Information search system

Abstract

An information search system (1000) comprises a selection unit (112b), a search unit (112d), a problem extraction unit (112f), and an output unit (112i). The selection unit (112b) selects one or more second objects that are different from yet similar to a first object. The search unit (112d) searches for one or more first patent documents respectively including the one or more second objects. The problem extraction unit (112f) extracts a problem to be solved by the invention according to the first patent document. The output unit (112i) outputs the problem extracted by the problem extraction unit 112f.

Description

Information retrieval system

[0001] This invention relates to an information retrieval system.

[0002] Conventionally, it is known that patent information matching patent information search criteria set based on user input is searched and displayed on the display unit of a client device (Patent Document 1). Patent Document 1 uses publications issued in the past for industrial property rights such as patents as patent information. Furthermore, Patent Document 1 states that the patent information search criteria are, for example, IPC classification, F-term, keywords, applicant or patentee, etc.

[0003] Japanese Patent Publication No. 2019-067330

[0004] For example, a product that a company is trying to develop, or a product that the company has already developed, can be used to address challenges that the company did not initially anticipate during product development.

[0005] However, the technology described in the above-mentioned patent document merely searches for patent information that matches the patent information search criteria. The above-mentioned patent document does not envision searching for new problems that the target product can solve from the patent information.

[0006] In light of the above issues, the purpose of this disclosure is to provide an information retrieval system that can search for new problems that the target product can solve.

[0007] The gist of this disclosure is as follows:

[0008] (1) The information retrieval system comprises a selection unit, a search unit, a problem extraction unit, and an output unit. The selection unit selects one or more second objects that are different from and similar to the first object. The search unit searches for one or more first patent documents that contain each of the one or more second objects. The problem extraction unit extracts the problem that the invention related to the first patent document aims to solve. The output unit outputs the problem extracted by the problem extraction unit.

[0009] (2) In the information retrieval system described in (1) above, the search unit further searches for one or more second patent documents. The one or more second patent documents correspond to the first problem, which is the problem that the invention related to the first patent document extracted by the problem extraction unit aims to solve. The problem extraction unit further extracts a second problem from the second patent documents, which is the problem that the invention related to the second patent document aims to solve. The output unit outputs the second problem.

[0010] (3) The information retrieval system further comprises a classification unit in (1) or (2) above. The classification unit classifies the first patent documents into predetermined groups based on the problems that the invention in the first patent document aims to solve.

[0011] (4) In any of (1) to (3) above, the information retrieval system selects a second object that competes with the first object using a trained model that has been machine-learned based on information from an academic institution or a competitor.

[0012] (5) In the information retrieval system, in (2) above, if the second issue includes the first issue, the issue extraction unit extracts the second issue excluding the first issue.

[0013] (6) The information retrieval system further comprises a solution extraction unit in (2) or (5) above. The solution extraction unit extracts the solution to the problem of the invention relating to the second patent document from the second patent document.

[0014] (7) In the information retrieval system, the search unit further searches for one or more new first patent documents with respect to the above-mentioned problem-solving means as new technical information. The one or more new first patent documents contain a description corresponding to the above-mentioned problem-solving means.

[0015] According to this disclosure, an information retrieval system is provided that allows users to search for new problems that the target product can solve.

[0016] This figure shows an information retrieval system according to one embodiment of the present disclosure. This figure shows an overview of the process for finding new problems that the product can solve. This figure shows another overview of the process for finding new problems that the product can solve. This is a schematic diagram showing the configuration of the search server. This is a schematic diagram showing the configuration of the user's terminal. This is a schematic block diagram showing the functional configuration of the search server. This is a schematic block diagram showing the functional configuration of the user's terminal. This figure shows a screen in which the user inputs technical information. This figure shows an example of a screen in which search results are displayed on the terminal's display device. This figure shows a sequence of processes performed by the search system.

[0017] Several embodiments relating to this disclosure will be described below with reference to the figures. However, these descriptions are intended to be merely illustrative of preferred embodiments of this disclosure and are not intended to limit this disclosure to such specific embodiments. In the following descriptions, similar components will be given the same reference numerals, and redundant descriptions will be omitted where appropriate.

[0018] An information retrieval system 1000 according to one embodiment of the present disclosure includes a search server 100 for retrieving information, as shown in Figure 1. The information retrieval system 1000 may further include a user terminal 200 and a generation AI server 300. The functions of the information retrieval system 1000 may be realized by the search server 100, the terminal 200, and the generation AI server 300. The information retrieval system 1000 may consist only of the search server 100. If it consists only of the search server 100, the search server 100 may have the functions of the terminal 200 or the generation AI server 300. If the information retrieval system 1000 further includes a user terminal 200 and a generation AI server 300 in addition to the search server 100, the search server 100, the terminal 200, and the generation AI server 300 may be able to communicate with each other via a communication network 500 which is composed of communication lines. The communication network 500 relays communication between the search server 100, the terminal 200, and the generation AI server 300. The double arrows shown in Figure 1 indicate that information is sent and received via communication. The generation AI server 300 is configured to execute generation AI processing that generates content by utilizing relationships learned from data. The generation AI can generate content based on prompts given by the user. For example, the generation AI can learn patterns and relationships using pre-collected data sets of conditions and correct answers as training data, and then generate content. A commercial generation AI may be used as the generation AI server 300. Such generation AIs can generate program code, images, videos, audio, etc. For example, the use of generation AIs that generate text data such as ChatGPT is becoming widespread, and examples of commercial text generation AIs include GPT-4 (registered trademark), Llama2, or PaLM2. By using a commercial generation AI, it is not necessary to collect training data individually, and the system can be built simply. The generation AI server 300 may be provided separately from the search server 100, or the functions of the generation AI server 300 may be integrated into the search server 100.

[0019] Incidentally, with respect to a product that a company is planning to develop, or a product that the company has already developed, it is conceivable that the product could be used to solve a new problem other than the one it is intended to solve. In other words, it is conceivable that the product could be used to solve problems that the company did not initially anticipate during product development. In such cases, when the company transmits technical information about the product from terminal 200 as a user, the search server 100 receives the technical information via the communication network 500. The technical information about the product may also be information about similar products that are similar to the product.

[0020] Similar products to the product may be products that share at least one of the following characteristics: structural features of the product, functions that the product can provide, problems that the product solves, and applications of the product. Here, the structural features of the product may also be the chemical formula features of the product. In other words, similar products may be products that share at least one of the one or more attributes of the technical information relating to the product described later. An example of a similar product is a peripheral compound similar to the product.

[0021] The search server 100 can search the literature database to present users with new problems that the product could solve but had not anticipated, along with solutions to those problems. Here, "product" is not limited to goods processed and manufactured for sale, but may be broadly interpreted to include raw materials such as marine products and agricultural products. Furthermore, "product" is not necessarily limited to items currently on sale, but may also include unsold samples intended for future sale.

[0022] In particular, in this embodiment, the technical information relating to the product includes information on one or more attributes. The one or more attributes include, for example, the names of the materials that make up the product. When the terminal 200 receives technical information including these attributes, for example, it transmits the technical information to the search server 100.

[0023] The search server 100 expands the problems that the product solves based on technical information about the product, and seeks new problems that the product can solve. Here, expanding the problems that the product solves and seeking new problems may include raising the conceptual level of the problems, extracting problems similar to the problems that the product solves, or extracting problems that have shifted from the problems that the product solves. The new problems that the product can solve may be problems that the company that developed the product did not originally anticipate being solved by the product. For example, the search server 100 can use patent information of similar products to the product to find new problems that the product can solve.

[0024] Information on similar products to the product is obtained, for example, by the search server 100, which collects data about the product and converts it into a format that can be processed by a machine learning model through morphological analysis and feature extraction. The machine learning model can use features such as chemical properties and physical properties. The search server 100 can then use the product's technical information and the information on similar products as training data to create a trained model that outputs information on similar products when the product's technical information is input. In this way, the search server 100 can obtain information on similar products to the product from the technical information of the product itself.

[0025] Similar products similar to the product include products with similar chemical properties. In this case, the similar product may be any of the following (1), (2), and (3). (1) When the original product is a simple compound, a product in which any of the elements constituting the original product is different. (2) When the original product is combined with other elements and forms a composite compound that forms other compounds. (3) When the original product is a simple compound, a compound having all the elements constituting the original product and further composed of other added elements. Similarly, (1) When the original product is a polymer, a polymer in which any of the monomers constituting the original product is different. (2) When the original product has a functional group that can be grafted onto the polymer, a polymer obtained by adding other chemical substances to the functional group. (3) When the original product is a polymer, a polymer having all the monomers constituting the original product and further composed of other added monomers. The degree of similarity is the highest for (3), followed by (2), and then (1). Similar products similar to the product may include each of (3), (2), and (1). It is more preferable that the similar product similar to the product includes (3) and (2). It is even more preferable that the similar product similar to the product is (3).

[0026] For example, as similar compounds as oxides, aluminum oxide (Al ２ O ３ ), iron oxide (Fe ２ O ３ ) can be mentioned. Also, for example, as oxides with similar structures, calcium oxide (CaO) and zinc oxide (ZnO) can be mentioned as similar compounds. Further, when exemplifying similar compounds as composite oxides containing magnesium oxide, spinel (MgAl ２ O ４ ), magnesium ferrite (MgFe ２ O ４ ), magnesium titanate (MgTiO ３ ), murchite (MgMn ２ O ４ ), magnesium chromate (MgCr ２ O ４ ), magnesium cobalt oxide (MgCo ２ O ４Examples include magnesium hydroxide (Mg(OH)) ３ ), magnesium peroxide (MgO ２ ), hydrotalcite is another example.

[0027] When new problems that a product can solve are identified, the product may be modified based on these new problems to resolve them. For example, if the product is an inorganic oxide, a small modification to its particle size, specific surface area, or BET may solve the identified new problem. The search server 100 presents the user with the new problem, along with solutions that serve as hints for how the product can solve it.

[0028] Figures 2 and 3 show an overview of the process by which the search server 100 seeks new problems that a product can solve based on technical information. Here, we will explain using the example of a case where the developed target product A is a novel substance and we seek new problems that target product A can solve. For example, when the search server 100 receives technical information on target product A, as shown in Figure 2, it lists first substances as similar products to the target product based on the product's technical information. Here, the first substance is a product similar to the target product, and may be, for example, a similar substance from a competitor, or a similar substance, similar compound, or similar article published by an academic institution. In Figure 2, the first substance is shown as a list of first substances: first substance A1, second substance A2, third substance A3, and fourth substance A4. The first substance may be, for example, magnesium oxide, magnesium hydroxide, or hydrotalcite. In particular, if the target product is magnesium oxide, the first substance may be aluminum oxide or boron nitride.

[0029] When listing similar products, the search server 100 may use a trained model that outputs the information of similar products as described above. Alternatively, for example, the search server 100 may send a prompt to the generation AI server 300 to retrieve similar products, and retrieve the listed similar products as a response to the prompt. On the other hand, the user may list the similar products. In this case, the search server 100 retrieves the information of similar products listed by the user and sent from the terminal 200.

[0030] The search server 100 then performs a first-stage search of patent documents. In the first-stage search, the corresponding first patent documents are searched for for each listed first substance. Figure 2 shows how the first-first patent document corresponding to the first first substance A1 has been retrieved. Figure 2 also shows how the second-first patent document corresponding to the second first substance A2 has been retrieved. Figure 2 also shows how the third-first patent document corresponding to the third first substance A3 has been retrieved. Figure 2 also shows how the fourth-first patent document corresponding to the fourth first substance A4 has been retrieved. Patent documents include patent gazettes, patent publications, or equivalents issued by the regulatory authorities of each country, or processed versions thereof. Patent documents may be digitized information. Processed versions may include summarized versions and translated versions.

[0031] Furthermore, when performing a first-stage search for patent documents, the search server 100 may utilize the generation AI server 300. For example, the search server 100 sends a prompt to the generation AI server 300 to extract the first patent document based on information about similar products, and in response to the prompt, obtains the first patent document corresponding to the similar product from the generation AI server 300.

[0032] The search server 100 then extracts the first problem that the invention aims to solve from the acquired first patent document. Furthermore, the search server 100 classifies the first problem that the invention aims to solve and groups similar problems together. In other words, it performs morphological analysis on the problem that the invention aims to solve in the textualized patent document and obtains morphemes. Based on the characteristic words identified based on the obtained morphemes, the problems can be classified into predetermined categories. In addition, a hierarchical structure may be formed in which more similar problems are grouped together. This makes it easier to understand similar problems when displaying them.

[0033] Figure 2 shows how the problem 1-1 that the invention related to Patent Document 1-1 aims to solve and the problem 2-1 that the invention related to Patent Document 2-1 aims to solve are classified into the same first problem group α. Also, Figure 2 shows how the problem 3-1 that the invention related to Patent Document 3-1 aims to solve and the problem 4-1 that the invention related to Patent Document 4-1 aims to solve are classified into the same second problem group β, which is different from the first problem group α. For example, a problem related to moisture resistance and a problem related to the property of being easily decomposed in water can be classified into the same problem group because they both have problems with moisture. Also, a problem related to bending resistance, a problem related to chipping, and a problem related to film fracture can be encompassed by similar problems because they both have problems with mechanical properties. By classifying problems in this way, it is possible to add similar problems that can be conceived from a given problem, for example, "moisture resistance," "resistance to hot water," and "resistance to boiling water" to "water resistance," and use that problem as a starting point to help users think of new "problems to be solved." Therefore, the user can recall improving product A based on the associated problem.

[0034] The search server 100 can then perform a second-stage search of patent documents. In the second-stage search, for each of the first-stage problems (1-1, 2-1, 3-1, and 4-1), the search is performed for second-stage patent documents that describe a second substance that solves these first-stage problems. Figure 3 shows that the first-stage patent document describing the first second substance B1 and the second-stage patent document describing the second second substance B2 were found as second-stage patent documents that can solve the first-stage problem. Similarly, Figure 3 shows that the third-stage patent document describing the third second substance B3 and the fourth second substance B4 were found as second-stage patent documents that can solve the second-stage problem. Furthermore, Figure 3 shows that two patent documents were found that could solve the problem described in Section 4-1: Patent Document 7-2, which describes the seventh second substance B7, and Patent Document 8-2, which describes the eighth second substance B8.

[0035] Then, the search server 100 extracts a second problem that the invention attempts to solve from the second patent document in which the second substance is described. In FIG. 3, the first-2 problems are extracted from the first-2 patent document as the problems that the first second substance B1 attempts to solve, and the state where the second-2 problems are extracted from the second-2 patent document as the problems that the second second substance B2 attempts to solve is shown. Also, in FIG. 3, the state where the third-2 problems are extracted from the third-2 patent document as the problems that the third second substance B3 attempts to solve, and the state where the fourth-2 problems are extracted from the fourth-2 patent document as the problems that the fourth second substance B4 attempts to solve is shown. Also, in FIG. 3, the state where the fifth-2 problems are extracted from the fifth-2 patent document as the problems that the fifth second substance B5 attempts to solve, and the state where the sixth-2 problems are extracted from the sixth-2 patent document as the problems that the sixth second substance B6 attempts to solve is shown. Also, in FIG. 3, the state where the seventh-2 problems are extracted from the seventh-2 patent document as the problems that the seventh second substance B7 attempts to solve, and the state where the eighth-2 problems are extracted from the eighth-2 patent document as the problems that the eighth second substance B8 attempts to solve is shown.

[0036] As described above, the search server 100 searches for the first patent document corresponding to the first substance by the first-stage search, and searches for the second patent document in which the second substance that can solve the first problem that the first substance attempts to solve is described by the second-stage search. Then, the search server 100 extracts a second problem that the invention attempts to solve from the second patent document. By searching for patent documents step by step and extracting problems in this way, there may be cases where problems that the target product A never assumed are found.

[0037] The first problem that the first substance attempts to solve may be the same as the problem that the target product attempts to solve, but may also be different from the problem that the target product attempts to solve. Also, the first problem that the first substance attempts to solve may include a problem different from the problem that the target product attempts to solve in addition to the problem that the target product attempts to solve. In such a case, the target product A may be able to solve the first problem that the first substance attempts to solve, which is different from the problem that the target product attempts to solve.

[0038] Similarly, the second problem that the second substance attempts to solve may be the same as the first problem that the first substance attempts to solve, but it may also be different from the first problem that the first substance attempts to solve. Further, the second problem that the second substance attempts to solve may include a problem different from the first problem that the first substance attempts to solve, in addition to the first problem that the first substance attempts to solve. In such a case, the target product A may be able to solve the second problem that the second substance attempts to solve, which is different from the first problem that the first substance attempts to solve.

[0039] For this reason, the search server 100 provides the user with the first problem that the first substance attempts to solve. Further, the search server 100 may provide the user with a solution to the first problem that the first substance attempts to solve. The search server 100 may also extract the use of the first substance from the first patent document and provide it to the user. Based on the provided information, the user can modify the target product A so that it can solve problems that were not originally assumed to be solved by the target product A. Thereby, the utility value of the target product A is increased.

[0040] Similarly, the search server 100 provides the user with the second problem that the second substance attempts to solve. Further, the search server 100 may provide the user with a solution to the second problem that the second substance attempts to solve. The search server 100 may also extract the use of the second substance from the second patent document and provide it to the user. Based on the provided information, the user can modify the target product A so that it can solve problems that were not originally assumed to be solved by the target product A. Thereby, the utility value of the target product A is increased.

[0041] For example, if product A is magnesium oxide, and aluminum oxide is extracted as the first substance, then a problem that product A did not anticipate may be found as a "problem to be solved" by the invention, such as "improvement of optical properties," in the first problem described in the first patent document describing aluminum oxide, or in the second problem in the second patent document found by further searching from the first patent document. The user can then modify product A to solve the problem of "improvement of optical properties," which product A was not originally intended to solve.

[0042] While Figures 2 and 3 illustrate examples where the target product is a material, compound, or substance, the same principle can be used to identify problems that were not originally anticipated in devices, circuits, or general objects.

[0043] As shown in Figure 4, the search server 100 includes a first control device 110 and a storage device 120. In other words, the search server 100 functions as an information retrieval device.

[0044] The first control device 110 includes a first processor 112, a first memory 114, and a first communication interface 116. The first processor 112, the first memory 114, and the first communication interface 116 are connected via a first communication bus 118. The first processor 112 has one or more CPUs (Central Processing Units) and their peripheral circuits. The first processor 112 may further have other arithmetic circuits such as a logic unit, a numerical unit, or a graphics processing unit. The first memory 114 includes, for example, a volatile semiconductor memory and a non-volatile semiconductor memory. The first communication interface 116 corresponds to the first communication I / F shown in Figure 4 and has an interface circuit for connecting the first control device 110 to a network within the search server 100 or to a communication network 500. The first communication interface 116 is configured to communicate with, for example, a terminal 200 and a generation AI server 300 via the communication network 500.

[0045] The storage device 120 may include, for example, a hard disk drive or an optical recording medium and its access device. In other words, the storage device 120 functions as a storage device. The storage device 120 may function as a patent database. The storage device 120 stores information from patent documents. The storage device 120 may also store non-patent information. Furthermore, the storage device 120 may store synonyms of keywords that are associated with keywords constituting the technical information. The storage device 120 may also store computer programs for executing processes performed on the first processor 112. In addition, the storage device 120 may store various trained models, including trained models that output information on similar products.

[0046] As shown in Figure 5, terminal 200 includes a second control device 210, a display device 220, an input device 230, and a speaker 240. Terminal 200 may be a personal computer or a mobile device. The second control device 210 includes a second processor 212, a second memory 214, and a second communication interface 216. The second processor 212, the second memory 214, and the second communication interface 216 are connected via a second communication bus 218. The second communication interface 216 corresponds to the second communication I / F shown in Figure 5. Terminal 200 may be configured similarly to the configuration of the first processor 112, the first memory 114, and the first communication interface 116 of the search server 100.

[0047] Each function of the information retrieval system 1000 is realized by a functional block executed by the processor. As an example, the first processor 112, as shown in Figure 6, includes an information acquisition unit 112a, a selection unit 112b, a search range specification unit 112c, a search unit 112d, a literature collection generation unit 112e, a problem extraction unit 112f, a solution means extraction unit 112g, a classification unit 112h, an output unit 112i, a prompt creation unit 112j, a prompt transmission unit 112k, and a learning unit 112l.

[0048] Each of these components of the first processor 112 is, for example, a functional module realized by a computer program running on the first processor 112. In other words, each of these components of the first processor 112 consists of the first processor 112 and a program, i.e., software, to make it function. The program may also be recorded in the first memory 114 of the first control device 110 or on a recording medium connected to the outside. Alternatively, each of these components of the first processor 112 may consist of a dedicated arithmetic circuit provided on the first processor 112. The second processor 212 of the terminal 200 shown in Figure 7 is realized by functional blocks executed by the second processor 212, similar to the first processor 112.

[0049] Note that the processor configurations shown in Figures 6 and 7 are examples, and individual components of one processor may be provided in other processors. For example, at least a portion of the functional blocks of the search server 100's processor may be provided in the terminal 200's processor. Also, individual components of one processor may be redundantly provided in other processors.

[0050] The following describes the functions of the information retrieval system 1000, which are realized by the functional blocks provided by the processor.

[0051] First, the functional blocks of the first processor 112 of the search server 100 will be described. The information acquisition unit 112a shown in Figure 6 acquires information transmitted from the terminal 200. Specifically, the information acquisition unit 112a includes a technical information acquisition unit 112a1. The technical information acquisition unit 112a1 acquires technical information transmitted from the terminal 200, which includes at least one or more attributes related to the target product.

[0052] The technical information may consist of only one or more attributes relating to the target product. The one or more attributes may include, for example, "the names of the materials constituting the product" as information about the target product. The attributes of the technical information may further include "the purpose of the product," "the technical field of the product," "the problem the product solves," "the structural features of the product," "the functions the product can provide," "the uses of the product," or "the characteristics of the materials constituting the product." The information acquisition unit 112a may further acquire identification information indicating which of the one or more attributes the technical information corresponds to. The information acquisition unit 112a may also acquire a search continuation command transmitted from the terminal 200. Furthermore, the information acquisition unit 112a may acquire a user's evaluation of the search results transmitted from the terminal 200. If a prompt for searching for similar products is sent to the generation AI server 300 based on the technical information of the target product, the information acquisition unit 112a acquires information on similar products from the generation AI server 300 as a response to the prompt. When a prompt for extracting a first patent document based on information about similar products is sent to the generation AI server 300, the information acquisition unit 112a acquires the first patent document corresponding to the similar product from the generation AI server 300 as a response to the prompt. Also, when a prompt for acquiring new technical information from the original technical information is sent to the generation AI server 300, the information acquisition unit 112a acquires the new technical information generated from the original technical information from the generation AI server 300 as a response to the prompt.

[0053] As described above, the technical information relating to the target product may be information about similar products that are similar to the target product. The information about similar products may be technical information about similar products that includes one or more attributes similar to those of the technical information relating to the target product. For example, the information about similar products corresponds to the information of the first substance A1, the second substance A2, the third substance A3, and the fourth substance A4 shown in Figure 2. The information about similar products may be set by the user themselves. For example, information about similar products that the user has searched for by some method may be input from the input device 230 of the terminal 200 and transmitted to the first communication interface 116 of the search server 100 via the second communication interface 216. The information about similar products transmitted in this manner is acquired by the information acquisition unit 112a.

[0054] The selection unit 112b selects one or more second objects that are different from and similar to the first object. The first object corresponds to the target product described above, and the second object corresponds to a similar product described above. The selection unit 112b may select the second object from the first object using a trained model that outputs information on similar products created by the machine learning described above. In this case, the selection unit 112b may select a second object that competes with the first object using a trained model that has been machine-learned based on information from the internet, academic institutions, or competitors. The selection unit 112b may obtain information from various external databases, the internet, academic institutions, or competitors. Alternatively, the selection unit 112b may select a second object from information on similar products obtained from the generation AI server 300. Furthermore, the selection unit 112b may select a second object from information on similar products entered by the user into the terminal 200.

[0055] The search range specification unit 112c specifies a predetermined area that constitutes the search range of the patent document based on the attributes of the technical information. For example, in the first stage of the search, the search range specification unit 112c may specify "modes for carrying out the invention" and "examples" of the patent document as predetermined areas. Alternatively, for example, in the second stage of the search, the search range specification unit 112c may specify "problems to be solved by the invention" of the patent document as predetermined areas.

[0056] The search range specification unit 112c may specify a predetermined location based on attribute identification information. Furthermore, if the patent document is an overseas document, the predetermined locations of the overseas document that are recognized as corresponding to each of the predetermined locations may be used.

[0057] The search unit 112d searches for one or more first patent documents that contain each of the one or more second objects. Specifically, the search unit 112d searches for one or more first patent documents that contain descriptions corresponding to technical information relating to the target product. The search unit 112d uses a patent database to search for patent documents that contain descriptions that match or approximate the technical information relating to the target product. As described above, the technical information relating to the target product may be information on similar products that are similar to the target product. Therefore, more specifically, in the first stage of the search, the search unit 112d searches for first patent documents that correspond to similar products.

[0058] Furthermore, in the second stage of the search, the search unit 112d further searches for one or more second patent documents containing descriptions corresponding to the first problem extracted by the problem extraction unit 112f. For example, in the examples of Figures 2 and 3, the search unit 112d uses the problem that the first substance aims to solve as technical information and searches for patent documents that contain descriptions that match or approximate the technical information.

[0059] The search unit 112d searches for patent documents containing descriptions that match or approximate the technical information by determining whether such descriptions match or approximate the technical information within the patent documents. The search unit 112d may also search predetermined locations in the patent documents corresponding to individual attributes more intensively than other locations, and determine whether such descriptions match or approximate the technical information within the patent documents. Specifically, the search unit 112d accesses each of the patent documents stored in the storage device 120, searches predetermined locations specified by the search range specification unit 112c within each patent document intensively, and determines whether such descriptions match or approximate the technical information within the patent documents. Alternatively, the search unit 112d may search only predetermined locations and determine whether such locations contain descriptions that match or approximate the technical information. As a patent database, an external patent database connected via the communication network 500 may be used. Examples of patent databases include databases containing patent information from publications issued by regulatory authorities of each country, and databases that compile patent information from published publications. The patent database may be either paid or free.

[0060] For example, in the first stage of the search, when the search unit 112d determines whether the names of the substances constituting similar products match or approximate the descriptions in the patent document, it weights the search for descriptions of "modes for carrying out the invention" or "examples" in the patent document, which are designated as predetermined locations, more heavily than other descriptions in that patent document. As a result, patent documents in which the names of the substances constituting similar products match or approximate the descriptions of "modes for carrying out the invention" or "examples" are determined to contain descriptions that match or approximate the technical information with a higher degree of certainty compared to patent documents in which the names of the substances constituting similar products match or approximate other descriptions.

[0061] Furthermore, for example, in the first stage of the search, when the search unit 112d determines whether the structural features of a similar product or the functions that a similar product can provide match or approximate the description in the patent document, it gives more weight to the search for the description of "modes for carrying out the invention" in the patent document designated as a predetermined location, compared to other descriptions in that patent document. As a result, patent documents in which the structural features of a similar product or the functions that a similar product can provide match or approximate the description of "modes for carrying out the invention" are determined to contain descriptions that match or approximate the technical information with a higher degree of certainty than patent documents in which the structural features of a similar product or the functions that a similar product can provide match or approximate other descriptions.

[0062] Furthermore, for example, in the second stage of the search, when the search unit 112d determines whether the description of the first problem and the description of the patent document match or are similar, it weights the search for the description of "the problem that the invention aims to solve" in the patent document, which is designated as a predetermined location, more heavily than other descriptions in that patent document. As a result, patent documents in which the description of the first problem and the description of "the problem that the invention aims to solve" match or are similar are determined to contain descriptions that match or are similar to the technical information with a higher degree of certainty compared to patent documents in which the description of the first problem matches or is similar to other descriptions.

[0063] The search unit 112d determines whether there are descriptions in the patent document that match or approximate the technical information, for example, by using a full-text search method. Alternatively, the search unit 112d may decompose both the technical information and the descriptions in the patent document into phrases using natural language processing, and then determine whether there are descriptions in the patent document that match or approximate the technical information. Specifically, the search unit 112d may analyze the morphological elements of both the technical information and the descriptions in the patent document, vectorize them, and determine the cosine similarity to determine whether there are descriptions in the patent document that match or approximate the technical information. In this case, the search unit 112c may determine whether there are descriptions in the patent document that match or approximate the technical information by setting a threshold for determining the similarity. Alternatively, for example, the search unit 112d may determine whether there are descriptions in the patent document that approximate the technical information based on similar words of individual keywords that are associated with each keyword constituting the technical information and stored in the storage device 120.

[0064] The search unit 112 may obtain a patent classification from the technical information and search for patent documents using the patent classification. The "patent classification" may include the International Patent Classification, FI, and F-terms. The search unit 112 may perform a search using a search expression or search algorithm that includes the patent classification and the technical information. The search unit 112c may obtain a patent classification based on a table that associates technical information with patent classifications, for example. These tables may be stored in the storage device 120 in advance. Alternatively, the search unit 112c may send a prompt to the generation AI server 300 to obtain a patent classification from the technical information, and obtain a patent classification from the response to the prompt.

[0065] Furthermore, the search unit 112d may acquire new technical information from the original technical information and determine, based on the new technical information, whether or not there is a description in the patent document that is similar to the technical information. The new technical information may be technical information similar to the original technical information, or information with attributes different from those of the original technical information. For example, the search unit 112d may acquire new technical information generated from the original technical information from the generating AI server 300 and determine, based on the acquired new technical information, whether or not there is a description in the patent document that is similar to the technical information. Alternatively, for example, the search unit 112d may acquire new technical information from the original technical information from a trained model and determine, based on the acquired new technical information, whether or not there is a description in the patent document that is similar to the technical information. In other words, the search unit 112d may decompose the technical information and the description in the patent document into phrases using natural language processing, and then extract technical information similar to the technical information from the patent document using machine learning. Specifically, the search unit 112d may, for example, search for different substances that have similar properties to the substance found from the substance's name, based on a trained model created using training data that associates the name of a substance with the properties of that substance, or properties similar to those of that substance. In other words, the description that approximates the technical information in the patent document searched by the search unit 112d may include not only descriptions with similar wording, but also descriptions with similar properties. Thus, the search unit 112d may perform searches not only based on technical information, but also based on new technical information obtained from the technical information.

[0066] The document collection generation unit 112e generates a set of patent documents containing descriptions that match or approximate the technical information, based on the search results of the search unit 112d. Specifically, the document collection generation unit 112e generates a set of patent documents that are determined to contain descriptions that match or approximate the attributes of the technical information. In other words, a patent document that matches the technical information may be a patent document that is determined to contain descriptions that match or approximate the attributes of the technical information. Each individual patent document included in the set of patent documents may include claims, specification, abstract, drawings, bibliographic information, and abstract.

[0067] The problem extraction unit 112f extracts a first problem, which is the problem that the invention related to the first patent document aims to solve, from the first patent document found through the first-stage search. The problem extraction unit 112f further extracts a second problem, which is the problem that the invention related to the second patent document aims to solve, from the second patent document found through the second-stage search. The problem extraction unit 112f extracts the problems that the invention aims to solve from each patent document found by the search unit 112d. Specifically, the problem extraction unit 112f may extract the description of the problem from the "Problem that the invention aims to solve" column of the patent document. The problem extraction unit 112f may extract the description from the "Problem that the invention aims to solve" column of the patent document as is. Alternatively, the problem extraction unit 112f may extract the description of the problem from the "Problem that the invention aims to solve" column of the patent document based on keywords related to the problem. Keywords related to the problem may include "objective," "problem," "difficult," and "room for improvement." The first problem extracted from the first patent document found in the first stage search is provided to the user. Alternatively, the extracted first problem is used in the second stage search. The second problem extracted from the second patent document found in the second stage search is provided to the user.

[0068] Furthermore, if the second issue includes the first issue, the issue extraction unit 112f extracts the second issue excluding the first issue. As described above, the second issue extracted from the second patent document found in the second stage search may include the first issue extracted from the first patent document found in the first stage search. In such cases, the issue extraction unit 112f may extract the second issue from the second patent document found in the second stage search, excluding the first issue extracted from the first patent document found in the first stage search.

[0069] The solution extraction unit 112g extracts the means for solving the problem of the invention related to the searched patent document from that patent document. The solution extraction unit 112g extracts means for solving the problem from each of the searched patent documents. Specifically, the problem extraction unit 112f may extract means for solving the problem from the "Claims," ​​"Means for Solving the Problem," "Modes for Carrying Out the Invention," and "Examples" sections of the patent document. The solution extraction unit 112g may extract the descriptions in the "Claims," ​​"Means for Solving the Problem," "Modes for Carrying Out the Invention," and "Examples" sections of the patent document as they are. The extracted means for solving the problem corresponds, for example, to the second substance shown in Figure 3. In particular, in patent documents relating to inventions of material patents and substance patents, since the names of specific substances are described in the "Examples" section as means for solving the problem, it is preferable for the solution extraction unit 112g to extract substances and materials as means for solving the problem from the descriptions in the "Examples" section.

[0070] The output unit 112i outputs a set of patent documents generated by the document set generation unit 112e. The output unit 112i outputs the issues extracted by the issue extraction unit 112f. The output unit 112i outputs the first or second issue extracted by the issue extraction unit 112f to provide to the user. The output unit 112i may further output the problem-solving means extracted by the solution means extraction unit 112g to provide to the user. Specifically, the output unit 112i transmits the set of patent documents, the issues extracted by the issue extraction unit 112f, and the problem-solving means extracted by the solution means extraction unit 112g to the terminal 200.

[0071] The search unit 112d may further search for one or more new first patent documents containing descriptions corresponding to the problem-solving means extracted from the second patent document by the solution extraction unit 112g as new technical information. The first and second stages of searching may be repeated by repeating this process. The more times the first and second stages of searching are repeated, the higher the probability that the problems extracted from the patent documents will include new problems that the original target product was not intended to solve. Therefore, it is possible to discover problems that the original target product was not intended to solve at all.

[0072] The classification unit 112h classifies patent documents into predetermined groups based on the problems that the inventions in the patent documents aim to solve. As a result, for example, as shown in Figure 2, the first problem is classified into the same first problem group α, which includes the first problem that the invention related to the first patent document aims to solve and the second problem that the invention related to the second patent document aims to solve. The classification unit 112h can also classify the second problem in the same way as the first problem.

[0073] The prompt creation unit 112j creates prompts for searching for similar products based on the technical information of the target product. A prompt refers to instructions or information input to the generating AI. For example, if the name of a substance constituting the target product is substance a, the prompt creation unit 112j creates a prompt such as, "Please tell me about products similar to substance a." The prompt creation unit 112j also creates prompts for obtaining new technical information from the generating AI server 300 based on the original technical information. For example, if the name of a substance constituting the product is substance a, the prompt creation unit 112j creates a prompt such as, "Please tell me about substances similar to substance a in terms of properties, structure, or function." The prompt creation unit 112j also creates prompts for extracting the first patent document based on information about similar products. For example, when extracting the first patent document corresponding to the first substance in Figure 2, the prompt creation unit 112j creates a prompt such as, "Please tell me the number of the patent document corresponding to the first substance."

[0074] The prompt transmission unit 112k sends the prompt created by the prompt creation unit 112j to the generation AI server 300.

[0075] The output unit 112i may output the issues extracted by the issue extraction unit 112f and the problem-solving means extracted by the solution means extraction unit 112g to the prompt creation unit 112j. In this case, the prompt creation unit 112j may create a prompt for modifying the target product based on the technical information of the target product, the issues extracted by the issue extraction unit 112f, and the problem-solving means extracted by the solution means extraction unit 112g. The information acquisition unit 112a may then acquire information regarding modifying the target product from the generation AI server 300 as a response to the prompt. The information regarding modifying the target product may include whether or not it is possible to modify the target product, or how to modify the target product. The output unit 112i may provide the user with the information regarding modifying the target product acquired from the generation AI server 300.

[0076] The learning unit 112l learns an algorithm for when the search unit 112d searches for patent documents, based on the user's evaluation of the set of patent documents. As an example, the learning unit 112l takes technical information about the target product as input and outputs a set of search results excluding patent documents that the user has evaluated as noise. Based on this training data, it trains a neural network to create a trained model. In subsequent searches, by using the trained model, patent documents that the user has evaluated as noise will not be included in the search results, and the user will obtain search results that are more to their liking.

[0077] In a pre-trained search algorithm, the search algorithm desired by the user has been learned through previously performed searches. Therefore, when the search unit 112d performs a search based on the search algorithm, the patent documents desired by the user can be found without generating a literature set, and can be provided to the user as the final output.

[0078] Specifically, when a user attempts multiple searches using the method described above, the learning unit 112l learns the search algorithm. When outputting the final output, the search unit 112d performs the search using the search algorithm learned by the learning unit 112l.

[0079] The learning unit 112l may perform training based on training data, which takes technical information about the target product as input and outputs a set of patent documents obtained from the search results, excluding those evaluated as noise by the user. The training data may also include the patent classification used during the search. When the user tries multiple searches and evaluates the set of patent documents, multiple training data sets are obtained. Since the neural network trained using machine learning based on the training data thus obtained reflects the user's evaluation, when new technical information is input into the neural network, patent documents corresponding to the user's evaluation are output.

[0080] Therefore, the search unit 112d can use the search algorithm learned by the learning unit 112l to search for patent documents based on new technical information. Since the generation of the set of patent documents has already been reflected in the search algorithm through learning, the number of patent documents in the search results is suppressed.

[0081] Furthermore, the learning unit 112l uses information from academic institutions or competitors to train a neural network using technical information about the product and information about similar products as training data, creating a trained model that outputs information about similar products when technical information about the product is input.

[0082] Next, the functional blocks of the second processor 212 of the terminal 200 will be described. As an example, the second processor 212, as shown in Figure 7, includes an input information acquisition unit 212a, a transmission unit 212b, a search result receiving unit 212c, and a display processing unit 212d.

[0083] The input information acquisition unit 212a acquires information entered by the user of the terminal 200 by operating the input device 230. Specifically, the input information acquisition unit 212a acquires the technical information of the target product entered by the user. In the terminal 200, the display processing unit 212d displays the screen on the display device 220 where the user enters the technical information. On the display device 220, one or more attributes of the technical information are displayed as shown in Figure 8, including "Name of the substance constituting the product," "Substances similar to the substance constituting the product," "Structural characteristics of the product," "Functions that the product can provide," "Problems that the product solves," "Applications of the product," "Purpose of the product," and "Technical field of the product." Note that "Substances similar to the substance constituting the product" is an attribute of the technical information of similar products.

[0084] The user can operate the input device 230 while referring to the screen shown in Figure 8 and input the attributes of the technical information corresponding to the target product into the input field 232. The user does not need to input all of the attributes shown in Figure 8; they may input only the attributes that correspond to the target product. The user can also input the attributes of the technical information of similar products other than "substances similar to the substances that make up the product" into the input field 232. In that case, the display processing unit 212d switches "product" to "similar product" in Figure 8 and displays a screen on the display device 220 for inputting the attributes of the technical information of the similar product.

[0085] Furthermore, the input information acquisition unit 212a acquires evaluations of the search results entered by the user.

[0086] The transmission unit 212b transmits the information acquired by the input information acquisition unit 212a to the first communication interface 116 of the search server 100. The transmission unit 212b transmits the information entered in the input field 232 shown in Figure 8 as technical information consisting of one or more attributes to the first communication interface 116 of the search server 100. The transmission unit 212b also transmits identification information to the first communication interface 116 of the search server 100 indicating which of the multiple input fields 232 the technical information was entered into. In other words, the identification information is information indicating which of the one or more attributes the technical information corresponds to. Therefore, the search server 100 can specify a predetermined location which is the search range for patent documents based on the identification information. The transmission unit 212b also transmits a search continuation command to the first communication interface 116 of the search server 100.

[0087] The search result receiving unit 212c receives search results transmitted from the first communication interface 116 of the search server 100. The search result receiving unit 212c receives patent information as search results. The search results include the problem extracted by the problem extraction unit 112f and the problem solving means extracted by the solution means extraction unit 112g.

[0088] The display processing unit 212d processes the search results received from the first communication interface 116 of the search server 100 to display them on the display device 220 of the terminal 200. The search results may also be output as sound by the speaker 240. As an example, Figure 9 shows an example of a screen on the display device 220 of the terminal 200 where the search results are displayed. In the example shown in Figure 9, when the name of the substance constituting the target product is "magnesium oxide" and the similar product is aluminum oxide, the search results show multiple patent documents and the problems that the inventions described in each patent document aim to solve. When the user clicks on the number of a patent document, the screen may transition to a link to the patent document database, and detailed information about that patent document may be displayed. The information or URL of the link displayed on the display device 220 is included in the set of patent documents transmitted from the search server 100 to the terminal 200. The similar product displayed in Figure 9, i.e., aluminum oxide, may be entered by the user in the input field 232 for "substances similar to the substance constituting the product" in Figure 8, or it may be selected by the selection unit 112b from the target product, "magnesium oxide". If a user sees a problem among those displayed that the product was not originally intended to solve, they can consider whether the product can solve that problem. Furthermore, in this process, the user can consider modifying the product, using the solutions displayed alongside the problem as a hint.

[0089] If none of the displayed problems are problems that the target product was not originally intended to solve, the user can operate the input device 230 and click the search continuation button 234 shown in Figure 9. When the search continuation button 234 is clicked, a search continuation command is sent from the terminal 200 to the first communication interface 116 of the search server 100. As a result, the search unit 112d of the search server 100 performs the first stage search again, using the problem-solving means extracted by the solution means extraction unit 112g as new technical information, and searches for patent documents that contain descriptions that match or approximate the technical information.

[0090] On the other hand, if the displayed problems include problems that the target product was not originally intended to solve, and the user does not wish to continue searching, the user can operate the input device 230 and click the search end button 236 shown in Figure 9. When the search end button 236 is clicked, a command to continue searching is not sent from the terminal 200 to the first communication interface 116 of the search server 100, and the search ends.

[0091] A user who views the search results in Figure 9 can evaluate the search results. As an example, Figure 9 shows an example of evaluation when the problems that the inventions of patents A and D aim to solve are useful to the user, while the problems that the inventions of patents B and C aim to solve are not useful to the user. In the example in Figure 9, the user evaluates the problems that the inventions of patents A and D aim to solve as "useful." The user also evaluates the problems that the inventions of patents B and C aim to solve as "noise." These evaluations are input by operating the input device 230 and acquired by the input information acquisition unit 212a. The evaluations are then transmitted by the transmission unit 212b to the first communication interface 116 of the search server 100.

[0092] Next, the time-series flow of processing performed by the information retrieval system 1000 will be explained. The basic sequence of processing performed by the information retrieval system 1000 is shown in Figure 10. First, the input information acquisition unit 212a of the user's terminal 200 acquires technical information about the target product (step S200). Next, the transmission unit 212b of the user's terminal 200 transmits the technical information about the target product to the first communication interface 116 of the search server 100 (step S202).

[0093] Next, the technical information acquisition unit 112a1 of the search server 100 acquires technical information about the target product transmitted from the terminal 200 (step S100). Next, the search range specification unit 112c of the search server 100 specifies a predetermined location in the patent document as the search range for the patent document (step S102). Next, the prompt creation unit 112j of the search server 100 creates a prompt (step S104), and the prompt transmission unit 112k transmits the prompt to the generation AI server 300 (step S106).

[0094] The generation AI server 300 generates information on similar products as a response to the prompt (step S300). The generation AI server 300 then transmits the information on similar products to the first communication interface 116 of the search server 100 (step S302).

[0095] Next, the search unit 112d of the search server 100 performs a first-stage search, focusing on predetermined locations in the patent documents and determining whether there are any descriptions in the patent documents that match or approximate the technical information (step S108). As mentioned above, the search unit 112d may also perform the search based on information obtained from the technical information. Next, the problem extraction unit 112f of the search server 100 extracts the first problem that the invention aims to solve from each of the first patent documents found in the first-stage search (step S110). Next, the search unit 112d performs a second-stage search, using the first problem extracted from the first patent documents found in the first-stage search as technical information, focusing on predetermined locations in the patent documents and determining whether there are any descriptions in the patent documents that match or approximate the technical information (step S112).

[0096] Next, the problem extraction unit 112f of the search server 100 extracts the second problem that the invention aims to solve from each of the second patent documents found in the second stage of the search (step S114). Next, the solution means extraction unit 112g of the search server 100 extracts means for solving the problem from each of the second patent documents found in the second stage of the search (step S116). Next, the output unit 112i of the search server 100 outputs the second problem extracted in step S114 and the problem-solving means extracted in step S116 to the user as search results (step S118). The search results are then transmitted to the user's terminal 200.

[0097] Next, the search result receiving unit 212c of the user's terminal 200 receives the search results transmitted from the first communication interface 116 of the search server 100, and the display processing unit 212d of the user's terminal 200 displays the search results on the screen of the display device 220 (step S204). Next, it is determined whether the user has operated the input device 230 and clicked the search continuation button 234 (step S206). If the user has clicked the search continuation button 234, the transmission unit 212b of the terminal 200 sends a search continuation command to the first communication interface 116 of the search server 100 (step S208).

[0098] When the first communication interface 116 of the search server 100 receives a command to continue the search (step S120), the search server 100 uses the problem-solving means extracted in step S116 as technical information and performs the processing from step S108 onwards again.

[0099] On the other hand, if in step S206 the user does not click the search continuation button 234 but clicks the search end button 236, the transmission unit 212b of the user's terminal 200 transmits the evaluation of the search results entered by the user in the input device 230 to the first communication interface 116 of the search server 100 (step S210).

[0100] Next, the learning unit 112l of the search server 100 learns the algorithm for when the search unit 112d searches for patent documents, based on the evaluation of the search results transmitted from the user's terminal 200 (step S124).

[0101] As explained above, according to this embodiment, when searching for patent documents based on technical information relating to the target product, by focusing on problems and performing a step-by-step search, it is possible to discover problems that the target product had not anticipated but which the target product can solve. Therefore, by modifying the target product based on the discovered problems, it becomes possible to further enhance the utility value of the target product.

[0102] 100 Search server 110 First control device 112 First processor 112a Information acquisition unit 112a1 Technical information acquisition unit 112b Selection unit 112c Search range specification unit 112d Search unit 112e Literature collection generation unit 112f Problem extraction unit 112g Solution means extraction unit 112h Classification unit 112i Output unit 112j Prompt creation unit 112k Prompt transmission unit 112l Learning unit 114 First memory 116 First communication interface 118 First communication bus 120 Storage device 200 Terminal 210 Second control device 212 Second processor 212a Input information acquisition unit 212b Transmission unit 212c Search result receiving unit 212d Display processing unit 214 Second memory 216 Second communication interface 218 Second communication bus 220 Display device 230 Input device 232 Input field 234 Continue search button 236 End search button 240 Speaker 300 Generation AI server 500 Communication network 1000 Information retrieval system

Claims

1. An information retrieval system comprising: a selection unit for selecting one or more second objects that are different from and similar to a first object; a search unit for searching one or more first patent documents that each of the one or more second objects contains; a problem extraction unit for extracting problems that the invention related to the first patent document aims to solve from the first patent document; and an output unit for outputting the problems extracted by the problem extraction unit.

2. The information retrieval system according to claim 1, wherein the search unit further searches for one or more second patent documents corresponding to the first problem which is the problem that the invention related to the first patent document extracts from the problem extraction unit, the problem extraction unit further extracts a second problem which is the problem that the invention related to the second patent document extracts from the second patent documents, and the output unit outputs the second problem.

3. The information retrieval system according to claim 1 or 2, further comprising a classification unit that classifies the first patent documents into predetermined groups based on the problems that the invention in the first patent document aims to solve.

4. The information retrieval system according to any one of claims 1 to 3, wherein the selection unit selects a second object that competes with the first object using a trained model that has been machine-learned based on information from an academic institution or a competitor.

5. The information retrieval system according to claim 2, wherein the problem extraction unit extracts the second problem excluding the first problem if the second problem includes the first problem.

6. The information retrieval system according to claim 2 or 5, further comprising a solution extraction unit for extracting solution solutions from the second patent document.

7. The information retrieval system according to claim 6, wherein the search unit further searches for one or more new first patent documents containing descriptions corresponding to the problem-solving means, with the problem-solving means as new technical information.

Images