is selected and is selected as the main hit document. After setting the cited document and the sub-cited document, the post-search processing is terminated.

  When the process proceeds to step 98, since there is no matching document in the full-text search file 157 and the concept search file 158, the document having the highest similarity among the documents in the concept search file 158 (the most similar) Document) is selected, set as the main cited document, and the process proceeds to step 99. In the subsequent step 99, the document stored in the full-text search file 157 is set as the sub-cited document, and thereafter the post-concept search process ends.

  On the other hand, when the process proceeds to step 96, the document number is not stored in the full-text search file 157 (the document was not hit in the main invention full-text search in step 21), and is stored in the concept search file 158. The document having the highest similarity (the most similar document) is selected from the documents, set as the main cited document, and the process proceeds to step 97. In step 97, the sub-cited invention search process described later is executed. After the subcitation invention search process is executed, the post concept search process ends.

The subcited invention search process is executed according to the flowchart shown in FIG. When the CPU 11 starts the secondary citation invention search process, the CPU 11 proceeds to step 111, refers to the sst table 154, and uses the secondary search terms set in the search term area 154a as a target for the search target publication data. A search process is executed (this full-text search is also called a sub-invention full-text search). As described above, the sst table 154 stores the sub search terms set to include at least one of n words from K3 ₁ to K3 _N. A sub-invention full-text search process is executed using each sub-search term.

In the following step 112, the document number hit in the sub-invention full-text search in step 111 is set in the document number (ref-no) area 154c of the sst table 154. In the next step 113, it is determined whether or not the sub-invention full-text search process has been completed for all the sub-search terms set in the sst table 154. If completed, the process proceeds to step 114. Otherwise, the process returns to step 111. In step 114, it is determined whether or not a document number is set in a record with the same complete flag (for example, a record in which the complete flag is P11). If it is set, the process proceeds to step 115, and so on. If not, the process proceeds to step 116. In step 115, the corresponding document number is set in the sub-cited document, and the sub-cited invention search process ends. In step 116, the document number is read from the document number area 154c of the record in which the maximum value is set in the word count area 154f in the sub-search terms set in the sst table 154, and is read as a reference document ( The reference document cannot be determined that the sub-cited invention is disclosed, but is set to a document in which relevance to the invention to be investigated is recognized (in the present embodiment, the variable area 154e starts from K3 ₁ at the maximum. Five types of words up to K3 ₅ can be set, but the sub-cited invention search process is executed when the document number is not set in the full-text search file 157, so the maximum value in the word count area 154f is 4 or less. Become). Thereafter, the sub-cited invention search process ends.

  As described above, when the novelty / expansion prior application prediction routine and the inventive step prediction routine are executed, the novelty / expansion prior application prediction data Nd and the inventive step prediction data Vd1 according to the respective results. Is output to the prediction result file generation unit 127. Further, since the requirement suitability document vector V4 is output from the machine learning unit 133, the prediction result file generation unit 127 generates a prediction result file shown in FIG. 9 using these, and stores the prediction result file in the prediction result storage unit 156.

  As shown in FIG. 9, the prediction result file stores data for each item of search document number, search document, search flag, hit document, and machine prediction for each investigation target invention. The machine prediction is data corresponding to the requirement conformity document vector V4 from the machine learning storage unit 133, and the reason for refusal of the inventive step violation that cited the main cited document and / or the sub-cited document found in the inventive step prediction routine. Is high or low (“H” when high, “L” when low).

  Further, the prediction result editing unit 105 reads the prediction result file, edits and outputs the patent requirement suitability prediction list L1 as shown in FIG. 20, and transmits it to the user terminal device 30. In the patent requirement suitability prediction list L1, the search document number, novelty (expansion prior application), and inventive step suitability of the search target invention are shown together with the documents on which the requirements are based (main cited reference and sub cited reference). . “X” is described when it is predicted that the requirement of novelty (expanded prior application) and inventive step will not be met (does not meet), and “A” is recorded when it is predicted that it will meet (satisfy). These are determined by the search flag of the prediction result file.

  “H” and “L” are also written in “X” (main citation and sub-citation) and “Y” (only main citation) when predicted not to meet the inventive step requirement. (In FIG. 20, only when “H” is written together is illustrated) This is in accordance with the requirement adequacy document vector V4 of the machine learning unit 133, and “H” is a violation of the inventive step in the main cited document. When the possibility of refusal is likely to be issued, “L” indicates a low case.

  As described above, in the patent requirement conformity prediction server 10 according to the embodiment of the present invention, search document data is generated using a search document used for search of the search target invention, and the main citation is used using the search document data. We conduct invention searches and sub-cited invention searches. The main citation invention search is common to the search target invention and its skeleton, that is, the process of searching for the main citation invention closest to the search target invention among the prior art inventions using search document data. Patent law, patent / utility model It is performed in accordance with the examination criteria. In the patent requirement suitability prediction server 10, in the main citation invention search (see FIG. 16), by executing the concept search, documents (1 or more) whose similarity is a certain value or more are selected as candidates for the main citation documents. These are recorded in the concept search file 158. If there is a document that matches the document hit in the main invention full-text search, the document with the highest similarity among the matched documents (above) The most similar hit document) is set as the main cited document (step 95 in FIG. 17). The document is selected with a concept search with a similarity equal to or greater than a certain value, and is also a document found by a full-text search of the main invention, so it is highly likely that the document matches the search target invention and may be cited as the main cited invention Is also expensive.

  In addition, if there is no document that matches the result of the main invention full-text search and the result of the concept search, the document selected as a similarity in the concept search with a certain value or more has a relevance to the main cited invention. Among them, the document with the highest similarity (the most similar document) is set as the main cited document (step 98). Furthermore, since it is recognized that the documents found in the main invention full-text search disclose the constituent requirements in common with the constituent requirements (K1, K2, K3) of the subject invention, it will be the main cited reference or not May also be a sub-cited document.

  On the other hand, if no matching document is found in the full-text search of the main invention, this corresponds to a case where a document that discloses the same component requirement as the component requirement (K1, K2, K3) of the subject invention is not found. In this case, the most similar document found by the concept search is set as the main cited document, and then the subcited invention search (FIG. 18) is performed. In that case, if a document is set in a record having the same complete flag, the document is set as a sub-cited document. Documents with records with the same complete flag are cited as at least a sub-cited document because they indicate that the presence of the same component requirement as the component requirement (K1, K2, K3) of the invention to be investigated can be recognized. It is possible that The process of step 99 or the process of searching for a sub-cited document in step 97 is a process of searching for a document that is recognized as including a difference between the subject invention and the main cited invention.

As described above, the patent requirement conformity prediction server 10 according to the embodiment of the present invention executes the main citation invention search process, executes the sub-citation invention search process according to the result, and the inventive step from both results. Therefore, the suitability of the inventive step is judged according to the Patent Law and the Patent / Utility Model Examination Criteria. Therefore, since the prediction regarding the suitability of the patent requirement is made in accordance with the examination practice, the preparation burden of the application documents for the patent application is effectively reduced. In addition, since a document with a similarity equal to or higher than a certain value is selected as a candidate for the main cited document, not only the document with the highest similarity but also the next document can be the main cited document. Prediction including the possibility of being performed is possible, and therefore the accuracy of the prediction is improved.

  On the other hand, as described above, the patent requirement suitability prediction processing unit 103 has the machine learning unit 133, and the machine learning unit 133 is constructed by an artificial intelligence program trained with learning data based on past examination results. ing.

  By the way, about 320,000 or more patent applications are filed annually on a performance basis in 2014, and the 1st action has already been issued for some or many patent applications before that. Among them, there are many applications (progressive rejection applications) in which the reasons for refusal of violation of inventive step are pointed out in the notice of reasons for refusal.

  In the inventive step rejection application, the examination results differed between the invention described in the claim at the time of examination and the main cited invention. However, the examiner determined that there was an inventive step based only on the difference. It means that there was no. On the other hand, in patent applications, applications for which a patent assessment was issued without the first action being issued, or a reason for refusal was issued, but the reasons for refusal of the inventive step violation were pointed out. There are applications that did not exist (applications without inventive step rejection).

For example, as shown in FIG. 19, if there is a patent application Pd to be examined and its filing date is t ₀ , the main cited invention or sub-cited invention for the patent application Pd can be known or publicly used. are known from the literature and the Internet known invention, is (in FIG. 19, rf1~rf6) Patent Laid-open application are mainly already published prior to the filing date t ₀ disclosed invention.

  Here, it is assumed that, as a result of examination, it is determined that the invention disclosed in the publication rf6 corresponds to the main cited invention. In this case, although there is a difference between the invention according to the patent application Pd and the invention disclosed in the publication rf6, the distance dp corresponding to the difference is the inventive step of the invention according to the patent application Pd. It is thought that it was not large enough to affirm. On the other hand, if it is determined that the invention disclosed in the publication rf6 does not fall under the main cited invention, it is considered that the distance dp is large enough to affirm the inventive step of the application related to the application Pd. It is done.

  If it is possible to determine how much the difference is when the inventive step is affirmed and what is the difference when the invention is denied, it can be considered as an objective judgment material for whether or not the patent requirement is met. In view of the above, for that purpose, it is considered effective to determine the distance dp according to the difference between the two inventions. The machine learning unit 133 determines this by learning the training data based on the past examination results, and gives a measure of whether the inventive step is highly likely to be denied or low.

  In the learning of the machine learning unit 133, in the invention according to the present application (the invention of the present application), the distance dp is regarded as a difference between two document vectors, and the distance dp is learned when there is a reason for refusal of inventive step and when there is no reason for rejection. Therefore, learning is performed using the training data based on the above-described HL pattern.

  Then, when predicting the suitability of patent requirements, the search target invention is searched for documents (1 or 2) having a similarity equal to or higher than a certain value by the concept search using the inspection document data. In addition, full text search (main invention full text search) is performed using the inspection document data to find hit documents. Since the similar document has a document vector having a high degree of similarity to the search target invention vector, it is recognized that there is a high possibility of becoming a main cited document in the published application. However, if there is a document (similar hit document) that matches the document hit in the full-text search among these, only the document with the highest similarity (the most similar hit document) is required for novelty and inventive step. Is likely to be denied. Therefore, the difference between the citation candidate vector RfV obtained from the most similar hit document and the gist vector EV obtained from the search document data of the search object invention is obtained to generate the invention movement vector V3, which is input to the machine learning unit 133. Thus, whether or not there is a reason for refusal of the inventive step violation that cites the most similar hit document is output. As a result, it is possible to give an indication of whether the reason for refusal of violating the inventive step is high or low.

  As described above, the patent requirement suitability prediction server 10 includes the machine learning unit 133, so that the patent requirement suitability is predicted in a manner that reflects the examination results of the Patent Office. Since the judgment results of artificial intelligence can be used for judgment results that must rely solely on the experience and intuition of experts such as examiners and patent attorneys as before, it is possible to make the prediction results objective become. In the conventional prior art search, only related prior art documents are picked up. However, according to the present embodiment, a prediction result of suitability of patent requirements is added to the search result, and the prediction is examined. The content is adapted to the actual business. Therefore, the patent requirement suitability prediction server 10 can effectively reduce the preparation burden of the application documents, and can be expected to improve the efficiency of the rights acquisition work.

  Further, the inventive step prediction processing unit 126 performs a concept search by main citation invention search. Whether or not there is a reason for refusal to violate the inventive step largely depends on whether or not the main cited invention is found, but if a full text search is performed in the main cited invention search to find the main cited invention, there is a possibility that multiple documents will be hit In some cases, the main cited invention (main cited document) cannot be specified. In this regard, in the concept search, similar documents are ordered according to the similarity obtained from the inner product of the document vectors, which is effective for specifying the main cited document. And by predicting whether the reason for refusal of violating the inventive step citing the main cited document is high or low by using the prediction by the machine learning unit 133 together with the main cited document found in the above manner Can do.

  In the above description, as a more preferred embodiment, a case where the cited invention search unit 131 of the inventive step prediction processing unit 126 has a main cited invention search unit and a sub-cited invention search unit is shown. As described above, when the main cited invention is found by the main cited invention search, there is a high possibility that the reason for refusal without inventive step will be found. The generation unit 132 and the machine learning unit 133 may be operated to output the requirement suitability document vector V4. Even in this case, by using together the inventive step prediction data Vd1 relating to the main cited document generated in step 76 and the requirement suitability document vector V4 generated by the machine learning unit 133, the content suitable for the examination practice can be obtained. Since the prediction can be performed and the prediction uses the judgment result of the artificial intelligence, the prediction result can have objectivity. Therefore, it is sufficient that the inventive step prediction processing unit 126 has the main citation invention search unit and does not have to have the sub citation invention search unit, but as described above, the progress having the sub citation invention search unit. The sex prediction processing unit 126 is more preferable.

Second Embodiment Subsequently, a patent requirement suitability prediction server 200 according to a second embodiment will be described with reference to FIGS. As shown in FIG. 21, the patent requirement suitability prediction server 200 is replaced with the patent requirement suitability prediction processing unit 103, the prediction result editing processing unit 105, and the prediction result storage unit 156, as compared with the patent requirement suitability prediction server 10 described above. 3 further includes a patent requirement suitability prediction processing unit 203, a prediction result editing processing unit 205, and a prediction result storage unit 256, and a point that the prediction result list L2 is output instead of the prediction result list L1.

  As shown in FIG. 22, the patent requirement suitability prediction processing unit 203 is different from the patent requirement suitability prediction processing unit 103 in that an inventive step prediction processing unit 226 is used instead of the inventive step prediction processing unit 126 and the prediction result file generation unit 127. And the prediction result file generation unit 227 is different.

  And the inventive step prediction processing unit 226 is different from the inventive step prediction processing unit 126 in that it has a cited invention search unit 231 and a machine learning unit 233 instead of the cited invention search unit 131 and the machine learning unit 133, and an input The difference is that the operation of the vector generation unit 132 is different.

In the inventive step predictive processing unit 126 according to the first embodiment described above, in the post-concept search process after the main cited invention search is performed, the result of comparing the result of the main invention full-text search and the result of the concept search Therefore, the main cited references are determined. Only one main citation is determined. On the other hand, the inventive step prediction processing unit 226 according to the second embodiment extracts n documents (n is an integer of 2 or more) including the most similar documents in descending order of similarity as similar documents. Is set in the main cited document without comparison with the result of the main invention full-text search. In addition, the machine learning unit 233 classifies the _n invention movement vectors V3 _{1 to} V3 _n into three classes, S class, H class, and L class, which will be described later.

  The cited invention search unit 231 has a main cited invention search unit and a sub-cited invention search unit similar to the cited invention search unit 131, but the operation of the main cited invention search unit compared to the cited invention search unit 131. Are different, and the output data is also different. The main cited invention search unit of the cited invention search unit 231 executes post-concept search processing shown in FIG. 28 instead of the post-concept search processing shown in FIG. That is, when the concept search post-processing is started, the CPU 11 advances the process to step 96A, and includes a plurality of documents including documents (most similar documents) having the highest similarity in descending order of similarity among documents in the concept search file 158. And set them in the main citation. Subsequently, the process proceeds to step 97. When the sub-cited invention search process is executed in the same manner as described above, the post-concept search process ends.

Thus, the main cited invention searching unit of the cited invention searching unit 231 extracts n documents including the most similar documents as similar documents, and sets them as main cited documents (doc _{1 to} doc _n ).

The main cited invention with the operation of the search part vary, and generates n matter of inventive step prediction data Vd1 ₁ ~Vd1 _n corresponding to the CPU11 at step 76 each similar document in advance prediction routine described above, They are output to the prediction result file generation unit 227. Further, the CPU 11 outputs the search document data ied to the input vector generation unit 132, but outputs _n concept search data Vd2 _{1 to} Vd2 n corresponding to each similar document to the input vector generation unit 132.

Compared with the machine learning unit 133, the machine learning unit 233 performs n invention movement vectors V3 ₁ to V3 ₁ corresponding to each similar document by machine learning (supervised learning) using an SHL pattern described below as a learning pattern. V3 _n is classified into one of S class, H class, and L class, respectively, and n output signals (requirement conformity document vectors V4 _{1 to} V4 _n ) corresponding to the classification result are output. . The S class, the H class, and the L class are classes that are highly likely not to meet the inventive step requirement, classes that are highly unlikely to be compatible, and conforming classes (the reason for refusal of inventive step violations for the subject invention) Class that is very likely to be found, a high class, a class that is not found).

The SHL pattern is a combination of the following three patterns S, H, and L.
Pattern S: The learning document vector is the first learning document vector, and the teacher vector is a vector indicating that there is a reason for rejection of novelty and inventive step (for example, only the dimension corresponding to the correct class is "1" and the others are "0") Combination pattern H with “vector” of: the learning document vector is the second learning document vector, the teacher vector is a vector indicating the reason for refusal of inventive step and no reason for refusal of novelty (for example, the above corresponding to the correct class) A combination pattern L: a vector in which the learning document vector is the third learning document vector and the teacher vector has no reason for refusal of inventive step. For example, it is a combination pattern with a vector whose only dimension different from the above two is “1” and the other is “0”.

  The first learning document vector is an application for which the first notice of reasons for refusal (1st action) has been issued as a result of examination by the Patent Office in a published application, and the same document is cited in the 1st action. An application (novelty / inventive step) in which the reason for refusal of violation of novelty or inventive step (the reason for refusal that the requirements of Article 29 (1) (iii) and Article 2 (2) of the Patent Act are not met) was pointed out The document vector corresponding to the claim (at the time of issuance of the reason for refusal) where the reason for refusal of the application for refusal was pointed out, and the cited reference 1 at that time (the first main cited publication cited as the main publication) The first moving document vector corresponding to the difference from the document vector corresponding to the object (first quoted document vector).

  The second learning document vector is an application in which the first notice of reasons for refusal (1st action) is issued as a result of examination by the Patent Office in the published application, and the reason for refusal of novelty ( The reason for refusal to violate the inventive step (rejection not satisfying the requirement of paragraph 2 of the same Article) is not pointed out, but the reason for refusal of violating the inventive step is not pointed out The document vector corresponding to the claim (at the time of issuance of the reason for refusal) and the cited reference 1 (main publication) where the reason for refusal of the application (reasonable refusal application) was pointed out The second moving document vector corresponding to the difference from the document vector (second cited document vector) corresponding to the second main cited publication cited as.

  The third learning document vector is the result of examination in the published application, where the first action is not issued and the patent decision is issued (non-rejection application) or the first action is issued but progresses to the reason for refusal Document vectors according to claim 1 (at the time of the notice of reasons for refusal) of applications for which no reason for refusal of sex violation has been pointed out (applications without inventive refusal), and those applications without refusal or inventive refusal As a result of the concept search for the application, a third moving document vector corresponding to the difference from the document vector (non-cited document vector) corresponding to the document (the most similar document for learning) that is considered to have the highest similarity. is there.

  As with the machine learning unit 133, the machine learning unit 233 can apply a neural network modeled on a cranial nerve network for information processing, and preferably uses a BP (back propagation) network among them.

As shown in FIG. 23, the input vector generation unit 132 of the inventive step prediction processing unit 226 includes a search target invention vector generation unit 132a, a citation candidate vector generation unit 132b, and a movement vector generation unit 132c. However, the operations of the citation candidate vector generation unit 132b and the movement vector generation unit 132c are different. That is, reference candidate vector generation unit 132b, since the concept search data Vd2 ₁ ~Vd2 _n of n matter from the cited invention search unit 231 is input, and inputs the publication data of each similar documents included in that each of which Feature words are extracted and weighted according to each word to generate _n document vectors (citation candidate vectors) RfV _{1 to} RfV _n corresponding to each similar document. The movement vector generation unit 132c calculates a difference between the search target invention vector EV and each of the citation candidate vectors RfV _{1 to} RfV _n, and n invention movement vectors V3 _{1 to} V3 according to the difference between the two document vectors. _n is generated.

As described above, each similar document is a document extracted in descending order from the highest similarity by the concept search by the main cited invention search unit. It is estimated that the probability of being cited as a document is high. For this reason, citation candidate vectors RfV _{1 to} RfV _n are obtained using each similar document as a citation candidate, and the difference between these and search object invention vector EV is calculated to obtain invention movement vectors V3 _{1 to} V3 _n. Inventive movement vectors V3 _{1 to} V3 _n corresponding to differences from the inventions disclosed in the similar documents are generated.

  Compared with the prediction result file generation unit 127, the prediction result file generation unit 227 generates a prediction result file having the layout shown in FIG. 24 and stores the prediction result file in the prediction result storage unit 256. The difference is that the operation as the non-conformance rate calculating unit according to the embodiment is performed to calculate the non-conformance rate Vr related to the investigation target invention.

The non-conformance rate Vr indicates the possibility of not meeting the inventive step requirement for the search target invention, and indicates the possibility of finding the reason for refusal of the inventive step violation for the search target invention. The requirement adequacy document vectors V4 _{1 to} V4 _n output from the machine learning unit 233 are generated with contents that are highly likely to be incompatible with the requirement of inventive step, are high, and are not included. It shows the possibility of finding reasons for refusal of violation of inventive step in relation to the literature. Therefore, using these, the prediction result file generation unit 227 calculates the non-conformance rate Vr related to the investigation target invention. In this case, the prediction result file generation unit 227 calculates the non-conformance rate Vr according to the non-conformance rate calculation rule. The non-conformance rate calculation rule is a rule by which the prediction result file generation unit 227 calculates the non-conformance rate Vr. In the present embodiment, the S class included in the requirement conformity document vectors V4 _{1 to} V4 _n , The calculation rule table (not shown) is set so that the numerical value of the non-conformance rate Vr is determined according to the number of cases of the H class and the M class. The non-conformance rate calculation rule can be set as follows, for example.

2 or more S class: Vr ≧ 85%
There is one S class and the number of H class is 50% or more: Vr ≧ 75%
One S class and less than 50% H class: Vr ≧ 65%
S class is 0, H class is more than 50%: Vr ≧ 50%
S class is 0, H class is less than 50%: Vr ≧ 40%
0 for both S and H classes: Vr ≧ 15%

  According to the non-conformance rate calculation rule, for example, when there are five similar documents (the above-mentioned integer n is “5”) and there are two S classes, Vr ≧ 85%. Also, if there is one S class and three H classes, Vr ≧ 75%, but if there are two H classes, Vr ≧ 65%, both S class and H class are zero (all cases are L class) ) Vr ≧ 15%.

  As shown in FIG. 24, the prediction result file generated by the prediction result file generation unit 227 includes a nonconformity rate Vr as compared with the prediction result file generated by the prediction result file generation unit 127. Is different.

  Further, the prediction result editing processing unit 205 is different from the prediction result editing processing unit 105 in that it reads a prediction result file, and edits and outputs a patent requirement suitability prediction list L2 as shown in FIG. . The patent requirement suitability prediction list L2 is different from the patent requirement suitability prediction list L1 in that an OA rate is added. The OA rate indicates the possibility of finding a reason for refusal that violates patent requirements (novelty or inventive step) in the examination process and issuing a notice of reason for refusal indicating that, as described above. A numerical value corresponding to the non-conformity rate Vr is shown.

As described above, in the patent requirement suitability prediction server 200 according to the second embodiment, the cited invention search unit 231 includes the most similar documents in the inventive step predictive processing unit 226 of the patent requirement suitability prediction processing unit 203. Is set as the main cited document. Further, the machine learning unit 233 classifies the inventive movement vectors V3 _{1 to} V3 _n corresponding to the respective similar documents into three classes of S class, H class, and L class, and the requirement suitability document vector V4 ₁ to V class according to the classification result. V4 _n is output.

  In the patent requirement conformity prediction server 10 according to the first embodiment, the prediction regarding patent requirement conformity is performed in a form that conforms to examination practice and reflects examination results of the Patent Office. This also applies to the patent requirement suitability prediction server 200.

  However, the patent requirement suitability prediction server 10 extracts a plurality of documents as similar documents in the main cited invention search, but only one of them (the most similar hit document or the most similar document) in the subsequent concept search post-processing. Was just the main reference. It is considered that the decided document is likely to be cited as the main cited document as a result of the actual examination, but it is not always cited in the actual examination, and the decided document and the next There is very little difference in similarity between documents with similarities (next similar documents), and the next similar documents are more appropriate as the main cited document. Therefore, if a plurality of documents including the most similar documents are extracted in the main cited invention search, and the document vectors are classified by the machine learning unit 233 for these documents, whether or not the patent requirement is met in consideration of the next similar documents. Predictions are made. Therefore, the prediction accuracy of the patent requirement suitability prediction server 200 is improved more than the prediction accuracy of the patent requirement suitability prediction server 10.

In actual examination practice, when a reason for refusal of inventive step violation is found for a patent application, the reason for refusal of inventive step violation and the reason for refusal of novelty violation is the same document (the document cited in this case is newly May also be pointed out by quoting refusal documents of sex / inventive step rejection). In such a case, since the examiner determines that there is no difference between the invention according to the claims of the patent application and the invention disclosed in the novelty / inventive step rejection citation, the invention movement vector V3 _When document vectors classified into the S class are included in _{1 to} V3 _n , it is considered that the possibility of finding the reason for refusal of the inventive step violation is higher than the case where it is not. Therefore, by constructing the machine learning unit 233 by performing machine learning using the SHL pattern so that the distance dp viewed from the novelty / inventive rejection application and the distance dp viewed from the inventive rejection application can be distinguished, It is possible to make a prediction that distinguishes between cases where the reason for refusal of an inventive step violation is very likely and cases where it is not. By doing so, it is possible to expect improvement in prediction accuracy by the patent requirement suitability prediction server 200 and further improvement in business efficiency.

(Modification 1)
In the inventive step prediction processing unit 226, one machine learning unit 233 classifies _n invention movement vectors V3 _{1 to} V3 _n . However, the progress in the patent requirement suitability prediction processing unit 213 shown in FIG. as sex prediction processing unit 227 includes a plurality of machine learning unit ₂₃₃ 1 _{~233 n} corresponding to the invention the movement vector _V3 1 to V3 _n, so each of which classifies the invention movement vector _V3 1 to V3 _n May be. Further, although not shown, a plurality of input vector generation units 132 may be provided according to the number of concept search data Vd2 _{1 to} Vd2 _n . By doing so, each of the machine learning units 233 _{1 to} 233 _n or each of the input vector generation units 132 executes the processing in parallel, so that the processing time can be shortened.

(Modification 2)
In each of the embodiments described above, an example in which the patent requirement suitability prediction server 10 or 200 functions as a patent requirement suitability prediction device by installing the patent requirement suitability prediction program in the patent requirement suitability prediction server 10 or 200 will be described. To explain. In addition, the present invention is also applicable to the case where the user terminal device 30 functions as a patent requirement suitability prediction device. In this case, the above-described patent requirement suitability prediction program is changed according to at least the following changes 1) and 2), and the changed patent requirement suitability prediction program is transferred from the patent requirement suitability prediction servers 10 and 200 to the user terminal device. It may be downloaded to 30 and installed in the user terminal device 30.

Modification 1) The image data for performing the input operation is displayed on the user terminal device 30 without being transmitted from the patent requirement suitability prediction servers 10 and 200 to the user terminal device 30.
Modification 2) The user terminal device 30 outputs a patent requirement suitability prediction list.

  The above description is the description of the embodiment of the present invention, and does not limit the apparatus and method of the present invention, and various modifications can be easily implemented. In addition, an apparatus or a method configured by appropriately combining components, functions, features, or method steps in each embodiment is also included in the present invention.

  For example, the user terminal device may be a notebook computer or a PDA instead of a high-function mobile phone or a tablet-type terminal device. The patent requirement suitability prediction program executed by the CPU 11 can be recorded on various recording media such as a magnetic recording medium, a CD-ROM, and a DVD, and can also be downloaded from a server (not shown) via a network.

  In the above embodiment, it is assumed that the search document is input using the search screen with prediction 181. However, the input area is not divided into one as in the search screen with prediction 181 and the area is determined. Then, the user may input a document as it is conceived, and the input document may be used as a search document as it is. However, by using the search screen with prediction 181, the name of the invention, the purpose / issue, and the solution are classified, and the document corresponding to each of them is input, so that the search document is useful for specifying the gist of the invention. Patterned (or stylized) to contain information. By doing this, the invention to be investigated can be more clearly identified, so that more effective search becomes possible.

  In the above embodiment, search document data is generated using the search document, and full-text search and concept search are executed using the search document data. However, full-text search and concept search are performed using the input search document. May be executed.

  In the first embodiment, in the post-concept search processing (FIG. 17), the main cited document is determined according to the comparison result between the main invention full-text search result and the concept search result. At this time, the document with the highest similarity is set as the main cited document (steps 95 and 98), but a plurality of documents including the document with the highest similarity in descending order of similarity in these steps 95, 98 and 96. May be the main cited document.

  By applying the present invention, it is possible to make a prediction on the suitability of patent requirements with contents that conform to examination practice for the invention subject to examination so that the preparation burden of application documents can be reduced. Enhanced. The present invention can be used in the field of patent requirement suitability prediction apparatus and patent requirement suitability prediction program.

  DESCRIPTION OF SYMBOLS 1 ... Patent requirement suitability prediction system, 10,200 ... Patent requirement suitability prediction server, 11, 31 ... CPU, 30 ... User terminal device, 101 ... Search document data generation unit, 102 ... Sst table generation unit, 103, 203, 213 ... Patent requirement suitability prediction processing unit, 105 ... Prediction result editing processing unit, 125 ... Novelty / expanded prior application prediction processing unit, 126, 226, 227 ... Inventive step prediction processing unit, 132 ... Input vector generation unit, 132a ... Search Target invention vector generation unit, 132b ... citation candidate vector generation unit, 132c ... movement vector generation unit, 133, 233 ... machine learning unit, 153 ... search document data storage unit, 154 ... sst table, 156, 256 ... prediction result storage unit , L1, L2 ... prediction result list.

Claims (4)

Using the search document used for the search of the search subject invention that is the subject of the prior art search, search the public gazette data that is the electronic data of the published patent gazette, and according to the search result, the novelty of the search subject invention Novelty prediction processing unit for generating novelty prediction data indicating whether the requirements of the invention are appropriate, an inventive step prediction processing unit for generating the inventive step prediction data indicating whether the inventive step is appropriate, and the novelty prediction data And a patent result propriety prediction processing means having a prediction result file generation unit that generates a prediction result file indicating the propriety of the patent requirement of the subject invention using the inventive step prediction data,
The inventive step predictive processing unit includes: a main citation invention search unit that searches for a main citation invention that is closest to the search target invention among prior art inventions identified by the publication gazette data; and a document classification that classifies document vectors And
The main citation invention search unit generates main search document data for searching the main citation invention using the search document, performs a concept search for the publication gazette data using the main search document data, As a result of the concept search, a plurality of similar documents including the most similar document with the highest similarity are extracted in descending order of the similarity, and the publication document data using the full-text search term generated from the search document is targeted If there is a similar hit document that matches the plurality of similar documents in the hit document hit in the full-text search, the main cited invention selects the most similar hit document having the highest similarity among the similar hit documents. The main cited reference is disclosed,
The document classification unit classifies an input invention movement vector as one that meets the requirement of inventive step by machine learning using a plurality of training data including a learning document vector and a teacher vector. It is constructed so as to output a requirement conformity document vector according to the classification result, and the invention movement vector is a difference between a search object invention vector according to the search document and a citation candidate vector according to the most similar hit document Patent requirement propriety prediction device which is a vector according to. Using the search document used for the search of the search subject invention that is the subject of the prior art search, search the public gazette data that is the electronic data of the published patent gazette, and according to the search result, the novelty of the search subject invention Novelty prediction processing unit for generating novelty prediction data indicating whether the requirements of the invention are appropriate, an inventive step prediction processing unit for generating the inventive step prediction data indicating whether the inventive step is appropriate, and the novelty prediction data And a patent result propriety prediction processing means having a prediction result file generation unit that generates a prediction result file indicating the propriety of the patent requirement of the subject invention using the inventive step prediction data,
The inventive step predictive processing unit includes: a main citation invention search unit that searches for a main citation invention that is closest to the search target invention among prior art inventions identified by the publication gazette data; and a document classification that classifies document vectors And
The main citation invention search unit generates main search document data for searching the main citation invention using the search document, performs a concept search for the publication gazette data using the main search document data, As a result of the concept search, a plurality of similar documents including the most similar document with the highest similarity are extracted in descending order of similarity, and each similar document is set as a main cited document in which the main cited invention is disclosed,
The document classification unit is a class that has a high possibility that the input invention movement vector does not conform to the requirement of inventive step by machine learning using a plurality of training data including a learning document vector and a teacher vector, a high class, The invention movement vector is constructed so as to be classified into one of the classes conforming to the requirement of the inventive step and a requirement conformity document vector corresponding to the classification result is output, and the invention movement vector is an investigation target invention vector corresponding to the search document And a plurality of vectors according to the difference between each citation candidate vector corresponding to each similar document,
The patent requirement conformity prediction processing means calculates a nonconformance rate indicating a possibility of not conforming to the inventive step requirement for the search target invention according to the plurality of requirement conformity document vectors output from the document classification unit. A patent requirement conformity prediction device further comprising a non-conformance rate calculation unit for calculating. The document classification unit includes a combination of the learning document vector as the training data, the learning document vector being a first learning document vector, the teacher vector being a vector indicating that there is a reason for rejection of novelty and inventive step, and the learning document vector being a first learning document vector. A combination of the learning vector of 2 with a vector indicating that the teacher vector has a reason for refusal of inventive step and no reason for refusal of novelty, and the learning document vector is a third learning document vector and the teacher vector has an inventive step. A combination with a vector indicating no reason for refusal is used,
The first learning document vector is an application in which a notice of reasons for refusal has been issued among published applications that have already been published, and the novel document has no novelty or inventive step by citing the same document in the notice of reasons for refusal The document vector corresponding to the claim for which the reason for refusal was pointed out in the novelty / inventive step refusal application in which the reason for refusal was pointed out, and the first main cited publication which is the above-mentioned document cited for the reason for refusal A first moving document vector corresponding to a difference from the first cited document vector corresponding to the object,
The second learning document vector is an application for which a notice of reason for refusal is issued in the published application, and the reason for refusal of novelty is not pointed out in the notice of reason for refusal, but there is a reason for refusal of inventive step. The document vector corresponding to the claim in which the reason for refusal of the inventive step rejection application pointed out was pointed out, and the second corresponding to the second main cited publication cited as the main publication for the reason for refusal A second moving document vector corresponding to the difference from the cited document vector of
The third learning document vector is a non-rejection application for which a patent decision has been issued without a notice of reason for refusal being issued or an application for which a notice of reason for refusal has been issued in the published application, the notice of reason for refusal And the document vector according to claim 1 of the non-progressive refusal application for which the reason for refusal of inventive step was not pointed out, and the result of the concept search for the non-refusal application or the non-progressive refusal application, The patent requirement conformity prediction apparatus according to claim 2, wherein the third requirement is a third moving document vector obtained from a difference from a non-cited document vector corresponding to the most similar document for learning that is said to be high. A patent requirement suitability prediction program for causing a computer to function as a patent requirement suitability prediction device, wherein the computer is used as a search document used for searching a search subject invention that is a subject of prior art search, A novelty prediction processing unit that searches publication gazette data that is data, and generates novelty prediction data indicating whether the novelty requirement of the searchable invention is appropriate according to the search result; and the inventive step of the searchable invention A predictive result for generating a predictive result data indicating the propriety of patent requirements of the subject invention using the novelty predictive data and the predictive predictive data Function as patent requirement suitability prediction processing means having a file generation unit,
Among the prior art inventions that the inventive step prediction processing unit is specified by the publication data, a main citation invention search unit that searches for a main citation invention that is closest to the search target invention, and a document classification unit that classifies document vectors And
The main citation invention search unit generates main search document data for searching the main citation invention using the search document, performs a concept search for the publication data using the main search document data, and As a result of the concept search, a plurality of similar documents including the most similar document having the highest similarity in descending order of similarity are extracted, and the publication document data using the full-text search term generated from the search document is targeted. When there is a similar hit document matching the plurality of similar documents in the hit document hit by the full-text search, the main cited invention is disclosed as the most similar hit document having the highest similarity among the similar hit documents. As the main cited reference
The document classification unit classifies the input invention movement vector as one that meets the requirement of inventive step by machine learning using a plurality of training data including a learning document vector and a teacher vector. It is constructed so as to output a requirement conformity document vector according to the classification result, and the invention movement vector is a difference between a search object invention vector according to the search document and a citation candidate vector according to the most similar hit document Patent requirement conformity prediction program which is a vector according to

Images