Text analysis device

Abstract

In order to analyze the content of a text in a free description style, a text analysis device 10 includes: an input reception unit 100 that receives an input of a text in a free description style; a shaping unit 200 that shapes the text received by the input reception unit 100 into a designated format; a tag generation unit 300 that generates a tag on the basis of the text shaped by the shaping unit 200; and an allocation unit 400 that allocates the tag generated by the tag generation unit 300 to the text received by the input reception unit 100 or to the text shaped by the shaping unit 200.

Description

Text analysis device

[0001] The present invention relates to a text analysis device.

[0002] The following technique is known as a technique for classifying multiple documents into documents with similar content: Specifically, this technique breaks down each document into words, tallying up the broken down words and combinations of broken down words, and classifying the documents into documents with similar content based on the tallying up results (see, for example, Patent Document 1).

[0003] JP 2024-28279 A

[0004] This technology can classify multiple documents into documents with similar content, but it has the problem that it cannot analyze the content of a single document. The present invention has been made in consideration of these circumstances, and its purpose is to provide a technology that processes documents collectively and associates with each document (text) a summary that summarizes the semantic content of the text and outputs the summary.

[0005] In order to achieve the above-mentioned object, a text analysis device according to one aspect of the present invention includes an input accepting unit that accepts input of free-form text, a formatting unit that formats the text accepted by the input accepting unit into a specified format, a tag generating unit that generates tags based on the text formatted by the formatting unit, and an assignment unit that assigns the tags generated by the tag generating unit to the text accepted by the input accepting unit.

[0006] 1 is a diagram showing the overall configuration of a system including a text analysis device according to an embodiment; FIG. 2 is a block diagram showing the configuration of an input reception unit in the text analysis device; FIG. 3 is a block diagram showing the configuration of a shaping unit in the text analysis device; FIG. 4 is a block diagram showing the configuration of a tag generation unit in the text analysis device; FIG. 5 is a block diagram showing the configuration of an assignment unit in the text analysis device; FIG. 6 is a diagram showing an operation sequence of the system; FIG. 7 is a diagram for explaining an example of a specific operation of the text analysis device; FIG. 8 is a diagram for explaining an example of the operation of a conversion unit and a clustering unit in the text analysis device; FIG. 9 is a diagram for explaining an example of the operation of tag generation and correction in the text analysis device; FIG. 10 is a block diagram showing the configuration of an input reception unit in an application example;

[0007] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that in each drawing, the dimensions and scale of each part are appropriately different from those of the actual parts. Furthermore, the embodiments described below are preferred examples of the present invention, and therefore various technically preferable limitations are applied. However, the scope of the present invention is not limited to these embodiments unless otherwise specified in the following description to the effect that the present invention is limited.

[0008] 1 is a block diagram showing the overall configuration of a system 1 including a text analysis device 10 according to an embodiment. As shown in this figure, the system 1 is configured such that the text analysis device 10 and terminal devices 60-1, ..., 60-n are connected to each other via a network 50 so as to be able to communicate with each other.

[0009] Terminal devices 60-1, ..., 60-n are information terminal devices, specifically, smartphones, personal computers, etc. Note that, although the figure shows an example in which multiple n terminal devices 60-1, ..., 60-n are provided, there is no particular need for multiple devices. In the following description, when a terminal device is not specified, the part after the hyphen is omitted and the reference number is simply "60."

[0010] The system 1 will now be outlined. The operator of the terminal device 60 is, for example, a survey participant who requests the system 1 to analyze multiple documents collected through the survey. In other words, the analysis targets multiple documents. The analysis targets not only survey responses but also various requests and opinions. The documents are free-form text. Free-form text refers to responses, requests, opinions, and other such responses written in a free format, rather than through a selection of options. In other words, free-form text refers to a series of meaningful letters, numbers, and symbols. Free-form text may contain redundant content or template-based fixed responses. Template-based responses may contain, for example, symbols with special meanings used to separate strings of characters inserted into fixed sentences. The inserted strings may contain meaningless characters or meaningful characters. Furthermore, the writing style of free-form text may include both formal and colloquial language.

[0011] A requester of analysis, who is an operator of terminal device 60, transmits multiple documents to be analyzed from terminal device 60 via network 50 to text analysis device 10. Text analysis device 10 formats the multiple documents transmitted from terminal device 60, i.e., the free-form text, classifies (groups) them into a specified number of classes, summarizes each class, and generates tags. Of the generated tags, text analysis device 10 associates one or more tags that are close in meaning to the free-form text and outputs them.

[0012] The text analysis device 10 includes an input receiving unit 100, a shaping unit 200, a tag generation unit 300, and an assignment unit 400. The input receiving unit 100 receives free-form text input transmitted from a terminal device 60 and instructs the shaping unit 200, the tag generation unit 300, and the assignment unit 400 to perform various processes. The shaping unit 200 formats the free-form text received by the input receiving unit 100 into a specified format and outputs it. The tag generation unit 300 classifies the formatted text into a specified number of classes, summarizes each class, and generates tags. The assignment unit 400 assigns tags output from the tag generation unit 300 to the free-form text received by the input receiving unit 100. In short, in this embodiment, the free-form text is assigned a tag that serves as a summary of the class to which the text belonged. The text to which the tag is assigned may not be free-form text received by the input receiving unit, but may be text that has been formatted by the formatting unit 200 .

[0013] In this embodiment, the input receiving unit 100, the shaping unit 200, the tag generating unit 300, and the allocation unit 400 are combined to form a single logical server, the text analysis device 10. However, the text analysis device 10 may be formed by a single server, or a logical text analysis device 10 may be formed by using cloud computing.

[0014] 2 is a block diagram showing the configuration of the input receiving unit 100. The input receiving unit 100 includes a control device 110, a storage device 120, a display device 130, an operation device 140, and a communication device 150.

[0015] The storage device 120 is a recording medium readable by the control device 110. It includes a volatile memory such as a random access memory (RAM) that functions as a working area for the storage device 120, and a non-volatile memory such as an electrically erasable programmable read-only memory (EEPROM) that stores various information, and stores a control program 122. The control program 122 includes both an operation system program that the control device 110 uses to control each component of the input receiving unit 100, and an application program for performing text analysis.

[0016] The control device 110 includes a processor. The processor provided in the control device 110 includes, for example, one or more central processing units (CPUs). Note that the processor provided in the control device 110 may include hardware such as a graphics processing unit (GPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA) in addition to or in place of one or more CPUs. The control device 110 executes application programs stored in the storage device 120 to accept input of free-form text, transfer data between the shaping unit 200, the tag generation unit 300, and the allocation unit 400, and instruct the execution of various processes.

[0017] The display device 130 displays various screens. The operation device 140 is a keyboard and pointing device that is operated by an operator. The communication device 150 is hardware for communicating with the terminal device 60, the shaping unit 200, the tag generation unit 300, or the allocation unit 400 via the network 50.

[0018] 3 is a block diagram showing the configuration of the shaping unit 200. The shaping unit 200 includes a control unit 210, a storage unit 220, and a communication unit 250.

[0019] Like storage device 120, storage device 220 is a recording medium readable by control device 210. A control program 222 is stored in storage device 220. Control program 222 includes both an operation system program that allows control device 210 to control each unit of formatting unit 200, and an application program for formatting text.

[0020] Similar to the control device 110, the control device 210 constructs a first large-scale language model 216 by executing an application program stored in the storage device 220. The first large-scale language model 216 formats the free-form text transferred from the input receiving unit 100 in accordance with specified requests. The control device 210 also issues instructions for transferring data, executing various processes, etc.

[0021] The communication device 250 is hardware for communicating with the input receiving unit 100 , the tag generating unit 300 , or the allocating unit 400 via the network 50 .

[0022] 4 is a block diagram showing the configuration of tag generation unit 300. Tag generation unit 300 includes a control unit 310, a storage unit 320, and a communication unit 350.

[0023] Like storage devices 120 and 220, storage device 320 is a recording medium readable by control device 310. A control program 322 is stored in storage device 320. Control program 322 includes both an operation system program that allows control device 310 to control each unit of tag generation unit 300, and an application program for generating tags.

[0024] Like the control devices 110 and 210, the control device 310 executes application programs stored in the storage device 320 to construct a conversion unit 312, a clustering unit 314, and a second large-scale language model 316. The control device 310 also instructs data transfer, execution of various processes, etc. The conversion unit 312 converts character strings of free-form text formatted by the formatting unit 200 into vectors. The clustering unit 314 clusters the converted vectors into a specified number of clusters. The second large-scale language model 316 summarizes each cluster clustered by the clustering unit 314 and generates them as tags.

[0025] The communication device 350 is hardware for communicating with the input receiving unit 100 , the shaping unit 200 , or the allocation unit 400 via the network 50 .

[0026] 5 is a block diagram showing the configuration of the allocation unit 400. The allocation unit 400 includes a control unit 410, a storage unit 420, and a communication unit 430.

[0027] Like storage devices 120, 220, and 320, storage device 420 is a recording medium readable by control device 410. Storage device 420 stores control program 422. Control program 422 includes both an operation system program that allows control device 410 to control each component of assignment unit 400, and an application program that assigns tags to text.

[0028] Similar to the control devices 110, 210, and 310, the control device 410 executes an application program stored in the storage device 420 to construct a third large-scale language model 416. The third large-scale language model 416 assigns tags generated by the tag generation unit 300 to free-form text input accepted by the input acceptance unit 100. The control device 410 also issues instructions for data transfer, execution of various processes, etc.

[0029] The communication device 350 is hardware for communicating with the input receiving unit 100 , the shaping unit 200 , or the allocation unit 400 via the network 50 .

[0030] In this embodiment, the reason why free-form text is sequentially processed through shaping, vector conversion, clustering, tag generation, and tag assignment is as follows: Generally, large-scale language models do not perform well in clustering objects. However, they are more accurate than clustering in generating tags as summaries for clusters that have already been clustered accurately using another method, and in classifying the generated tags.

[0031] On the other hand, as described above, free-form text contains redundant content, uses templates, and has a variety of writing styles, so it is difficult to directly cluster free-form text, or the accuracy is low. Therefore, in this embodiment, first, the free-form text accepted by the input accepting unit 100 is first shaped by the first large-scale language model 216 and converted into a unified description in natural language.

[0032] Second, in this embodiment, in order to cluster the formatted text with high accuracy, the text is converted into vectors and the converted vectors are clustered. Third, in this embodiment, a summary of each cluster is generated as a tag using the second large-scale language model 316, and the generated tag is assigned to the original free-form text (before formatting) or the formatted text using the third large-scale language model 416.

[0033] 6 is a diagram showing the operation sequence of the system 1, particularly the data flow. First, an operator requesting analysis operates the terminal device 60 to transmit multiple documents to be analyzed to the input receiving unit 100 (step Sc1). That is, the multiple documents to be analyzed are transmitted to the input receiving unit 100 via the network 50. The input receiving unit 100 receives input of the multiple documents transmitted from the terminal device 60 (step Sc2). As described above, the documents are free-form text.

[0034] The input receiving unit 100 transmits the received free-form text to the shaping unit 200 and the allocation unit 400 (step Sc3), and instructs the shaping unit 200 to shape the received free-form text. In the shaping unit 200 that has received the free-form text, the first large-scale language model 216 shapes the free-form text in accordance with the instructions from the input receiving unit 100 (step Sc4). An example of shaping the free-form text will be described later.

[0035] The text shaped by the shaping unit 200 is transmitted to the tag generation unit 300 (step Sc5a). Specifically, when the text is shaped by the shaping unit 200, the control unit 210 notifies the input reception unit 100 that the text shaping is complete. In response to this notification, the control unit 110 of the input reception unit 100 instructs the shaping unit 200 to transmit the shaped text to the tag generation unit 300. In accordance with this instruction, the shaped text is transmitted to the tag generation unit 300.

[0036] In the tag generation unit 300, the conversion unit 312 converts the character strings of the formatted text into vectors, for example, using a trained model (step Sc6). Note that converting a character string into a vector means converting the semantic features of the character string into an array of numerical values. Converting a character string into a vector not only makes it possible to quantitatively analyze the similarity and relevance between character strings, but also makes it possible to know the distribution of meanings in the character strings. Note that an example of the distribution of vectors converted from character strings will be described later.

[0037] The converted vectors are transferred to the clustering unit 314. The clustering unit 314 clusters the converted vectors into a specified number of clusters using, for example, the k-means method (step Sc7). Next, the second large-scale language model 316 generates a summary of each cluster as a tag (step Sc8).

[0038] The generated tag is transmitted to the terminal device 60 (step Sc9). Specifically, when the tag is generated in the tag generation unit 300, the control device 310 notifies the input reception unit 100 that the tag generation is complete. In response to this notification, the control device 110 of the input reception unit 100 instructs the tag generation unit 300 to transmit the generated tag to the terminal device 60. In accordance with this instruction, the generated tag is transmitted to the terminal device 60 operated by the person requesting the analysis (step Sc9).

[0039] The assigning unit 400 is responsible for assigning tags to free-form text, but tags generated by the tag generating unit 300 (second large-scale language model 316) may have unnatural semantics or expressions. Furthermore, similar tags may be generated depending on the situation. In such cases, it may be better to align similar tags. Therefore, the tags generated by the tag generating unit 300 are first sent to the terminal device 60, where they are corrected or confirmed by the operator who requested the analysis (step Sc10). The tags corrected or confirmed by the operator are then sent to the assigning unit 400 (step Sc11). The tags sent to the assigning unit 400 become candidates for assignment to the free-form text.

[0040] Since there are multiple free-text entries, multiple tags are also generated. However, even if the free-text entries are different, the tag generation unit 300 may generate the same tag if the meanings of the character strings in the texts are similar. Furthermore, due to revisions by the person requesting the analysis, tags based on two or more different free-text entries may end up being the same. For this reason, the number of free-text entries received by the input receiving unit 100 does not necessarily match the number of tags selected as candidates for assignment by the assignment unit 400.

[0041] In the assignment unit 400, the third large-scale language model 412 assigns tags from among the tags that have been selected as candidates for assignment that are close in meaning to the multiple free-form texts sent from the input receiving unit 100 in step Sc3 (step Sc12).

[0042] As described above, the text to which tags are assigned by the assignment unit 400 may be that which has been shaped by the shaping unit 200. Therefore, tags may be assigned to the shaped text transmitted from the shaping unit 200 in step Sc5b indicated by the dashed line. Whether to use the unshaped or shaped text as the text to be assigned tags is selected, for example, by an instruction from the person requesting the analysis to the terminal device 60. This selection may be made collectively for all texts to be analyzed, or may be an instruction to select individual texts.

[0043] The output result of the allocation unit 400, i.e., the free-form text with tags assigned, is sent to the terminal device 60 (step Sc13). This allows the person requesting the analysis to obtain the analysis result with tags assigned to the free-form text. The output result of the allocation unit 400 may also be sent to a separate server for aggregation.

[0044] 7 is a diagram showing an example of the process from accepting input of free-form text to assigning tags. 1. to 3. in (A) of the figure are examples of free-form text accepted by the input accepting unit 100, and in this case, requests for telecommunications services and the like are described. 2. and 3. are examples in which a template was used. 1. to 3. in (B) are examples in which 1. to 3. in (A) have been shaped in order by the shaping unit 200 (first large-scale language model 216), taking into account the expression using the template.

[0045] 1. and 2. in (C) are examples of tags generated as a summary by the tag generation unit 300 based on 1. to 3. in (B). Here, the tag shown as 1. in (C) is shared for text formatted as in 1. and 2. in (B). Note that the summary generated for text formatted as in 1. and 2 in (B) is "content related to poor signal reception," but in order to standardize the tags, for example, this is an example where the summary is modified to "content related to ease of signal reception," as shown in 1. in (C).

[0046] (D) is a diagram showing an example in which tags have been assigned to free-text text by the assignment unit 400 (third large-scale language model 416). In this example, the tag 1 in (C) has been assigned to the free-text text 1. and the free-text text 2. in (A), and the tag 2 in (C) has been assigned to the free-text text 3. in (A).

[0047] FIG. 8 is a diagram showing an example distribution of vectors converted by the conversion unit 312 in the tag generation unit 300. As described above, the converted vectors represent the semantic features of a character string as a numerical array. FIG. 9 is a visualization of this numerical array. In the diagram, coordinates indicating the numerical array are shown as black dots. Therefore, the converted vectors are represented as vectors whose start point is the origin and whose end point is the black dot. An example of clustering by the clustering unit 314 is shown by the dashed lines in the diagram. The example in the diagram is an example of vectors clustered into two clusters A and B as "groups."

[0048] Note that some numeric sequences are not clustered because they are semantically different from other strings. In other words, when vector conversion and clustering are performed on the strings of formatted text, sequences with different meanings are separated from other numeric sequences and excluded from the class, so they do not affect the generation of summaries.

[0049] 9 is a diagram showing an example in which tags are generated individually for two clusters A and B by the second large-scale language model 316. Note that the diagram shows an example in which the tags generated for cluster A by the second large-scale language model 316 have been corrected.

[0050] In this embodiment, a tag that is closest in meaning to the free-written text is assigned to the free-written text from among multiple tags that are candidates for assignment. Therefore, the tag assigned to the free-written text may be a tag generated (and modified) based on the text, or may be a tag generated (and modified) based on text other than the text. In other words, if a tag (conveniently referred to as N) that is closer in meaning to the text than a tag generated based on the free-written text (conveniently referred to as M) is selected as a candidate for assignment, the free-written text is assigned the tag N, which is closer in meaning to the text than the tag M. Therefore, according to this embodiment, a tag that accurately indicates the meaning of the text is assigned to the free-written text and output.

[0051] In the embodiment, the flow of free-form text accepted by the input accepting unit 100 and data based on that text is as shown in FIG. 6 , but is not limited to this. For example, if the shaping unit 200 is not subject to control or management by the input accepting unit 100, the input accepting unit 100 will be unable to cause the shaping unit 200 to directly output the shaped text to the tag generating unit 300. A similar issue arises when the tag generating unit 300 and the allocation unit 400 are not subject to control or management by the input accepting unit 100. Therefore, next, an application example that improves this issue will be described. In this application example, the accepted free-form text and data based on that text are transmitted and received between the shaping unit 200, the tag generating unit 300, and the allocation unit 400, with the input accepting unit 100 as the core.

[0052] 10 is a block diagram showing the configuration of the input reception unit 100 in the application example. In the input reception unit 100 in the application example, the control device 110 executes an application program stored in the storage device 120 to construct a first acquisition unit 112, a second acquisition unit 114, and a third acquisition unit 116. Note that the display device 130 and the operation device 140 in the input reception unit 100 are omitted.

[0053] The first acquisition unit 112 accepts free-form text sent from the terminal device 60, transmits it to the shaping unit 200, and acquires the text shaped into a specified format from the shaping unit 200. The second acquisition unit 114 transmits the text acquired by the first acquisition unit 112, i.e., the text shaped by the shaping unit 200, to the tag generation unit 300 and acquires tags generated based on the shaped text from the tag generation unit 300. The third acquisition unit 116 transmits the tags acquired by the second acquisition unit 114 to the terminal device 60 and transmits the tags (tags based on the second acquisition unit 114) that have been corrected or confirmed by the person requesting the analysis in the terminal device 60 to the assignment unit 400. Furthermore, the third acquisition unit 116 acquires the text to which the tags have been assigned (the text whose input has been accepted or the shaped text) from the assignment unit 400 and transmits it to the terminal device 60.

[0054] In such an application example, as shown in FIG. 10, even when the formatting unit 200, tag generation unit 300, and assignment unit 400 are not under the control or management of the input receiving unit 100, analysis results of the freely written text that has been received as input can be obtained, just as in the embodiment.

[0055] In the above-described embodiment and application example (hereinafter referred to as "embodiment, etc."), the system 1 is configured to request analysis of multiple documents at once. However, after the analysis request, there may be cases where an additional document is requested to be analyzed. In this case, since a certain number of tags already exist as assignment candidates, the processing of the tag generation unit 300, specifically, the processing of steps Sc5a to Sc9 in FIG. 6, may be skipped. Furthermore, when an additional document is requested to be analyzed, if tags that exist as assignment candidates are to be assigned to the text formatted by the formatting unit 200, the processing of steps Sc5a and Sc6 to Sc9 may be skipped.

[0056] <Others> (1) In the embodiments, storage devices 120, 220, 320, and 420 are exemplified by ROM and RAM, but may also be flexible disks, magneto-optical disks (e.g., compact disks, digital versatile disks, Blu-ray (registered trademark) disks), smart cards, flash memory devices (e.g., cards, sticks, key drives), CD-ROMs (Compact Disc-ROMs), registers, removable disks, hard disks, floppy (registered trademark) disks, magnetic strips, databases, servers, or other suitable storage media.

[0057] (2) In the above-described embodiments, the described information, signals, etc. may be represented using any of a variety of different technologies. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0058] (3) In the above-described embodiments, input and output information may be stored in a specific location (for example, a memory) or may be managed using a management table. Input and output information may be overwritten, updated, or added to. Output information may be deleted. Input information may be transmitted to another device.

[0059] (4) In the above-described embodiments, the determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a comparison of numerical values ​​(e.g., a comparison with a predetermined value).

[0060] (5) The order of the exemplary process procedures, sequences, flowcharts, etc. illustrated in the above-described embodiments may be rearranged unless it is consistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0061] (6) Each function illustrated in Figures 2 to 5 is realized by any combination of hardware and / or software. Furthermore, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using a single device that is physically or logically coupled, or may be realized using two or more physically or logically separated devices that are directly or indirectly connected (e.g., wired, wireless, etc.) and these multiple devices. A functional block may be realized by combining software with the single device or the multiple devices.

[0062] (7) The programs exemplified in the above-described embodiments, etc., whether called software, firmware, middleware, microcode, hardware description language, or by other names, should be broadly construed to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc.

[0063] Software, instructions, information, etc. may also be transmitted or received over a transmission medium. For example, if software is transmitted from a website, server, or other remote source using wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave), then these wired and / or wireless technologies are included within the definition of transmission media.

[0064] (8) In the above-described embodiments, the terms "system" and "network" are used interchangeably.

[0065] (9) The information, parameters, etc. described in this disclosure may be expressed using absolute values, relative values ​​from a predetermined value, or corresponding other information.

[0066] (10) In the above-described embodiments and the like, the portable device includes a mobile station (MS). A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other appropriate term. In this disclosure, terms such as "mobile station," "user terminal," "user equipment (UE)," and "terminal" may be used interchangeably.

[0067] (11) In the above-described embodiments, the terms "connected" and "coupled," or any variations thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using at least one of one or more electric wires, cables, and printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0068] (12) In the above-described embodiments, the phrase "based on" does not mean "based only on," unless otherwise specified. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0069] (13) As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, inquiring (e.g., searching in a table, database, or other data structure), ascertaining something as "determining" or "determining," and the like. Furthermore, "judgment" and "decision" may include regarding receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, and accessing (e.g., accessing data in memory) as having been "judgment" or "decision." Furthermore, "judgment" and "decision" may include regarding resolving, selecting, choosing, establishing, comparing, etc. as having been "judgment" or "decision." In other words, "judgment" and "decision" may include regarding some action as having been "judgment" or "decision." Furthermore, "judgment" may be interpreted as "assuming," "expecting," "considering," etc.

[0070] (14) In the above-described embodiments, when the terms "include," "including," and variations thereof are used, these terms are intended to be inclusive, similar to the term "comprising." Furthermore, the term "or," as used in this disclosure, is not intended to mean an exclusive or.

[0071] (15) In this disclosure, where articles are added by translation, such as a, an, and the in English, this disclosure may include that the nouns following these articles are plural.

[0072] (16) In this disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "combined" may also be interpreted in the same way as "different."

[0073] (17) Each aspect / embodiment described in this disclosure may be used alone, in combination, or switched depending on the implementation. Furthermore, notification of predetermined information (e.g., notification that "X is true") is not limited to explicit notification, but may be implicit (e.g., not notifying the predetermined information). Multiple aspects selected from the above may be combined.

[0074] 1...text analysis device, 100...input receiving unit, 200...shaping unit, 216...first large-scale language model, 300...tag generation unit, 312...conversion unit, 314...clustering unit, 316...second large-scale language model, 400...assignment unit, 416...third large-scale language model.

Claims

1. A text analysis device comprising: an input accepting unit that accepts input of free-form text; a formatting unit that formats the text accepted by the input accepting unit into a specified format; a tag generating unit that generates tags based on the text formatted by the formatting unit; and an assignment unit that assigns the tags generated by the tag generating unit to the text accepted by the input accepting unit or to the text formatted by the formatting unit.

2. The text analysis device according to claim 1, wherein the formatting unit is a first large-scale language model and outputs the text received by the input receiving unit in a natural language.

3. A text analysis device as described in claim 2, including: a conversion unit that converts character strings of text formatted by the formatting unit into vectors; and a clustering unit that clusters the vectors, wherein the tag generation unit generates the tags by inputting the clustered vectors into a second large-scale language model.

4. The text analysis device according to claim 3, wherein the tags generated by the tag generation unit are modifiable.

5. The text analysis device according to claim 4, wherein the assignment unit assigns tags generated or modified by the tag generation unit to the text accepted by the input acceptance unit using a third large-scale language model.

6. A text analysis device comprising: a first acquisition unit that sends input-accepted free-form text to a shaping unit and acquires text shaped into a specified format from the shaping unit; a second acquisition unit that sends the text acquired by the first acquisition unit to a tag generation unit and acquires a tag generated based on the shaped text from the tag generation unit; and a third acquisition unit that sends the tag based on the second acquisition unit to an assignment unit and acquires from the shaping unit the text acquired by the input acceptance unit to which the tag has been assigned or the text shaped by the shaping unit.

Images