Document search system

Abstract

To achieve a document search system that enables efficient document search regardless of the ability of a user.SOLUTION: Document search is performed using a document search system in which database document data is stored. After first document data and second document data are input to the document search system, the document search system extracts a plurality of terms from the first document data. The extraction of the terms is performed using morphological analysis, for example. Next, the extracted terms are weighted based on the second document data. For example, texts included in a document represented by the second document data are classified into first and second texts. Among the terms extracted from the first document data, the weight of the term included in the first text is set larger than the weights of the other terms. The classification of the texts can be performed in accordance with a rule basis or using machine learning. After that, the similarity of the database document data to the first document data is calculated based on the weighted term.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] One aspect of the present invention relates to a document search system and a document search method. [Background technology]

[0002] For an invention for which a patent has been obtained, it is possible to investigate whether or not there is related prior art by investigating patent application documents and non-patent literature such as papers that were published or filed before the filing date of the application for the patent. The domestic and foreign patent literature and non-patent literature such as papers obtained by conducting the investigation can be used to determine whether there is a risk that one's own patent right will be invalidated, or whether a patent right owned by another person can be invalidated.

[0003] In addition, the use of artificial intelligence (AI) is being considered for various applications. In particular, it is expected that the use of machine learning will lead to the realization of computers with higher performance than conventional von Neumann type computers, and research on machine learning has been progressing in recent years.

[0004] For example, Patent Document 1 discloses an invention that uses an artificial neural network (hereinafter, referred to as a neural network) to search for information on intellectual property similar to input document data. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2018-206376 A Summary of the Invention [Problem to be solved by the invention]

[0006] For example, when searching prior art documents to find reasons for invalidating a patent, it is necessary to create a search query centered on the invention defined in the scope of the patent claims. Although it is expected that the quality of the search query will improve by understanding the invention and the examination history, it is a time-consuming task that requires the searcher's experience and understanding.

[0007] An object of one embodiment of the present invention is to provide a document search system capable of efficiently searching for documents regardless of the ability of a searcher.An object of one embodiment of the present invention is to provide a document search system capable of searching for documents in a simple manner.An object of one embodiment of the present invention is to provide a document search system capable of searching for documents with high accuracy.An object of one embodiment of the present invention is to provide a novel document search system.

[0008] Alternatively, an object of one embodiment of the present invention is to provide a document retrieval method capable of efficiently retrieving documents regardless of the ability of a searcher. Alternatively, an object of one embodiment of the present invention is to provide a document retrieval method capable of retrieving documents in a simple manner. Alternatively, an object of one embodiment of the present invention is to provide a document retrieval method capable of retrieving documents with high accuracy. Alternatively, an object of one embodiment of the present invention is to provide a novel document retrieval method.

[0009] It should be noted that the description of multiple problems does not preclude the existence of each other's problems. One embodiment of the present invention does not need to solve all of the problems exemplified. In addition, problems other than those listed will become apparent from the description of this specification, and such problems may also be problems of one embodiment of the present invention. [Means for solving the problem]

[0010] One aspect of the present invention is a document search system having a database and a processing unit, where the database has a function of storing database document data, the processing unit has a function of extracting a plurality of terms from first document data, the processing unit has a function of weighting at least one of the extracted terms based on second document data, and the processing unit has a function of calculating a similarity between the database document data and the first document data based on the weighted terms.

[0011] Alternatively, in the above aspect, the processing unit may have a function of dividing the sentences included in the document represented by the second document data into first sentences and second sentences, and the processing unit may have a function of assigning a greater weight to terms included in the first sentence among terms extracted from the first document data than to terms not included in the first sentence.

[0012] Alternatively, in the above aspect, the processing unit may have a function of performing machine learning, and the processing unit may have a function of classifying sentences based on the learning results of the machine learning.

[0013] Alternatively, in the above aspect, machine learning is performed so that data output from the processing unit by inputting first training document data into the processing unit approaches second training document data, the first training document data being the same type of document data as the second document data, and the second training document data being document data that has been labeled with the first training document data.

[0014] Alternatively, in the above aspect, the processing unit may have a function of extracting terms using morphological analysis.

[0015] Alternatively, one aspect of the present invention is a document retrieval method using a document retrieval system in which database document data is stored, in which after first document data and second document data are input into the document retrieval system, the document retrieval method extracts a plurality of terms from the first document data, weights at least one of the extracted terms based on the second document data, and calculates a similarity of the database document data to the first document data based on the weighted term.

[0016] Alternatively, in the above aspect, after extracting the terms, the sentences included in the document represented by the second document data may be divided into first sentences and second sentences, and weighting may be performed so that, among the terms extracted from the first document data, the weight of the terms included in the first sentence is greater than the weight of the terms not included in the first sentence.

[0017] Alternatively, in the above aspect, the sentences may be segmented using machine learning.

[0018] Alternatively, in the above aspect, machine learning is performed by inputting first training document data into a processing unit having a function of performing processing using machine learning, so that the data output from the processing unit approaches the second training document data, the first training document data being the same type of document data as the second document data, and the second training document data being document data that has been labeled with the first training document data.

[0019] Alternatively, in the above aspect, the extraction of terms may be performed using morphological analysis. Effect of the Invention

[0020] According to one aspect of the present invention, it is possible to provide a document search system capable of efficiently searching for documents regardless of the ability of a searcher, or a document search system capable of searching for documents with high accuracy, or a novel document search system.

[0021] Alternatively, according to one aspect of the present invention, it is possible to provide a document retrieval method capable of efficiently retrieving documents regardless of the ability of a searcher, or a document retrieval method capable of retrieving documents with high accuracy, or a novel document retrieval method.

[0022] Note that the description of multiple effects does not preclude the existence of other effects. Moreover, one embodiment of the present invention does not necessarily have all of the exemplified effects. Furthermore, problems, effects, and novel features of one embodiment of the present invention other than those described above will become apparent from the description and drawings in this specification. [Brief description of the drawings]

[0023] [Figure 1] FIG. 1 is a block diagram showing an example of a document retrieval system. [Diagram 2] FIG. 2 is a flowchart illustrating an example of a document search method. [Diagram 3] FIG. 3 is a schematic diagram for explaining an example of a document search method. [Figure 4] 4A, 4B1, 4B2, and 4C are tables for explaining an example of a document search method. [Diagram 5] 5A and 5B are tables for explaining an example of a document search method. [Figure 6] Fig. 6A is a flow chart for explaining an example of a document search method, and Fig. 6B is a table for explaining an example of the document search method. [Figure 7] 7A and 7B are schematic diagrams for explaining an example of a document search method. [Figure 8] FIG. 8 is a schematic diagram for explaining an example of a document search method. [Figure 9] FIG. 9 is a schematic diagram for explaining an example of a document search method. [Figure 10]10A and 10B are schematic diagrams for explaining an example of a document search method. [Figure 11] FIG. 11 is a schematic diagram for explaining an example of a document search method. [Figure 12] FIG. 12 is a schematic diagram for explaining an example of a document search method. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] Hereinafter, an embodiment of the present invention will be described. However, one aspect of the present invention is not limited to the following description, and it will be easily understood by those skilled in the art that the form and details of the present invention can be modified in various ways without departing from the spirit and scope of the present invention. Therefore, one aspect of the present invention should not be interpreted as being limited to the description of the embodiment shown below.

[0025] In this specification, the ordinal numbers "first," "second," and "third" are used to avoid confusion between components. Therefore, they do not limit the number of components, nor the order of the components. For example, a "first" component in this specification may be described as a "second" component in the claims. For example, a "first" component in this specification may be omitted in the claims.

[0026] In addition, when the same symbol is used for multiple elements, particularly when it is necessary to distinguish between them, a distinguishing symbol such as "[1]", "[2]", "[n]" may be added to the symbol.

[0027] (Embodiment) In this embodiment, a document retrieval system and a document retrieval method according to one embodiment of the present invention will be described with reference to FIGS.

[0028] In this embodiment, a document search system that can be used to search for intellectual property and a document search method using the document search system will be described as an example of the document search system. Note that the document search system according to one embodiment of the present invention is not limited to searching for intellectual property, and can also be used for searches other than intellectual property.

[0029] In the document retrieval method according to an embodiment of the present invention, first, a user of the document retrieval system according to an embodiment of the present invention inputs first document data and second document data to the document retrieval system according to an embodiment of the present invention. The first document data may be document data representing a document in which an invention is described, and the second document data may be document data representing a document in which the contents described in the document represented by the first document data are discussed and explained. For example, the second document data may be document data representing a feature of the invention described in the document represented by the first document data. For example, if the first document data represents the claims of a patent application, the second document data may represent an argument in which a rebuttal to the reasons for rejection of the patent application is described.

[0030] When the first and second document data are input to a document retrieval system according to an embodiment of the present invention, the document retrieval system extracts a plurality of terms from the first document data. For example, the plurality of terms can be extracted by performing a morphological analysis on the first document data.

[0031] Next, the document search system weights each of the terms extracted from the first document data based on the second document data. For example, the document included in the document represented by the second document data is divided into a first sentence and a second sentence. For example, if the first document data represents the claims of a patent application as described above, and the second document data represents an opinion in which a counterargument against the reason for refusal is described, a sentence representing a feature of an invention asserted to avoid the reason for refusal against the patent application related to the first document data can be regarded as the first sentence. In this case, for example, among the sentences included in the document represented by the second document data, a sentence other than the first sentence can be regarded as the second sentence. Here, the division (also called classification) into the first sentence and the second sentence can be performed using, for example, a rule base or machine learning.

[0032] In this specification, the term "text" refers to a collection of characters expressing a coherent content. Furthermore, the term "document" refers to a medium containing text. For example, patent documents (unexamined patent publications, patent gazettes, etc.), papers, rejection notices, opinions, etc. are documents. For example, the claims, specifications, and abstracts contained in patent documents can each be said to be documents. It should be noted that "documents" may include drawings, images, etc. in addition to text.

[0033] Among the terms extracted from the first document data, the weight of the terms included in the first sentence is set to be greater than the weight of the terms not included in the first sentence. In other words, the weight of the terms included in the first sentence is set to be greater than the weight of the terms not included in the first sentence but included only in the second sentence, and the weight of the terms not included in either the first or second sentence.

[0034] Then, based on the weighted terms, the similarity of database document data, which is document data stored in a database, to the first document data is calculated. For example, it is determined whether or not the terms extracted from the first document data by the above method and weighted are included in the database document data. If the terms extracted from the first document data and weighted are included in the database document data, the weight of the terms is set as the score. If the terms extracted from the first document data and weighted are not included in the database document data, the score is set as 0. The above determination and score calculation are performed, for example, for all the terms extracted from the first document data. The sum of the scores can be used as the similarity of the database document data to the first document data. That is, for example, the sum of the weights of the terms included in the database document data among the terms extracted from the first document data can be used as the similarity of the database document data to the first document data.

[0035] By calculating the similarity using the above method, it is possible to obtain a similarity that reflects the features of the invention described in the document represented by the first document data more than when no weighting is applied to the terms extracted from the first document data. This makes it possible to efficiently search for documents regardless of the capabilities of the user of the document search system according to one embodiment of the present invention, who is the searcher.

[0036] <1. Example of document search system configuration> In this embodiment, a document search system that can be used to search for intellectual property will be described as an example of the document search system. Note that the document search system according to one embodiment of the present invention is not limited to searching for intellectual property, and can also be used for searching other than intellectual property.

[0037] Fig. 1 is a block diagram showing an example of the configuration of a document search system 10, which is a document search system according to one embodiment of the present invention. In Fig. 1, arrows indicate the flow of data, etc. Note that the flow of data, etc. shown in Fig. 1 is only an example, and data, etc. may flow in a direction not shown by the arrows.

[0038] In the drawings attached to this specification, the components are classified by function and shown as independent blocks in the block diagrams, but in actuality, it is difficult to completely separate the components by function, and one component may be involved in multiple functions, or one function may be involved in multiple components.

[0039] FIG. 1 is a block diagram showing a configuration example of a document search system 10. The document search system 10 has at least a processing unit 20, and the processing unit 20 has a weighting unit 21 and a search execution unit 23. The document search system 10 can have an input unit 11, a storage unit 15, a database 17, and an output unit 19 in addition to the processing unit 20. The weighting unit 21, the search execution unit 23, the input unit 11, the storage unit 15, the database 17, and the output unit 19 may be provided in a PC (Personal Computer) of a user of the document search system 10, or may be provided in a server connected to the PC via a network. For example, some of the weighting unit 21, the search execution unit 23, the input unit 11, the storage unit 15, the database 17, and the output unit 19 may be provided in the PC of the user of the document search system 10, and the other parts may be provided in a server connected to the PC via a network.

[0040] Data is input to the input unit 11. For example, a user of the document retrieval system 10 can input data to the input unit 11. The data input to the input unit 11 can be supplied to the weighting unit 21. The data input to the input unit 11 may also be supplied to the search execution unit 23, the storage unit 15, the database 17, etc. A part of the data input to the input unit 11 may also be supplied to the weighting unit 21, the search execution unit 23, the storage unit 15, the database 17, etc.

[0041] The storage unit 15 has a function of storing a program executed by the processing unit 20. The storage unit 15 may also have a function of storing the processing results of the processing unit 20, data input to the input unit 11, and the like.

[0042] The storage unit 15 has at least one of a volatile memory and a non-volatile memory. The storage unit 15 may have a volatile memory such as a DRAM or an SRAM. The storage unit 15 may have a non-volatile memory such as a resistance change type memory such as a resistive random access memory (ReRAM), a magnetoresistive type memory such as a phase change random access memory (PRAM), a ferroelectric random access memory (FeRAM), a magnetoresistive random access memory (MRAM), or a flash memory. The storage unit 15 may also have a recording media drive such as a hard disk drive (HDD) or a solid state drive (SSD).

[0043] The database 17 has at least a function of storing database document data to be searched. The database 17 may also have a function of storing the processing results by the processing unit 20, data input to the input unit 11, etc. The storage unit 15 and the database 17 do not need to be separated from each other. For example, the document search system 10 may have a storage unit having the functions of both the storage unit 15 and the database 17.

[0044] The processing unit 20 has a function of performing processing such as calculation and inference using data supplied from the input unit 11, the storage unit 15, the database 17, etc. The processing unit 20 has a function of supplying the processing result to the output unit 19. The processing unit 20 may also have a function of supplying the processing result to the storage unit 15, the database 17, etc.

[0045] Specifically, of the weighting unit 21 and search execution unit 23 included in the processing unit 20, the weighting unit 21 has a function of performing processing using data supplied from the input unit 11 and the storage unit 15. The search execution unit 23 has a function of performing search processing using the processing result of the weighting unit 21 and data supplied from the input unit 11, the storage unit 15, and the database 17. The search results by the search execution unit 23 can be supplied to the output unit 19. The specific processing that can be performed by the weighting unit 21 and the search execution unit 23 will be described later.

[0046] The processing unit 20 includes, for example, an arithmetic circuit or a central processing unit (CPU: Central Processing Unit).

[0047] The processing unit 20 may have a microprocessor such as a DSP (Digital Signal Processor) or a GPU (Graphics Processing Unit). The microprocessor may be configured with a PLD (Programmable Logic Device) such as an FPGA (Field Programmable Gate Array) or an FPAA (Field Programmable Analog Array). The processing unit 20 can perform various data processing and program control by interpreting and executing instructions from various programs using the processor. Programs that can be executed by the processor are stored in at least one of a memory area of ​​the processor and the storage unit 15.

[0048] The processing unit 20 may have a main memory. The main memory has at least one of a volatile memory such as a random access memory (RAM) and a non-volatile memory such as a read only memory (ROM).

[0049] The RAM may be, for example, a dynamic random access memory (DRAM) or a static random access memory (SRAM), and a virtual memory space is allocated and used as a working space for the processing unit 20. The operating system, application programs, program modules, program data, lookup tables, and the like stored in the storage unit 15 are loaded into the RAM for execution. The data, programs, and program modules loaded into the RAM are each directly accessed and operated by the processing unit 20.

[0050] ROM can store BIOS (Basic Input / Output System) and firmware that do not require rewriting. Examples of ROM include mask ROM, OTPROM (One Time Programmable Read Only Memory), EPROM (Erasable Programmable Read Only Memory), etc. Examples of EPROM include UV-EPROM (Ultra-Violet Erasable Programmable Read Only Memory), which allows the erasure of stored data by exposure to ultraviolet light, EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory, etc.

[0051] The output unit 19 has a function of outputting data of the document search system 10. Specifically, for example, it can output the processing result of the processing unit 20. For example, the search result by the search execution unit 23 may be displayed on a display device, or the search result may be output in the form of an electronic file.

[0052] <2. Document search method 1> A document search method using the document search system 10 will be described. FIG. 2 is a flow chart showing an example of a document search method using the document search system 10. FIG. 3 is a schematic diagram explaining the operation of step S01 shown in FIG. 2. FIG. 4A is a schematic diagram explaining the operation of step S02 shown in FIG. 2. FIGS. 4B1 and 4B2 are schematic diagrams explaining the operation of step S03 shown in FIG. 2. FIGS. 4C, 5A, and 5B are schematic diagrams explaining the operation of step S04 shown in FIG. 2.

[0053] [Step S01] First, a user of the document retrieval system 10 inputs the document data TD and the reference document data RTD to the input unit 11. The document data TD and the reference document data RTD may be input simultaneously or separately.

[0054] The document data TD and the reference document data RTD can be supplied from the input unit 11 to the weighting unit 21 .

[0055] The document data TD includes, for example, documents explaining inventions, devices, designs, pre-release industrial products, technical information, or technical ideas. In particular, the document data TD may include patent claims, abstracts, or documents explaining the outline of an invention. Since such documents have a smaller amount of text than the full text of a specification, it is easy to extract characteristic keywords included in the document data TD.

[0056] The reference document data RTD can be document data representing a document that discusses and explains the contents described in the document represented by the document data TD. For example, the reference document data RTD can be document data representing the features of the invention, etc. described in the document represented by the document data TD. For example, the written opinion describes the features of the invention, etc. related to the patent application, etc., in order to argue against the reasons for refusal of the patent application, etc. Specifically, for example, in order to resolve the reasons for refusal for violation of novelty and inventive step, when a point of differentiation from the cited document is described in the written opinion, the point of differentiation can be the feature of the invention, etc. related to the patent application, etc. Thus, for example, the reference document data RTD can be a written opinion describing a counterargument against the reasons for refusal of the patent application, etc. related to the document data TD. Or, when an objection is filed against the patent, etc. related to the document data TD, the reference document data RTD can be a written opinion describing a counterargument against the reasons for objection for violation of novelty and inventive step. Alternatively, when an invalidation trial is filed against a patent, etc., related to the document data TD, the reference document data RTD can be used as a written reply in which arguments against the invalidation grounds for lack of novelty and lack of inventive step are described. Alternatively, when the features of the invention, etc., described in the document represented by the document data TD are described in a written petition, the reference document data RTD can be used as the written petition. In FIG. 3, the document data TD is the scope of the patent claims, and the reference document data RTD is the written opinion. In the following description, the document data TD is the scope of the patent claims, and the reference document data RTD is the written opinion in which arguments against the grounds for rejection notified against the patent application related to the document data TD are described.

[0057] In the document retrieval method using the document retrieval system 10, document data similar to the document data TD can be retrieved from the database document data while taking into account the reference document data RTD.

[0058] [Step S02] Next, the weighting unit 21 extracts terms from the document data TD to obtain the term data TmD. For example, the terms are extracted by performing a morphological analysis on the document data TD. Of the extracted terms, for example, nouns can be regarded as terms included in the term data TmD, and for example, all nouns can be regarded as terms included in the term data TmD. FIG. 4A shows an example in which the terms "aaa", "bbb", "ccc", and "ddd" are extracted from the document data TD and are included in the term data TmD. Note that weights will be described later.

[0059] Here, the "term" extracted from the document data TD etc. may be the smallest unit that can be divided into morphemes, such as a "noun," a "verb," ​​or a "particle." Alternatively, one "term" may be composed of two or more words. Alternatively, a phrase may be a "term."

[0060] [Step S03] Next, the weighting unit 21 weights each term included in the term data TmD based on the reference document data RTD. The weighted term data is referred to as the term data TmD. w For example, the sentences included in the document represented by the reference document data RTD are divided (also called classification) into a first sentence and a second sentence, and the weight of the term included in the first sentence is set to be greater than the weight of the term not included in the first sentence. That is, the weight of the term included in the first sentence is set to be greater than the weight of the term not included in the first sentence but only in the second sentence, and the weight of the term not included in either the first or second sentence. Note that the sentences included in the document represented by the reference document data RTD may be divided into first to third sentences. In this case, for example, the weight of the term included in the first sentence may be set to be greater than the weight of the term not included in the first sentence but included in the second sentence. Also, the weight of the term not included in the first sentence but included in the second sentence may be set to be greater than the weight of the term not included in the first sentence but included in the second sentence. Note that the sentences included in the document represented by the reference document data RTD may be divided into first to k-th sentences (k is an integer of 2 or more).

[0061] Here, for example, among the sentences included in the document represented by the reference document data RTD, a sentence that strongly expresses the characteristics of the invention, etc., related to the document data TD can be set as the first sentence, and the other sentences can be set as the second sentence. For example, if the document data TD is a patent claim and the reference document data RTD is an opinion in which a counterargument against the reasons for refusal is described, among the sentences included in the reference document data RTD, a sentence that strongly asserts the characteristics of the invention, etc., in order to avoid the reasons for refusal against the patent application related to the document data TD can be set as the first sentence. For example, a sentence or paragraph containing the words "difference," "not disclosed," and "different" can be set as the first sentence. The classification of the sentences included in the reference document data can be performed on a rule basis using such specific terms as keywords.

[0062] In addition, the first sentence included in the reference document data RTD does not have to be one continuous sentence. In other words, the reference document data RTD may include a plurality of first sentences. The same applies to the second to k-th sentences.

[0063] Fig. 3 shows an example in which the term "bbb" is included in a first sentence 31, and the terms "aaa" and "ddd" are included in a second sentence 32. Fig. 4B1 shows a case in which, among the terms included in the term data TmD (terms extracted from the document data TD in step S02), the weight of the term "bbb" included in the first sentence 31 is made larger than the weight of the other terms.

[0064] For example, an example of a method for calculating the weight of each term included in the term data TmD when dividing the sentences included in the document represented by the reference document data RTD into a first sentence 31 and a second sentence 32 will be described. Of the terms included in the term data TmD, those that are not included in the first sentence 31 are assigned a weight of 1. The weight of the terms included in the first sentence 31 is calculated using the following formula. Here, the weight of the terms included in the first sentence 31 is W, the number of terms included in the term data TmD is N, and the number of terms included in the term data TmD that are included in the first sentence 31 is N. ftr Let us assume that.

[0065]

number

[0066] By calculating the weight of the term included in the first sentence 31 using Equation 1, the greater the number N of terms included in the term data TmD, the greater the weight of the term included in the first sentence 31. In addition, the number N of terms included in the term data TmD that are included in the first sentence 31 is ftr The smaller the number of terms in the first sentence 31, the greater the weight of the term.

[0067] FIG. 4B1 shows an example of the weight of the term “bbb” calculated using Formula 1. In Formula 1, N is 4, N ftr is 1, so the weight W is 3.

[0068] Note that, among the terms included in the term data TmD, the weight of a term that is not included in either the first sentence 31 or the second sentence 32 may be set to be smaller than the weight of a term that is not included in the first sentence 31 but is included in the second sentence 32. Fig. 4B2 shows an example in which the weight of the term "ccc" that is not included in either the first sentence 31 or the second sentence 32 shown in Fig. 3 is set to 0, and the weight of the terms "aaa" and "ddd" that are not included in the first sentence 31 but included in the second sentence 32 is set to 1.

[0069] [Step S04] Next, the term data TmD w Based on this, the search execution unit 23 calculates the similarity of the database document data DbTD to the document data TD. For example, w It is determined whether the term included in the database document data DbTD is included in the term data TmD. w If a term included in the term data TmD is included in the database document data DbTD, the weight of the term is used as the score. wIf the term included in the term data TmD is not included in the database document data DbTD, the score is set to 0. w The sum of the scores can be used as the similarity of the database document data DbTD to the document data TD. w The sum of the weights of the terms included in the database document data DbTD among the terms included in can be regarded as the similarity of the database document data to the document data TD. The similarity calculation method described above can be expressed by the following formula. Here, the similarity of the database document data DbTD to the document data TD is SIM, and the term data TmD is w Let m be the number of terms in and W be the weight of the i-th term. i In addition, if the database document data DbTD contains the i-th term, then Y i If not included, set it to 1. i Let be 0.

[0070]

number

[0071] 4C shows an example of the calculation result of the similarity of the database document data DbTD[1] to the database document data DbTD[n] (n is an integer equal to or greater than 1) to the document data TD. w The terms included in and their weights are as shown in Fig. 4B1. In Fig. 4C, terms included in the database document data DbTD are indicated with "◯", and terms not included in the database document data DbTD are indicated with "X".

[0072] In the example shown in FIG. 4C, database document data DbTD[1] includes the terms "aaa", "ccc", and "ddd", but does not include the term "bbb". Since the weights of the terms "aaa", "ccc", and "ddd" are all 1, the similarity of database document data DbTD[1] to document data TD can be set to 3. Furthermore, database document data DbTD[2] includes the terms "aaa" and "bbb", but does not include the terms "ccc" and "ddd". Since the weight of the term "aaa" is 1 and the weight of the term "bbb" is 3, the similarity of database document data DbTD[2] to document data TD can be set to 4.

[0073] Moreover, the database document data DbTD[n-1] contains all of the terms "aaa", "bbb", "ccc", and "ddd". Since the weights of the terms "aaa", "ccc", and "ddd" are 1 and the weight of the term "bbb" is 3, the similarity of the database document data DbTD[n-1] to the document data TD can be set to 6. Furthermore, the database document data DbTD[n] does not contain any of the terms "aaa", "bbb", "ccc", or "ddd". In this case, the similarity of the database document data DbTD[n] to the document data TD can be set to 0.

[0074] For the entire text included in the database document data DbTD, the term data TmD w For example, instead of determining whether the term included in the database document data DbTD is included in the term data TmD, the determination may be performed for a part of the sentences included in the database document data DbTD. w Alternatively, for example, among the sentences included in the database document data DbTD, only sentences other than those expressing bibliographic items may be included in the term data TmD. w It may also be possible to determine whether the document contains terms included in the

[0075] In the example shown in FIG. 4C, even if the same term appears multiple times in the same database document data DbTD, the number of occurrences is not taken into consideration when calculating the similarity, but the number of occurrences may be taken into consideration. For example, w The score is calculated by multiplying the number of occurrences of a term included in the term data TmD by the weight of the term. w The sum of the scores can be regarded as the similarity of the database document data DbTD to the document data TD. The above-mentioned method of calculating the similarity can be expressed by the following formula. Here, the similarity of the database document data DbTD to the document data TD is SIM, and the term data TmD is w Let m be the number of terms in and W be the weight of the i-th term. i Let T be the number of occurrences of the i-th term in the database document data DbTD. i Let us assume that.

[0076]

number

[0077] 5A shows an example of a calculation result of the similarity of the database document data DbTD[1] to DbTD[n] to the document data TD. In the example shown in FIG. 5A, the similarity is calculated by Equation 3. In FIG. 5A, the term data TmD w 4C, the term data TmD indicates the number of occurrences in the database document data DbTD for each term (term extracted from the document data TD in step S02). w contains the terms "aaa", "bbb", "ccc", and "ddd", and the terms "aaa", "ccc", and "ddd" have a weight of 1, and the term "bbb" has a weight of 3.

[0078] In the example shown in FIG. 5A, in the database document data DbTD[1], the term "aaa" appears 10 times, the term "ccc" appears 3 times, and the term "ddd" appears 5 times. Meanwhile, the term "bbb" does not appear even once. Since the weights of the terms "aaa", "ccc", and "ddd" are all 1, the similarity of the database document data DbTD[1] to the document data TD can be set to 18. Furthermore, in the database document data DbTD[2], the term "aaa" appears 4 times, and the term "bbb" appears 7 times. Meanwhile, the terms "ccc" and "ddd" do not appear even once. Since the weight of the term "aaa" is 1, and the weight of the term "bbb" is 3, the similarity of the database document data DbTD[2] to the document data TD can be set to 25.

[0079] In addition, in the database document data DbTD[n-1], the term "aaa" appears 15 times, the term "bbb" appears 10 times, the term "ccc" appears 20 times, and the term "ddd" appears 25 times. Since the weights of the terms "aaa", "ccc", and "ddd" are 1 and the weight of the term "bbb" is 3, the similarity of the database document data DbTD[n-1] to the document data TD can be set to 90. Furthermore, in the database document data DbTD[n], none of the terms "aaa", "bbb", "ccc", and "ddd" appears even once. In this case, the similarity of the database document data DbTD[n] to the document data TD can be set to 0.

[0080] When a term extracted from document data TD appears repeatedly in database document data DbTD, there is a high possibility that the contents described in the document represented by document data TD are similar to the contents described in the document represented by database document data DbTD. For example, there is a high possibility that an invention, etc. described in the document represented by document data TD is similar to the invention, etc. described in the document represented by database document data DbTD. Therefore, by calculating the similarity taking into account the number of times the same term appears in database document data DbTD, the document retrieval system 10 can calculate the similarity with high accuracy.

[0081] In the example shown in FIG. 5A, the term data TmD w For all terms included in the database document data DbTD, the number of occurrences of the term was taken into consideration when calculating the similarity, but one aspect of the present invention is not limited to this. For example, the consideration of the number of occurrences of the term may be changed depending on the magnitude of the weight. For example, the number of occurrences of the term may be considered only for terms whose weight is equal to or greater than a specified value when calculating the similarity. Alternatively, the number of occurrences of the term specified by the user of the document search system 10 may be considered only for terms whose weight is equal to or greater than a specified value when calculating the similarity. Alternatively, an upper limit value for the number of occurrences to be considered may be determined. For example, if the upper limit value is set to 10, the similarity can be calculated by regarding a term that has appeared 11 or more times as having appeared only 10 times.

[0082] 5B shows an example of the calculation result of the similarity of the database document data DbTD[1] to the database document data DbTD[n] to the document data TD. Note that, as in the example shown in FIG. 5A, the term data TmD w includes the terms "aaa", "bbb", "ccc", and "ddd", and the weights of the terms "aaa", "ccc", and "ddd" are set to 1, and the weight of the term "bbb" is set to 3. The number of times the terms "aaa", "bbb", "ccc", and "ddd" appear in the database document data DbTD[1] to database document data DbTD[n] is also set to the same as the example shown in FIG. 5A.

[0083] In the example shown in Fig. 5B, the term "bbb" with a weight of 3 takes into consideration the number of times it appears in the database document data DbTD, while the terms "aaa", "ccc", and "ddd" with a weight of 1 do not take into consideration the number of times it appears in the database document data DbTD. Specifically, for the terms "aaa", "ccc", and "ddd", if they appear once or more in the database document data DbTD, the score is set to 1, and if they do not appear even once, the score is set to 0. On the other hand, for the term "bbb", the product of the weight and the number of times it appears in the database document data DbTD is set as the score. The sum of the scores for the terms "aaa", "bbb", "ccc", and "ddd" is set as the similarity of the database document data DbTD to the document data TD.

[0084] 5B, the similarity of the database document data DbTD[1] to the document data TD can be 3, and the similarity of the database document data DbTD[2] to the document data TD can be 22. In addition, the similarity of the database document data DbTD[n-1] to the document data TD can be 33, and the similarity of the database document data DbTD[n] to the document data TD can be 0.

[0085] Even if the term is extracted from the document data TD, it is highly likely that a term with a low weight does not strongly express the characteristics of the invention, etc. related to the document data TD. For example, when the reference document data RTD is a written opinion and the points of differentiation from the cited documents described in the written opinion are the characteristics of the invention, etc. related to the document data TD, a term with a low weight may be used only in a counterargument against the violation of the description requirements, etc. In this case, a term with a low weight may not express any points of differentiation from the cited documents at all. Therefore, if the number of occurrences in the database document data DbTD is taken into consideration even for a term with a low weight, it is possible that the similarity of the database document data DbTD to the document data TD cannot be calculated with high accuracy. Therefore, by changing whether or not to consider the number of occurrences of the term depending on the magnitude of the weight, it is possible to increase the similarity of the database document data DbTD representing a database document in which an invention, etc. having a similar characteristic to the characteristic of the invention, etc. related to the document data TD, to the document data TD. Therefore, the document search system 10 can calculate the similarity with high accuracy.

[0086] As described above, after calculating the similarity by the methods shown in Figures 4C, 5A, 5B, etc., the document retrieval system 10 outputs the database document data DbTD based on the calculated similarity. For example, the document retrieval system 10 outputs a predetermined number of database document data DbTD in descending order of similarity. Or, for example, the document retrieval system 10 outputs database document data DbTD whose similarity is equal to or greater than a specified value. As described above, the document retrieval system 10 can search for database document data DbTD similar to document data TD while taking into account the reference document data RTD.

[0087] As described above, the document retrieval method according to one embodiment of the present invention extracts terms from document data TD, weights the extracted terms based on reference document data RTD, and calculates the similarity of database document data DbTD to document data TD while taking the weights into consideration. By calculating the similarity in this manner, it is possible to obtain a similarity that reflects the features of the invention or the like described in the document represented by document data TD more effectively than when the terms extracted from document data TD are not weighted. This allows documents to be efficiently searched for, regardless of the capabilities of the user of document retrieval system 10, who is the searcher.

[0088] <3. Document search method 2> Fig. 6A is a flowchart showing an example of a document retrieval method using the document retrieval system 10. In the method shown in Fig. 6A, the operations of steps S11 and S12 are similar to the operations of steps S01 and S02 shown in Fig. 2 and the like.

[0089] [Step S13] After the operation of step S12 is completed, the search execution unit 23 calculates the similarity of the database document data DbTD to the document data TD based on the term data TmD acquired by the weighting unit 21 by extracting terms from the document data TD. This similarity is set as the first similarity. The first similarity can be calculated in the same manner as in step S04 shown in FIG. 2 etc. Here, the weights of the terms included in the term data TmD can all be the same. That is, for example, the weights of all terms can be set to 1. From the above, it can be said that the first similarity is a similarity calculated without taking into account the reference document data RTD.

[0090] The operations in steps S14 and S15 are similar to the operations in steps S03 and S04 shown in FIG.

[0091] [Step S16] After the operation of step S15 is completed, the search execution unit 23 compares the first similarity with the second similarity. For example, database document data DbTD in which the second similarity is higher than the first similarity is presented to the user of the document search system 10. For example, database document data DbTD in which the difference between the second similarity and the first similarity is equal to or greater than a specified value is presented to the user of the document search system 10.

[0092] Alternatively, for example, the search execution unit 23 creates a first ranking in which the database document data DbTD are arranged in descending order from the highest first similarity, and a second ranking in which the database document data DbTD are arranged in descending order from the highest second similarity. Thereafter, the database document data DbTD whose rank in the second ranking is different from that in the first ranking is presented to the user of the document search system 10. For example, the database document data DbTD whose rank in the second ranking is higher than that in the first ranking is presented to the user of the document search system 10. For example, the database document data DbTD whose rank in the second ranking is higher than that in the first ranking by a specified value or more is presented to the user of the document search system 10. The above is an example of the document search method shown in FIG. 6A.

[0093] 6B shows an example of the calculation results of the first and second similarities of the database document data DbTD[1] to DbTD[n] to the document data TD. Here, the terms included in the term data TmD (i.e., the terms extracted from the document data TD in step S12) are the four terms "aaa", "bbb", "ccc", and "ddd", similar to the case shown in FIG. 4C.

[0094] The first and second similarities are calculated by Equation 2. Here, the first similarity is calculated by setting the weights of these four terms to 1, and the second similarity is calculated by setting the weights of these four terms to the values ​​shown in FIG. 4B1. In FIG. 6B, terms included in the database document data DbTD are indicated by "◯", and terms not included in the database document data DbTD are indicated by "X". Whether or not each term is included in the database document data DbTD[1] to DbTD[n] is the same as the example shown in FIG. 4C.

[0095] In the example shown in FIG. 6B, the first similarity of the database document data DbTD[1] to the document data TD can be 3, and the second similarity can also be 3. The first similarity of the database document data DbTD[2] to the document data TD can be 2, and the second similarity can also be 4. The first similarity of the database document data DbTD[n-1] to the document data TD can be 4, and the second similarity can also be 6. The first similarity of the database document data DbTD[n] to the document data TD can be 0, and the second similarity can also be 0. As a result, the second similarity of the database document data DbTD[2] and the database document data DbTD[n-1] is greater than the first similarity. FIG. 6B shows an example in which the second similarity of the database document data DbTD[2] and the second similarity of the database document data DbTD[n-1] are highlighted by encircling them.

[0096] <4. Document search method 3> In the document retrieval method according to one embodiment of the present invention, in order to weight the terms (terms extracted from the document data TD) included in the term data TmD, the sentences included in the reference document data RTD are classified into a first sentence 31 and a second sentence 32, but this classification may be performed using machine learning. For example, weighting may be performed using algorithms such as Naive Bayes, Support Vector Machine, and Conditional Random Fields (CRFs). Alternatively, a circuit having a neural network may be provided in the weighting unit 21, and weighting may be performed using the neural network. In addition, when converting the terms extracted from the document data TD into distributed representations in order to perform machine learning, Bidirectional Encoder Representations from Transformer (BERT), which can obtain different distributed representations for the same term depending on the context, may be used. When the document retrieval system 10 has a function of performing machine learning, for example, the weighting unit 21 may have a function of performing machine learning.

[0097] [study] An example of the learning method will be described. First, learning reference document data LRTD is prepared. The learning reference document data LRTD can be the same type of document data as the reference document data RTD. That is, the learning reference document data LRTD can be, for example, a written opinion (see FIG. 7A).

[0098] Next, the sentences included in the learning reference document data LRTD are labeled. The learning reference document data LRTD including the labeled sentences is called the learning reference document data LRTD. LB (See FIG. 7A.)

[0099] For example, label A is assigned to a sentence that strongly represents the characteristics of an invention, etc., related to document data (e.g., claims) linked to the learning reference document data LRTD, and label B is assigned to other sentences. That is, the sentences included in the learning reference document data LRTD can be divided into first sentences and second sentences, and label A is assigned to the first sentences and label B is assigned to the second sentences. Labeling can be performed, for example, on a paragraph-by-paragraph basis. That is, for example, label A can be assigned to a paragraph containing a sentence that strongly represents the characteristics of an invention, etc., related to document data linked to the learning reference document data LRTD, and label B can be assigned to other paragraphs.

[0100] When labeling is performed using the above method, it is preferable that the sentence to which label A is assigned contains terms contained in the document data linked to the learning reference document data LRTD. This allows accurate weighting of the terms contained in the term data TmD. Therefore, the document retrieval system 10 can calculate the similarity with high accuracy. Although the above describes a case where two types of labels (label A and label B) are assigned to the learning reference document data LRTD, three or more types of labels may be assigned.

[0101] In addition, terms included in both the learning reference document data LRTD and the document data linked to the learning reference document data LRTD may be replaced with a predetermined character string, etc. This makes it possible to suppress overlearning.

[0102] Next, the learning reference document data LRTD is input to the weighting unit 21. The weighting unit 21 outputs the learning reference document data LRTD LB Specifically, the label attached to the learning reference document data LRTD is LBThe weighting unit 21 performs learning so as to approach the label attached to the neural network. This enables the weighting unit 21 to obtain the learning result P (see FIG. 7B). For example, when the weighting unit 21 has a circuit in which a neural network is configured, the learning result P can be a weighting coefficient corresponding to each neuron included in the neural network. From the above, it can be said that the weighting unit 21 can perform supervised learning.

[0103] [Weighting for terms extracted from document data TD] In step S03 shown in FIG. 2 etc., or step S14 shown in FIG. 6A etc., when the reference document data RTD is input to the weighting unit 21 that has been trained by the above method, labeling is performed based on the training result P. For example, a label A can be assigned to a sentence that strongly expresses the characteristics of the invention, etc. related to the document data TD, and a label B can be assigned to other sentences. The labeled reference document data RTD is LB (See FIG. 8). From the above, it is possible to classify the sentences included in the document represented by the reference document data RTD. For example, the sentence with label A attached can be regarded as the first sentence 31, and the sentence with label B attached can be regarded as the second sentence 32.

[0104] Thereafter, the weighting unit 21 assigns a weight to each term included in the term data TmD. For example, the weight of a term included in a sentence with label A is made larger than the weight of a term not included in a sentence with label A. By the above method, it is possible to weight the terms extracted from the document data TD.

[0105] <5. Document Search Method 4> The method of weighting the terms included in the term data TmD (terms extracted from the document data TD) by using a neural network is not limited to the above method. Below, examples of a neural network learning method and a weighting method for terms extracted from the document data TD that are different from the above method will be described. Note that, as with the above method, Naive Bayes, Support Vector Machine, CRFs, etc. can be used as the learning algorithm.

[0106] [study] An example of a neural network learning method will be described. First, learning document data LTD is prepared. Also, learning reference document data LRTD linked to the learning document data LTD is prepared.

[0107] The learning document data LTD can be the same type of document data as the document data TD. That is, the learning document data LTD can be, for example, the scope of a patent claim. Also, as described above, the learning reference document data LRTD can be the same type of document data as the reference document data RTD. That is, the learning reference document data LRTD can be, for example, a written argument that describes a rebuttal to the reasons for rejection of a patent application or the like related to the learning document data LTD.

[0108] Next, the training term data LTmD is obtained by extracting terms from the training document data LTD (see FIG. 9). The extraction of terms from the training document data LTD can be performed in the same manner as the extraction of terms from the document data TD by the weighting unit 21. For example, the terms are extracted by performing a morphological analysis on the training document data LTD. Of the extracted terms, for example, nouns can be regarded as terms included in the training term data LTmD, and for example, all nouns can be regarded as terms included in the training term data LTmD.

[0109] In addition, the sentences represented by the training reference document data LRTD are divided. For example, the sentences represented by the training reference document data LRTD are divided into first sentences and second sentences. For example, of the sentences represented by the training reference document data LRTD, sentences that strongly represent the features of the invention, etc. related to the training document data LTD are highlighted. The highlighted sentences can be set as first sentences, and the other sentences can be set as second sentences. The training reference document data LRTD including the highlighted sentences is divided into first sentences and second sentences. HL (See FIG. 10A.) The learning reference document data LRTD shown in FIG. 10A is HL In the example, the text included in the highlighted portion HL is referred to as the highlighted text.

[0110] Next, the learning reference document data LRTD HL Based on this, the terms included in the training term data LTmD are labeled. For example, of the terms included in the training term data LTmD, label A can be assigned to those included in the highlighted sentence, i.e., the first sentence, and label B can be assigned to the other terms. This makes it possible to assign label A to those terms included in a sentence that strongly represents the characteristics of the invention, etc. related to the training document data LTD, and label B to the other terms. The training term data LTmD including the labeled terms is then stored as the training term data LTmD LB (See FIG. 10B.) Although the above description has been given of a case in which two types of labels (label A and label B) are attached to the terms included in the training terminology data LTmD, three or more types of labels may be attached.

[0111] Next, the training term data LTmD and the training reference document data LRTD are input to the weighting unit 21. The weighting unit 21 outputs the training term data LTmD LB , and learning reference document data LRTD HL Specifically, for example, the labels given to the terms included in the training term data LTmD input to the neural network are LBThe weighting unit 21 performs learning so as to approach the labels attached to the terms included in the neural network. This enables the weighting unit 21 to obtain the learning result P (see FIG. 11). For example, when the weighting unit 21 has a circuit in which a neural network is configured, the learning result P can be a weighting coefficient corresponding to each neuron included in the neural network. In this way, the weighting unit 21 can perform supervised learning.

[0112] [Weighting for terms extracted from document data TD] In step S03 shown in FIG. 2 etc., or step S14 shown in FIG. 6A etc., when the term data TmD and the reference document data RTD are input to the neural network trained by the above method, labeling is performed based on the learning result P. For example, label A can be assigned to terms included in the reference document data RTD and contained in sentences that strongly express the characteristics of the invention, etc. related to the document data TD, and label B can be assigned to other terms. The term data including the labeled terms is referred to as the term data TmD LB (See FIG. 12.)

[0113] Then, the weighting unit 21 weights the term data TmD LB A weight is assigned to each term included in the document data TD. For example, the weight of a term with label A is made larger than the weight of a term with label B. By using the above method, it is possible to weight terms extracted from the document data TD. [Explanation of symbols]

[0114] S01: step, S02: step, S03: step, S04: step, S11: step, S12: step, S13: step, S14: step, S15: step, S16: step, 10: document search system, 11: input unit, 15: memory unit, 17: database, 19: output unit, 20: processing unit, 21: assignment unit, 23: search execution unit, 31: text, 32: text

Claims

[Claim 1] A database and a processing unit are provided. The database has a function of storing database document data, The processing unit has a function of extracting a plurality of terms from claims of a patent application contained in the first document data, the processing unit has a function of classifying a sentence including the words "difference," "not disclosed," or "different" in a written argument in response to a notice of rejection of the patent application contained in the second document data as a first sentence, and classifying the rest as a second sentence; the processing unit has a function of weighting the terms extracted from the first document data so that a weight of the term included in the first sentence is greater than a weight of the term not included in the first sentence; The processing unit has a function of calculating a similarity between the database document data and the first document data based on the weighted terms.

Images