Information search program, information search method, and information search device

Abstract

To satisfy a demand for enabling knowledge in a specific language to be searched in a language other than the specific language, while solving the following conventional problem: a difference in search performance may occur in vocabulary between a search query specified in Japanese that native Japanese may use and a search query in translated Japanese, causing a large difference between the respective search results.SOLUTION: A computer executes processing of: in an information search using a trained machine learning model generated by machine learning using training data in a first language, when a search condition is in the first language, translating the search condition to a second language different from the first language; re-translating the search condition translated to the second language to the first language; and performing the information search by inputting the search condition re-translated to the first language to the machine learning model; and when the search condition is in a language different from the first language, translating the search condition to the first language, and performing the information search by inputting the search condition translated to the first language to the machine learning model.SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to an information search program, an information search method, and an information search device. [Background technology]

[0002] To support inquiry operations, there are products that have a knowledge utilization function that can search for knowledge such as documents and FAQs (Frequently Asked Questions) that exist within a company. The products centrally manage all incidents that occur daily, such as inquiries from users, and the history of interactions with users as inquiry information. In addition, to achieve highly accurate knowledge search, the products use a model generated by machine learning, for example, with the search statement as a feature and the knowledge to be displayed for the search statement as the correct answer label.

[0003] In such a product, there is a demand for making it possible to search and utilize knowledge accumulated in a specific language in languages ​​other than the specific language. As a method for satisfying this demand, for example, when searching for a search target in a specific language in another language, there is a method of translating a search query into the specific language using a translation engine, performing a search, translating the search results into the other language, and presenting them to a user. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2020-71865 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the conventional method, there may be a difference in search performance between a search query in the same language as the search target and a search query in a different language. More specifically, for example, in the conventional method, when searching for information in Japanese, there may be a difference in vocabulary between a search query specified in Japanese that sounds like a Japanese person and a search query in Japanese that has been translated, resulting in a large difference in search results.

[0006] In one aspect, an object of the present invention is to improve the accuracy of information retrieval using machine learning models. [Means for solving the problem]

[0007] In the first proposal, in an information search using a trained machine learning model generated by machine learning using training data in a first language, if the search conditions are in the first language, a computer is caused to execute a process of translating the search conditions into a second language different from the first language, retranslating the search conditions translated into the second language into the first language, and inputting the search conditions retranslated into the first language into the machine learning model to perform an information search, and if the search conditions are in a language different from the first language, translating the search conditions into the first language, and inputting the search conditions translated into the first language into the machine learning model to perform an information search. Effect of the Invention

[0008] In one aspect, the accuracy of information retrieval is improved using machine learning models. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing an example of behavioral differences between languages ​​in document retrieval according to a conventional method. [Diagram 2] FIG. 2 is a diagram illustrating an example of an overall configuration of the information processing system according to the first embodiment. [Diagram 3] FIG. 3 is a block diagram of a functional configuration of the information retrieval device 100 according to the first embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of the search target document 121 according to the first embodiment. [Diagram 5] FIG. 5 is a diagram illustrating an example of the replacement document 122 according to the first embodiment. [Figure 6] FIG. 6 is a diagram illustrating an example of the feedback data 123 and the training data 124 according to the first embodiment. [Figure 7] FIG. 7 is a diagram illustrating an example of an overall processing configuration according to the first embodiment. [Figure 8] FIG. 8 is a diagram illustrating an example of a document search according to the first embodiment. [Figure 9] FIG. 9 is a flowchart of the information search process according to the first embodiment. [Figure 10] FIG. 10 is a diagram illustrating an example of a hardware configuration. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the information search program, the information search method, and the information search device disclosed in the present application will be described in detail with reference to the accompanying drawings. Note that the present invention is not limited to these embodiments. In addition, each embodiment can be appropriately combined within a range that does not cause a contradiction.

[0011] First, the problems with the conventional method will be described in more detail with reference to FIG. 1. FIG. 1 is a diagram showing an example of behavioral differences between languages ​​in document searches according to the conventional method. The example in FIG. 1 shows a case where documents stored in Japanese are searched using a search engine. Note that the search engine, including the conventional method, may be one included in a search system via the Web, or may be one of a local search system.

[0012] As shown in Fig. 1, in the conventional method, for example, when performing a document search, an English-speaking user inputs a search query in English to a search system. The input English search query is then translated into Japanese by a translation engine included in the search system. Then, Japanese documents are searched for by a search engine using the search query translated into Japanese.

[0013] On the other hand, for users who speak Japanese, the Japanese search query they entered is used as is to search for Japanese documents. In this case, the Japanese search query may contain Japanese expressions used by native Japanese speakers. As a result, as shown in Figure 1, even when an English user and a Japanese user search for similar documents, there may be differences in the search results due to differences in expressions between the Japanese entered by the Japanese user and the Japanese that has been translated.

[0014] Therefore, the information retrieval program, information retrieval method, and information retrieval device disclosed in this application aim to reduce such differences in search results and improve the accuracy of information retrieval. Note that in the following description, the subject of information retrieval is a document, but the subject of information retrieval may include various information written in a specific language, such as documents and FAQs.

[0015] [Overall configuration example] Fig. 2 is a diagram showing an example of the overall configuration of the information processing system according to the first embodiment. As shown in Fig. 2, the information processing system according to the first embodiment is a system in which an information search device 100 and a user terminal 200 are connected to each other via a network N so as to be able to communicate with each other. The network N may be any of various communication networks, such as the Internet or an intranet, whether wired or wireless. The network N may not be a single network, and may be configured, for example, by connecting an intranet and the Internet via a network device such as a gateway or other device (not shown).

[0016] The information search device 100 is the entity that executes this embodiment including Example 1, and searches for information written in a first language based on a search query input via the user terminal 200, and transmits the search results to the user terminal 200.

[0017] When the search query received from the user terminal 200 is in a language other than the first language, the information search device 100 translates it into the first language and performs information search using the translated search query. On the other hand, when the search query received from the user terminal 200 is in the first language, the information search device 100 translates it into a second language, retranslates it into the first language, and performs information search using the retranslated search query. Here, the first language may be, for example, Japanese, and the second language may be English, but each may be a language other than those. However, as described above, the information search device 100 still searches for information written in the first language using a search query in the first language that has been translated or retranslated.

[0018] In this way, the information retrieval device 100 can search for information in a simple first language by translating or retranslating the search query into a first language, regardless of the language of the search query, and can reduce the difference in search performance between languages. Note that, although the information retrieval device 100 is shown as a single computer in FIG. 2, it may be a distributed computing system composed of multiple computers. Furthermore, the information retrieval device 100 may be a cloud server device managed by a cloud service provider that provides cloud computing services.

[0019] The user terminal 200 is an information processing terminal used by a user to search for information such as documents and FAQs. The user terminal 200 may be a desktop PC (Personal Computer) or a notebook PC installed at the user's workplace or home, or may be a mobile terminal such as a smartphone or a tablet PC.

[0020] A user accesses, for example, a website or an application provided by the information retrieval device 100 via the user terminal 200, inputs a search query, which is a search condition for information retrieval, and obtains search results. Here, the user may be a user from various language areas other than Japanese, and the search query may be input in various languages. In addition, when information retrieval is performed by an application, the application may be downloaded in advance from the information retrieval device 100 and installed in the user terminal 200.

[0021] [Functional configuration of information search device 100] Next, a functional configuration of the information search device 100 shown in Fig. 2 will be described. Fig. 3 is a block diagram showing the functional configuration of the information search device 100 according to the first embodiment. As shown in Fig. 3, the information search device 100 includes a communication unit 110, a storage unit 120, and a control unit 130.

[0022] The communication unit 110 is a processing unit that controls communications with other devices such as the user terminal 200, and is, for example, a communication interface such as a network interface card.

[0023] The storage unit 120 is an example of a storage device, such as a memory or a hard disk, that stores various data and programs executed by the control unit 130. The storage unit 120 stores a search target document 121, a replacement document 122, feedback data 123, training data 124, and a search model 125.

[0024] The search target document 121 stores, for example, information related to documents, FAQs, etc. that are the subject of information search. Note that the search target documents, etc. are written in a first language, for example, Japanese.

[0025] Fig. 4 is a diagram showing an example of the search target document 121 according to the embodiment 1. As shown in Fig. 4, the search target document 121 stores "ID, title, text, language" and the like in association with each other.

[0026] The "ID" stored in the search target document 121 is, for example, an identifier that uniquely identifies the document to be searched. The "title" and "text" are the title and text of the document to be searched, respectively. The "language" is an abbreviation that indicates the first language in which the document to be searched is written, for example, Ja for Japanese and En for English.

[0027] The replacement document 122 stores, for example, information about a replacement document to be replaced with the search target document stored in the search target document 121 and output to the user terminal 200. The replacement document is written in a second language, for example, English. The replacement document is a document to be presented to a user who wishes to obtain a document in the second language rather than the first language, instead of or together with the search target document.

[0028] Fig. 5 is a diagram illustrating an example of the replacement document 122 according to the embodiment 1. As illustrated in Fig. 5, the replacement document 122 stores "ID, title, text, language, corresponding ID" and the like in association with each other.

[0029] The "ID" stored in the replacement document 122 is, for example, an identifier that uniquely identifies the replacement document. The "title" and "body" are the title and body of the replacement document, respectively. The "language" is, for example, an abbreviation indicating the second language in which the replacement document is written. The "corresponding ID" is an identifier of the corresponding document to be searched, for example, the "ID" of the search target document 121, which allows the search target document 121 and the replacement document 122 to be linked.

[0030] The feedback data 123 stores, for example, information on feedback from the user regarding the document to be searched or the document for replacement presented to the user. The feedback data 123 is data generated, for example, by the information retrieval device 100 confirming with the user via the user terminal 200 whether the presented document was correct or useful.

[0031] Fig. 6 is a diagram illustrating an example of the feedback data 123 and the training data 124 according to the first embodiment. As illustrated in Fig. 6, the feedback data 123 stores "a search query, a translated / retranslated query, a correctly hit document," and the like in association with each other.

[0032] The "search query" stored in the feedback data 123 is a search condition input by a user, and may be, for example, a search query or a search word. The "search query" may be input in various languages, not just the first language. The "query after translation / retranslation" is a search query or a search word obtained by translating or retranslating the "search query" into the first language. If the "search query" is in a language other than the first language, it is translated into the first language, and if the "search query" is in the first language, it is translated into a language other than the first language and then retranslated into the first language to generate the "query after translation / retranslation." The "document that was correctly hit" is, for example, an identifier of a document indicated by the user as being correct or useful for the search query. Since the "document that was correctly hit" is an identifier of a document presented to the user, it may be an identifier of a document to be searched or a document to be replaced, which is output to the user terminal 200. The "document that was correctly hit" may be an answer to the correctness or incorrectness of the presented document.

[0033] The training data 124 is training data for a machine learning model that is trained using, for example, a translated or retranslated search query in a first language as a feature and a document in the first language output as a result of information search for the search query as a correct answer label. The training data 124 is training data written in the first language.

[0034] As shown in FIG. 6, the training data 124 stores "search statements, documents to be displayed at the top" in association with each other. The "search statements" stored in the training data 124 are search queries in a first language that have been translated or retranslated. The "documents to be displayed at the top" are identifiers of documents that have been indicated by the user as being correct or useful for the search query, for example. The "documents to be displayed at the top" are identifiers of documents to be searched that are written in the first language. Therefore, as shown in FIG. 6, when the "documents that were correctly hit" in the feedback data 123 are in a language other than the first language, they are converted into the identifiers of the corresponding documents in the first language and stored in the "documents to be displayed at the top".

[0035] The search model 125 is a machine learning model that is trained and generated by machine learning using the training data 124 written in the first language. The search model 125 can store model parameters for constructing the machine learning model.

[0036] It should be noted that various information other than the above information can be stored in the storage unit 120. Furthermore, the data configuration of each piece of information is not limited to the above.

[0037] The control unit 130 is a processing unit, such as a processor, that controls the entire information search device 100. The control unit 130 includes a search unit 131, a translation unit 132, and a machine learning unit 133. Each processing unit is an example of an electronic circuit included in a processor or an example of a process executed by a processor.

[0038] In an information search using a trained machine learning model generated by machine learning using training data in a first language, the search unit 131 inputs search conditions, which are search queries input by a user, to the machine learning model to perform information search. Here, when the search conditions are in the first language, the search unit 131 inputs the search conditions translated by the translation unit 132 into a language other than the first language and then retranslated into the first language to the machine learning model to perform information search. On the other hand, when the search conditions are in a language different from the first language, the search unit 131 inputs the search conditions translated into the first language by the translation unit 132 to the machine learning model to perform information search.

[0039] When the search conditions are in a first language, the translation unit 132 translates the search conditions into a second language different from the first language, and retranslates the search conditions translated into the second language into the first language. On the other hand, when the search conditions are in a language different from the first language, the translation unit 132 translates the search conditions into the first language.

[0040] Furthermore, for a user who wishes to obtain a document in a second language different from the first language as a result of an information search, the translation unit 132 translates the document into the second language and transmits it to the information processing terminal that requested the information search, i.e., the user terminal 200. Note that the document may be translated into the second language in advance, rather than when an information search is requested. This can improve search response compared to translating the document each time an information search is performed.

[0041] The machine learning unit 133 generates training data in the first language using the search conditions translated or retranslated into the first language by the translation unit 132. More specifically, the machine learning unit 133 generates training data in which the search conditions in the first language are treated as features and the data in the first language output as a result of an information search for the search conditions are treated as correct answer labels. Then, the machine learning unit 133 trains or retrains a machine learning model using the generated training data.

[0042] [Function Overview] Next, an overall information search process according to a first embodiment for reducing the difference in search results between languages ​​of search conditions and improving the accuracy of information search will be described. Fig. 7 is a diagram showing an example of an overall processing configuration according to the first embodiment. The example in Fig. 7 shows a case where a search query is input in a first language to search for documents in the first language.

[0043] 7, the machine learning unit 133 performs model learning using the search query retranslated into the first language as a feature and the document in the first language output as a result of information search for the search query as a correct answer label (step S1). As a result, the search model 125 is generated. The translation of the search query is performed by the translation unit 132.

[0044] Here, the search query retranslated into the first language may be a search query of the feedback data, or may be a manually generated search query in the first language that is translated into an arbitrary second language different from the first language and retranslated into the first language. The retranslated search query is stored in the training data 124. In addition, the document in the first language that is to be the correct label may be an identifier that uniquely identifies the document.

[0045] Next, when the search unit 131 receives the search query in the first language input by the user terminal 200, it transmits it to the translation unit 132. As a result, the search query in the first language is translated by the translation unit 132 into an arbitrary second language different from the first language, and is retranslated into the first language (step S2). Note that the search unit 131 can receive language information of the search query from the user terminal 200 together with the search query in the first language or separately. The language information of the search query is data indicating in which language the search query is written, and is specified by the user via the user terminal 200, for example. As a result, the information search device 100 can recognize in which language the search query is written, but the information search device 100 may determine the language in which the search query is written by an existing technology such as language analysis, for example.

[0046] Next, the search unit 131 inputs the search query retranslated into the first language to the search model 125, and obtains documents in the first language as search results from the search target documents 121 using the identifiers of the documents output (step S3).

[0047] Here, the re-translation of the search query in step S2 and the document search in step S3 will be described with a concrete example. Fig. 8 is a diagram showing an example of document search according to the first embodiment. Fig. 8 shows an example in which the first language is Japanese and the second language is English.

[0048] As shown in FIG. 8, a Japanese search query inputted through the user terminal 200 is translated into English. Note that, as shown in FIG. 8, the Japanese search query inputted through the user terminal 200 may include Japanese expressions used by native Japanese speakers. Next, the search query translated into English is retranslated into Japanese. In this way, the information retrieval device 100 can convert the search query into a simple expression by translating the search query into another language and then retranslating it into the original language. This allows the information retrieval device 100 to search for information in a first language with a simple expression, and can reduce the difference in search results caused by the difference between the expression in the first language as it is inputted and the expression in the first language after translation, as described with reference to FIG. 1.

[0049] Next, when the language information of the search result specified by the user using the user terminal 200 is a language other than the first language, the search unit 131 obtains a document translated into the specified language corresponding to the document of the search result (step S4). The translated document obtained here is a document that is obtained by translating each document in the first language stored in the search target document 121 into various languages ​​by the translation unit 132 and stored in the replacement document 122. When the language information of the search result specified by the user is the first language or there is no particular specification, step S4 may not be executed. The document of the search result is transmitted to the user terminal 200 by the search unit 131.

[0050] Next, if the received search result document is appropriate, such as correct or useful, the user terminal 200 transmits an identifier of the appropriate document to the information retrieval device 100 as user feedback data 123. The identifier of the appropriate document transmitted here is the identifier of the document presented to the user as the search result. Therefore, if a document translated into a language other than the first language is presented in step S4, the identifier of the appropriate document is the identifier of the translated document.

[0051] Next, if the document presented to the user is in a language other than the first language, the search unit 131 converts the appropriate document identifier received from the user terminal 200 into a corresponding base language identifier (step S5). Here, the base language identifier is the identifier of the document in the first language stored in the search target documents 121. Note that if the document presented to the user is in the first language, step S5 does not need to be executed.

[0052] Next, the search unit 131 adds the search query retranslated into the first language and the appropriate document identifiers to the training data 124 in order to retrain the search model 125 (step S6). As a result, the machine learning unit 133 retrains the search model 125 using the added training data 124 (step S1). In this way, the search model is generated, documents are searched using the generated search model, and the search model is retrained using the search results.

[0053] [Processing flow] Next, the information search process according to the first embodiment will be described in more detail along the flow. Fig. 9 is a flowchart showing the flow of the information search process according to the first embodiment. The information search process described in Fig. 9 starts, for example, when a user inputs a search query into an input form for information search provided to the user terminal 200 by the information search device 100.

[0054] 9, a user inputs a search query for information search via the user terminal 200 (step S101). In addition, language information of the search query or search results may be input together with or separately from the search query.

[0055] Next, the user terminal 200 transmits the search query input in step S101 to the information retrieval device 100 (step S102). Note that, if language information of the search query or the search results has been input, it is transmitted to the information retrieval device 100 together with the search query or separately.

[0056] Next, if the information retrieval device 100 has not received a search query from the user terminal 200 (step S103: No), it waits for reception of the search query. On the other hand, if the information retrieval device 100 has received a search query (step S103: Yes), it determines whether the received search query is in the first language (step S104). This determination may be made based on language information of the search query input by the user, or may be made by determining the language in which the search query is written by linguistic analysis or the like.

[0057] If the search query is in the first language (step S104: Yes), the information retrieval device 100 translates the search query into a second language different from the first language (step S105). On the other hand, if the search query is in a language other than the first language (step S104: No), step S105 is not executed and the process proceeds to step S106.

[0058] Next, the information retrieval device 100 translates the search query translated into the second language in step S105 or the search query in a language other than the first language received from the user terminal 200 into the first language (step S106).

[0059] Next, the information retrieval device 100 executes an information search using the search query translated into the first language in step S106 (step S107). This searches for documents, etc., but if the search does not return any hits, for example, the device may notify the user terminal 200 of this and prompt the user to re-enter the search query.

[0060] Next, if the search results from the information search in step S107 need to be translated (step S108: Yes), information retrieval device 100 replaces the search results with information in the language specified by the user (step S109). A case in which the search results need to be translated is, for example, when the language information of the search results is specified by the user. The information replaced here is, for example, a document in a first language that is the search result, which has been translated in advance into a language other than the first language. On the other hand, if the search results do not need to be translated (step S108: No), step S109 is not executed and the process proceeds to step S110.

[0061] Next, information retrieval device 100 transmits the information replaced in step S109 as the search result, or, if step S109 has not been executed, the search result in step S107 to user terminal 200 (step S110).

[0062] Next, if the user terminal 200 has not received the search results from the information retrieval device 100 (step S111: No), it waits for the search results to be received. On the other hand, if the user terminal 200 has received the search results (step S111: Yes), the user terminal 200 displays the search results on the display of the user terminal 200 or the like (step S112). After executing step S112, the information retrieval process shown in FIG. 9 ends.

[0063] [effect] As described above, in an information search using a trained machine learning model generated by machine learning using training data in a first language, the information search program causes information search device 100 to execute a process of translating the search conditions into a second language different from the first language, retranslating the search conditions translated into the second language into the first language, and inputting the search conditions retranslated into the first language into the machine learning model to perform an information search if the search conditions are in a language different from the first language, and translating the search conditions into the first language, and inputting the search conditions translated into the first language into the machine learning model to perform an information search if the search conditions are in a language different from the first language.

[0064] As a result, the information retrieval device 100 can search for information in a simple first language by translating or retranslating the search condition into a search query in the first language, regardless of the language of the search condition, and can reduce the difference in search performance between languages. In this way, the information retrieval device 100 can improve the accuracy of information retrieval using a machine learning model.

[0065] In addition, the information retrieval program further causes the information retrieval device 100 to execute a process of generating training data using the search conditions translated or retranslated into the first language, and retraining the machine learning model using the generated training data.

[0066] This enables the information retrieval device 100 to improve the accuracy of information retrieval using a machine learning model.

[0067] In addition, the process of generating training data executed by the information search device 100 includes a process of generating training data in which search conditions in a first language are used as features and data in the first language output as a result of an information search for the search conditions is used as a correct answer label.

[0068] This enables the information retrieval device 100 to improve the accuracy of information retrieval using a machine learning model.

[0069] The information search program also causes the information search device 100 to execute a process of transmitting the information search result, translated into a language different from the first language, to the information processing terminal that has requested the information search.

[0070] This enables the information retrieval device 100 to provide information in a language desired by the user while improving the accuracy of information retrieval using a machine learning model.

[0071] In addition, the process of transmitting the results of the information search to the information processing terminal, which is executed by the information search device 100, includes a process of transmitting data that has been pre-translated into a language different from the first language and that corresponds to the results of the information search in the first language, to the information processing terminal as the results of the information search.

[0072] This allows the information retrieval device 100 to provide information in a language desired by the user, while improving search response compared to translating the information each time an information search is performed.

[0073] [system] The information including the processing procedures, control procedures, specific names, various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. In addition, the specific examples, distributions, values, etc. described in the embodiments are merely examples and can be changed as desired.

[0074] Moreover, each component of each device shown in the figure is a functional concept, and does not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure. In other words, all or part of them can be functionally or physically distributed and integrated in any unit according to various loads, usage conditions, etc. For example, the search unit 131 and the translation unit 132 of the information search device 100 can be integrated.

[0075] Furthermore, each processing function performed by each device may be realized, in whole or in part, by a CPU and a program analyzed and executed by the CPU, or may be realized as hardware using wired logic.

[0076] [Hardware] The hardware configuration of the above-mentioned information search device 100 will be described. Fig. 10 is a diagram showing an example of the hardware configuration. As shown in Fig. 10, the information search device 100 has a communication unit 100a, a HDD (Hard Disk Drive) 100b, a memory 100c, and a processor 100d. Moreover, each unit shown in Fig. 10 is connected to each other via a bus or the like.

[0077] The communication unit 100a is a network interface card or the like, and communicates with other servers. The HDD 100b stores programs and databases (DBs) that operate the functions shown in FIG.

[0078] The processor 100d reads out a program that executes the same processes as the respective processing units shown in Fig. 3 from the HDD 100b or the like and loads it in the memory 100c, thereby operating a process that executes each function described in Fig. 3. For example, this process executes a function similar to that of each processing unit of the information search device 100. Specifically, for example, the processor 100d reads out a program having a function similar to that of the search unit 131, the translation unit 132, or the like from the HDD 100b or the like. Then, the processor 100d executes a process that executes a process similar to that of the search unit 131, the translation unit 132, or the like.

[0079] In this way, the information retrieval device 100 operates as an information processing device that executes each process by reading and executing a program. The information retrieval device 100 can also realize functions similar to those of the above-mentioned embodiment by reading the program from a recording medium using a medium reading device and executing the read program. Note that the program in these other embodiments is not limited to being executed by the information retrieval device 100. For example, the present invention can be similarly applied to cases where another computer or server executes a program, or where these execute a program in cooperation with each other.

[0080] This program can be distributed via a network such as the Internet. This program can also be recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, a magneto-optical disk (MO), or a digital versatile disk (DVD), and can be executed by being read out from the recording medium by a computer.

[0081] In addition, the following supplementary notes are disclosed regarding the embodiments including the above-mentioned Example 1.

[0082] (Appendix 1) In information retrieval using a trained machine learning model generated by machine learning using training data in the first language, If the search criteria is in the first language, translating the search terms into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search terms into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval program that causes a computer to execute a process.

[0083] (Appendix 2) Generating the training data using the search terms translated or retranslated into the first language; Retraining the machine learning model using the generated training data 2. The information retrieval program according to claim 1, further causing the computer to execute a process.

[0084] (Appendix 3) The information retrieval program according to appendix 1 or 2, characterized in that the process of generating the training data includes a process of generating the training data in which the search conditions in the first language are treated as features, and the data in the first language output as a result of the information retrieval for the search conditions is treated as a correct answer label.

[0085] (Appendix 4) An information search program as described in any one of Appendices 1 to 3, characterized in that the computer is further caused to execute a process of transmitting the results of the information search translated into a language different from the first language to the information processing terminal that requested the information search.

[0086] (Appendix 5) The information search program according to appendix 4, characterized in that the process of transmitting the results of the information search to the information processing terminal includes a process of transmitting data that has been pre-translated into a language different from the first language and that corresponds to the results of the information search in the first language to the information processing terminal as the results of the information search.

[0087] (Appendix 6) In information retrieval using a trained machine learning model generated by machine learning using training data in a first language, If the search criteria is in the first language, translating the search terms into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search terms into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval method characterized in that the processing is executed by a computer.

[0088] (Appendix 7) Generating the training data using the search terms translated or retranslated into the first language; Retraining the machine learning model using the generated training data 7. The information retrieval method according to claim 6, further comprising the steps of:

[0089] (Appendix 8) The information retrieval method according to appendix 6 or 7, characterized in that the process of generating the training data includes a process of generating the training data in which the search conditions in the first language are treated as features, and the data in the first language output as a result of the information retrieval for the search conditions is treated as a correct answer label.

[0090] (Appendix 9) An information search method as described in any one of Appendices 6 to 8, characterized in that the computer further executes a process of transmitting the results of the information search translated into a language different from the first language to the information processing terminal that requested the information search.

[0091] (Appendix 10) The information search method described in Appendix 9, characterized in that the process of transmitting the results of the information search to the information processing terminal includes a process of transmitting data that has been pre-translated into a language different from the first language and that corresponds to the results of the information search in the first language to the information processing terminal as the results of the information search.

[0092] (Appendix 11) In information retrieval using a trained machine learning model generated by machine learning using training data in a first language, If the search criteria is in the first language, translating the search terms into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search terms into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval device comprising a control unit.

[0093] (Appendix 12) Generating the training data using the search terms translated or retranslated into the first language; Retraining the machine learning model using the generated training data 12. The information retrieval device according to claim 11, further comprising the control unit executing a process.

[0094] (Appendix 13) The information retrieval device according to appendix 11 or 12, characterized in that the process of generating the training data includes a process of generating the training data in which the search conditions in the first language are treated as features, and the data in the first language output as a result of the information search for the search conditions is treated as a correct answer label.

[0095] (Appendix 14) An information search device as described in any one of Appendices 11 to 13, characterized in that the control unit is further made to execute a process of transmitting the results of the information search translated into a language different from the first language to the information processing terminal that requested the information search.

[0096] (Appendix 15) The information search device described in Appendix 14, characterized in that the process of transmitting the results of the information search to the information processing terminal includes a process of transmitting data that has been pre-translated into a language different from the first language and that corresponds to the results of the information search in the first language to the information processing terminal as the result of the information search.

[0097] (Appendix 16) A processor; a memory operatively connected to said processor; An information retrieval device comprising: In information retrieval using a trained machine learning model generated by machine learning using training data in a first language, If the search criteria is in the first language, translating the search terms into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search terms into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval device that executes processing. [Explanation of symbols]

[0098] 100 Information retrieval device 100a Communications Department 100b HDD 100c Memory 100d Processor 110 Communications Department 120 Storage section 121 searchable documents 122 Replacement Documents 123 Feedback Data 124 Training Data 125 Search Models 130 Control section 131 Search Department 132 Translation Department 133 Machine Learning Department 200 User terminals

Claims

1. In information retrieval using a trained machine learning model generated by machine learning using training data in a first language, If the search criteria is in the first language, translating the search criteria into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search criteria into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval program that causes a computer to execute a process.

2. generating the training data using the search terms translated or retranslated into the first language; Retraining the machine learning model using the generated training data 2. The information retrieval program according to claim 1, further comprising instructions to cause the computer to execute a process.

3. The information retrieval program according to claim 1 or 2, characterized in that the process of generating the training data includes a process of generating the training data in which the search conditions in the first language are treated as features and the data in the first language output as a result of the information search for the search conditions are treated as correct answer labels.

4. The information search program according to any one of claims 1 to 3, further comprising causing the computer to execute a process of transmitting the results of the information search translated into a language different from the first language to the information processing terminal that requested the information search.

5. The information search program according to claim 4, characterized in that the process of transmitting the results of the information search to the information processing terminal includes a process of transmitting data that has been pre-translated into a language different from the first language and that corresponds to the results of the information search in the first language to the information processing terminal as the results of the information search.

6. In information retrieval using a trained machine learning model generated by machine learning using training data in a first language, If the search criteria is in the first language, translating the search criteria into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search criteria into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval method characterized in that the processing is executed by a computer.

7. In information retrieval using a trained machine learning model generated by machine learning using training data in a first language, If the search criteria is in the first language, translating the search criteria into a second language different from the first language; retranslating the search terms translated into the second language into the first language; inputting the search terms retranslated into the first language into the machine learning model to perform the information search; If the search criteria is in a language different from the first language, translating the search criteria into the first language; The search criteria translated into the first language are input to the machine learning model to perform the information search. An information retrieval device comprising a control unit.

Images