Information processing device, information processing method, and program

Abstract

Improve accuracy when using generative models to organize data. [Solution] The information processing device 10 includes an extraction unit 102 that extracts multiple elements from a structured document based on multiple query expressions, and an organizing unit 103 that provides internal text related to each of the extracted elements as input data to the generation AI server 20 model and outputs output data in response to a prompt that instructs the model to organize the multiple item texts by associating them with multiple item names.

Description

[Technical Field]

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

[0002] A technique has been proposed for collecting data from web pages and extracting predetermined information from the collected data (for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2008-226235 Summary of the Invention [Problem to be solved by the invention]

[0004] Recently, generative AI (Artificial Intelligence) has become popular, and it is possible to use generative AI to organize predetermined data. However, when predetermined data is input directly into generative AI, the desired output may not be obtained, and there are issues with the accuracy of the output results.

[0005] The present invention has been made in view of the above-mentioned circumstances, and aims to provide an information processing device and the like that improves the accuracy when organizing data using a generative model. [Means for solving the problem]

[0006] In order to achieve the above object, the information processing device of the present invention comprises: an extracting unit that extracts a plurality of elements from the structured document based on a plurality of query expressions, respectively; a reorganization unit that provides the internal texts relating to each of the extracted elements as input data to the generative model, and outputs output data in response to a prompt that instructs that the item texts should be organized by associating them with the item names, respectively; Equipped with. [Effects of the Invention]

[0007] According to the present invention, it is possible to improve the accuracy when organizing data using a generative model. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a functional block diagram of an information processing device according to an embodiment of the present invention; [Figure 2] FIG. 1 is a diagram illustrating an example of a hardware configuration of an information processing device according to an embodiment of the present invention. [Figure 3] 10 is a flowchart illustrating an example of a metadata acquisition process. [Figure 4] FIG. 10 is a diagram showing an example of XPath information for each website. [Figure 5] FIG. 10 is a diagram showing an example of the locations of items extracted based on XPath in a web page. [Figure 6] FIG. 1 illustrates an example of inner text extracted from a web page. [Figure 7] FIG. 10 is a diagram showing an example of a prompt for a generation AI server. [Figure 8] 10A and 10B are diagrams showing examples of display on a web browser. [Figure 9] 10A and 10B are diagrams showing examples of display on a web browser. [Figure 10] FIG. 10 is a diagram illustrating an example of a display on a web browser. DETAILED DESCRIPTION OF THE INVENTION

[0009] An information processing device 10 according to an embodiment of the present invention will be described below with reference to the drawings, in which the same or corresponding parts are designated by the same reference numerals.

[0010] When extracting multiple elements related to a specific item from a structured document related to the specific item (for example, a web page or document related to the specific item), a person can manually extract the elements. However, considering the work time and accuracy of the extraction, it is desirable to automate the extraction. Furthermore, when the extracted elements are to be reused elsewhere, it is necessary to organize the extracted elements by listing them and assigning item names to each element. The information processing device 10 of this embodiment automatically extracts elements from such a structured document and organizes the extracted elements. Furthermore, a generation AI is used to organize the extracted elements.

[0011] FIG. 1 is a functional block diagram showing the functional configuration of an information processing device 10 according to this embodiment. The information processing device 10 according to this embodiment, together with a generation AI server 20, constitutes an information processing system 100. The information processing device 10 is connected to a communication network such as the Internet or a mobile communication network, and transmits and receives various information via the communication network. The information processing device 10 and the generation AI server 20 are also connected via the communication network, and transmit and receive various information to and from each other via the communication network.

[0012] The information processing device 10 processes a structured document. It acquires a structured document and extracts multiple elements from the structured document based on a predetermined query expression. The information processing device 10 then provides internal text related to the extracted multiple elements as input data to the generation AI server 20, and outputs output data consisting of multiple item texts in response to a prompt that instructs the generation AI server 20 to organize the multiple item texts by associating multiple item names with each item text.

[0013] The generation AI server 20 is a server equipped with a generation AI. The generation AI server 20 is a computer equipped with, for example, a large language model (LLM) such as ChatGPT (Generative Pre-trained Transformer) as a generation model, and executes processing of character strings composed of input natural language. The information processing device 10 accesses the generation AI server 20, for example, by using an API (Application Programming Interface) provided by the generation AI server 20.

[0014] In this embodiment, the generation AI server 20 receives text data of internal text and a prompt that instructs processing of the text data from the information processing device 10. Then, the generation AI server 20 collects and organizes item text from the text data based on the prompt, and returns the organized data to the information processing device 10 as output data.

[0015] Elements extracted from a structured document are data corresponding to items included in the structured document, for example, text data of words. The text data extracted from this structured document becomes input data to the generation AI server 20. In this embodiment, the text data that becomes input data to the generation AI server 20 is called internal text. Furthermore, the output data output from the generation AI server 20 using this internal text as input data is an organized version of the internal text, and is composed of item names and text data corresponding to those item names. In this embodiment, text data corresponding to such item names is called item text. Note that elements and items included in a structured document are roughly synonymous, but the term "element" may refer to what is before being organized by the generation AI server 20, and the term "item" may refer to what is organized after being organized by the generation AI server 20.

[0016] Data relating to multiple elements and items contained in a structured document, or text data (item text) organized and output by the generation AI server 20, are also referred to as metadata. The information processing device 10 of this embodiment aims to ultimately obtain metadata from a structured document. Furthermore, it aims to collect metadata from multiple structured documents and make it available as organized information in other systems, etc.

[0017] The information processing device 10 is functionally composed of a document acquisition unit 101, an extraction unit 102, a sorting unit 103, an inspection unit 104, a warning unit 105, and an output unit 106, as shown in FIG.

[0018] The document acquisition unit 101 acquires a structured document from outside the information processing device 10 via a communication network, a recording medium, or the like. The structured document is, for example, an SGML (Standard Generalized Markup Language) document, an XML (Extensible Markup Language) document, or an HTML (HyperText Markup Language) document. Note that the structured document may be another structured document composed of a document and tags that indicate the structure of the document, or may be a markup document.

[0019] For example, the document acquisition unit 101 receives input from a user of a URL (Uniform Resource Locator) of a web page that is the source of a structured document, and acquires the structured document of the web page at the input URL. The document acquisition unit 101 may acquire a structured document that is directly input to the information processing device 10 via a communication network or a recording medium.

[0020] Extraction unit 102 extracts multiple elements from the structured document acquired by document acquisition unit 101 based on multiple query expressions that instruct the extraction of multiple elements. In this embodiment, each of the multiple query expressions is expressed by XPath (XML Path Language), which indicates the locations of the multiple elements. Extraction unit 102 then extracts multiple elements located at the positions indicated by the XPath from the structured document.

[0021] XPath is a language for specifying specific parts of a document, such as elements and attribute values ​​in an XML document, and is expressed using a path notation similar to a URL. XPath can also be applied to other structured documents, such as HTML documents.

[0022] In this embodiment, XPath information indicating the portion (path) where multiple elements are written is prepared in advance in a website that is the source of a structured document. Specifically, XPath indicating the location of each element for each website that is the source of a structured document is stored in advance in information processing device 10. Note that in this embodiment, a website refers to a collection of web pages that provide predetermined service information and the like as content under a specific domain on the Internet. An individual page on a website where individual content and the like is written is called a webpage.

[0023] In this embodiment, the website from which the structured document is obtained is assumed to be a website that provides information on a wide variety of content, such as various database sites and catalog sites. When the same website provides information on multiple contents, individual pages tend to have the same format from the standpoint of ease of viewing and page creation. Therefore, the location of each element on each individual page is easily determined depending on the website. Therefore, for such websites, it is possible to predefine XPaths that indicate the location of each element for each website from which the structured document is obtained.

[0024] Based on such XPath, the extraction unit 102 extracts inner text related to multiple elements from the structured document acquired by the document acquisition unit 101. For example, the extraction unit 102 extracts the textContent attribute (simple text) or innerText attribute (text reflecting escape characters) related to each of the multiple elements as the inner text related to each of the elements.

[0025] The internal text extracted by the extraction unit 102 is a plurality of elements contained in a structured document, and can therefore be considered metadata contained in the structured document. However, in a structured document such as a web page, a plurality of related elements are displayed in a group to some extent. The extraction unit 102 then extracts the grouped elements in a certain number of chunks based on XPath. In order to further organize the internal text, which is metadata extracted as a group by the extraction unit 102, the extraction unit 102 passes the internal text extracted from the structured document to the organization unit 103.

[0026] The organizing unit 103 provides the generation AI server 20 with the internal text for each of the multiple elements extracted by the extraction unit 102 as input data. The organizing unit 103 then provides the internal text and a prompt instructing the generation AI server 20 to organize and output the internal text by associating multiple item texts with multiple item names, respectively. The organizing unit 103 then obtains the item text organized based on the prompt from the generation AI server 20 as output data. Furthermore, upon receiving the output data, the organizing unit 103 formats the output data (item text) into an output format for the output unit 106.

[0027] The inspection unit 104 checks whether there are any errors in the output data by comparing the input data to the generation AI server 20 with the output data from the generation AI server 20. Specifically, the inspection unit 104 checks whether the item text associated with multiple item names in the output data from the generation AI server 20 is included in the internal text of the structured document, i.e., the input data to the generation AI server 20. If the generation AI server 20 outputs item text that is not included in the input data, hallucination by the generation AI server 20 is considered. As a result, the inspection unit 104 inspects and determines whether any malfunctions or hallucination have occurred in the generation AI server 20. In this way, the inspection unit 104 inspects the consistency of the input and output data. Note that the inspection unit 104 may inspect the output data using other inspection methods, and there may be inspection items other than the inspection of the consistency of the input and output data.

[0028] When the inspection unit 104 determines that there is item text that is not included in the input data, the warning unit 105 generates and outputs warning information that warns of the item text that is determined not to be included.

[0029] The output unit 106 receives the formatted output data from the organizing unit 103 and outputs it to the outside as metadata. The output unit 106 also receives warning information generated by the warning unit 105 and outputs the warning information to the outside. For example, the output unit 106 outputs the metadata organized by the organizing unit 103 and the warning information generated by the warning unit 105 to a display device included in the information processing device 10 or a display device connected to the information processing device 10. For example, the output unit 106 also outputs data (files) corresponding to the metadata organized by the organizing unit 103 and the warning information generated by the warning unit 105 to a terminal device or the like designated by the user so that the data (files) can be downloaded.

[0030] The information processing device 10 is physically configured, for example, as a personal computer (PC), a workstation, or the like. The information processing device 10 may also be a dedicated device. FIG. 2 is a diagram showing an example of the hardware configuration of the information processing device 10. The information processing device 10 has a processor 11 such as a CPU (Central Processing Unit) or a GPU (Graphical Processing Unit), a storage device 12 such as a memory, an HDD (Hard Disk Drive) and / or an SSD (Solid State Drive), a communication interface 13 for wired or wireless communication, an input device 14 for accepting input operations, and an output device 15 for outputting information. The input device 14 is, for example, a keyboard, a touch panel, a camera, a mouse, and / or a microphone, etc. The output device 15 is, for example, a display and / or a speaker, etc.

[0031] The processor 11 of the information processing device 10 reads a program stored in the storage device 12, and operates as the various functional units shown in FIG.

[0032] In this embodiment, it is assumed that a user uses a terminal device such as a smartphone, tablet, or PC to connect to the information processing device 10 via a communication network and use the functions of the information processing device 10. The functions of the information processing device 10 can be used from the terminal device using a dedicated application or a general-purpose web browser. However, this is not limiting, and the user may directly operate the information processing device 10 to use the functions of the information processing device 10.

[0033] Next, the operation of the information processing device 10 will be described. Fig. 3 is a flowchart showing an example of metadata acquisition processing executed by the information processing device 10. The metadata acquisition processing is executed by the processor 11 of the information processing device 10 reading a program stored in the storage device 12. The metadata acquisition processing is started, for example, based on a metadata acquisition operation by a user.

[0034] In the metadata acquisition process, first, the document acquisition unit 101 of the information processing device 10 acquires the URL of the structured document (step S101). The URL may be input to the information processing device 10 individually, or a list of multiple URLs may be input.

[0035] Next, the document acquisition unit 101 accesses the web page of the acquired URL via the Internet (step S102).

[0036] Then, extraction unit 102 acquires XPath information corresponding to the website name (e.g., domain) of the accessed webpage (step S103). FIG. 4 shows an example of XPath information for each website. As shown in FIG. 4, the XPath for each website is information associating the website name, element names (frame names) corresponding to the locations of multiple elements in the structured document of the webpage on the website, and XPaths indicating the locations of the elements indicated by the element names on the website. XPath information for each website that provides a webpage from which the structured document is obtained is pre-stored in storage device 12 of information processing device 10. Extraction unit 102 then identifies the website from which the structured document is obtained based on the acquired URL, and acquires XPath information corresponding to the website from which the structured document is obtained. In other words, extraction unit 102 switches the query expression for extracting internal text depending on the website from which the structured document is obtained. This makes it possible to extract internal text appropriately depending on the source of the structured document.

[0037] FIG. 5 is a diagram showing an example of the locations of items extracted based on XPath in a web page. FIG. 5 shows an example of a web page provided by the website named "AAA" in FIG. 4. The website named "AAA" is a website that provides movie information, and as shown in FIG. 4, the element names "movie info," "staff info," "cast info," and "movie spec" are defined, and their XPaths are also defined. In the web page shown in FIG. 5, multiple related elements are grouped together to a certain extent, and the grouped parts are treated as frames, and element names are defined, and their positions are defined by XPath. The XPath of "AAA" shown in FIG. 4 indicates the locations (frames) f1 to f4 of the elements "movie info," "staff info," "cast info," and "movie info" in the web page of "AAA" as shown in FIG. 5.

[0038] After the process of step S103, the extraction unit 102 extracts inner text corresponding to each element name from the structured document of the web page based on such XPath information (step S104).

[0039] Steps S101 to S104 correspond to the part where document acquisition unit 101 of information processing device 10 acquires a structured document and extraction unit 102 extracts internal text, and show an example where the structured document is a web page. In other cases, for example, when a structured document is input directly, the processes of steps S101 and S102 can be omitted. Also, when the structured document is not a web page, multiple query expressions corresponding to the structured document can be acquired in step S103, and internal text can be extracted from the structured document based on the multiple query expressions in step S104.

[0040] Then, the extraction unit 102 cleanses the extracted internal text (step S105). Specifically, the extraction unit 102 adds element names (frame names) to the extracted internal text to organize it and prepare input data for the generation AI server 20. Note that the internal text may be cleansed by detecting, deleting, or correcting erroneous data, duplicate data, etc.

[0041] Figure 6 shows an example of internal text extracted from a web page on the "AAA" website, which provides movie information, and organized by adding element names (frame names). As shown in Figure 6, adding item names and labels associated with the XPath used for extraction to the list of extracted internal text increases the identifiability of the internal text, thereby improving the accuracy of information understanding by the generation AI server 20.

[0042] Next, the organizing unit 103 transmits the internal text cleansed in step S105 to the generation AI server 20 as input data together with a prompt indicating instructions to the generation AI server 20 (step S106). Note that only the internal text may be referred to as input data.

[0043] Figure 7 shows an example of a prompt 200 for the generation AI server 20. The prompt 200 is a command or instruction to the generation AI server 20. Specifically, the prompt 200 shown in Figure 7 instructs the extraction of metadata (item text) for the item name specified in the #metadata template 201 from the internal text (**text** in Figure 7) that is the input data, as described in the #task. In addition, specific processing procedures are defined in each step of the #instructions. The description of the #instructions may be such that the content of the #task can be executed appropriately.

[0044] In Figure 7, the steps of the #instructions include an instruction to use a **metadata template** (step 1), an instruction to extract items described in the **metadata template** from **text** (step 2), an instruction to understand the rules for extraction (step 3), an instruction to extract related information (step 4), an instruction to review the extraction results (step 5), and an instruction to output the extraction results (step 6). Other steps may also be included.

[0045] Symbols such as # indicate headings and are used to make the prompt easier for the generation AI server 20 to understand. In the explanation, these symbols will be used as is. Note that the method of writing the prompt 200 and the method of using the symbols are not limited to this, and any description suitable for the large-scale language model installed in the generation AI server 20 will suffice.

[0046] The item names specified in the #metadata template 201 are item names corresponding to the input data, and the item names of the metadata to be collected from the structured document (web page) that is the source of the input data are described. Therefore, the #metadata template 201 only needs to describe item names corresponding to the source of the input data. The #metadata template 201 in FIG. 7 shows an example in which the internal text of the movie information shown in FIG. 6 is the input data, and describes item names corresponding to the items included in the movie information, i.e., the metadata. In the information processing device 10, by rewriting the #metadata template 201, it is possible to acquire metadata corresponding to the information source (the source of the input data) without rewriting the entire prompt.

[0047] Step 3 of the # instruction in prompt 200 defines rules that the generation AI server 20 must understand when extracting metadata. In this embodiment, these rules include a nuance list 202 that defines synonymous and non-synonymous relationships between terms that may be included in the input data. Synonymous relationships in the nuance list 202 include alternative expressions for terms (item names), such as "original author" and "creator," "director," "chief director," and "series director." Non-synonymous relationships in the nuance list 202 include a rule such as "the year of production should not be extracted from the release date," since "year of production" is not necessarily the same as "release date." The nuance list 202 may contain at least one of synonymous and non-synonymous relationships between terms, depending on the source of the input data. This nuance list 202 defines synonymous and non-synonymous relationships between terms that may be included in the input data, which is expected to improve the accuracy of metadata extraction by the generation AI server 20. Furthermore, in the information processing device 10, by rewriting the nuance list 202, it is possible to acquire metadata according to the information source without rewriting the entire prompt.

[0048] In addition, the rules for extraction that the generation AI server 20 must understand include character conversion, date conversion, adding or deleting "·" or spaces between first and last names, and standardizing notation.

[0049] Furthermore, the related information in the instruction in step 4 of prompt 200 includes, for example, error information during metadata extraction. For example, the instruction in step 4 may include instructions to extract as error information, such as the fact that metadata (item text) corresponding to an item name could not be extracted, or that the instruction in prompt 200 could not be understood and executed.

[0050] The instruction to output the extraction results of step 6 of prompt 200 includes, for example, an instruction to output the item names included in the #metadata template 201 in association with the item text extracted from the internal text. The instruction to output the extraction results of step 6 may also include a specification of the metadata output format. Examples of metadata output formats include plain text, CSV (Comma-Separated Values), and JSON (JavaScript Object Notation).

[0051] When the generation AI server 20 receives the internal text (e.g., Figure 6) and the prompt (e.g., Figure 7) sent from the information processing device 10 in step S106, it processes the internal text in accordance with the prompt to extract metadata (item text corresponding to the item name) and returns the item text to the information processing device 10 as output data in the specified output format.

[0052] Thereafter, the organizer 103 waits to receive the internal text as input data and output data corresponding to the prompt from the generation AI server 20 (step S107).

[0053] When output data is received from the generation AI server 20 (step S107; Yes), the inspection unit 104 inspects whether the output data is included in the input data (step S108). The inspection unit 104 inspects whether the item text (metadata) of each item name that will be output data from the generation AI server 20 is included in the internal text that will be input data that the organization unit 103 sent to the generation AI server 20 in step S106.

[0054] As another check, the check unit 104 may check whether the item name has been misidentified by, for example, comparing the item name with the item text. For example, if the item names include "staff" and "cast," it may be checked whether "cast" has been mistaken for "staff" or vice versa.

[0055] In addition, if the inspection unit 104 determines that the output data contains item text that is not included in the input data, it may specify the relevant part of the item text that is the output data and instruct the generation AI server 20 to correct it.

[0056] After step S108, the organizing unit 103 formats the item text (metadata), which is output data from the generation AI server 20, into a predetermined output format (step S109). The output format may be, for example, TSV (Tab-Separated Values), CSV, etc. The user may be able to specify the output format. The generation AI server 20 may also be allowed to perform the formatting into the output format, or if the predetermined output format is a format that can be output from the generation AI server 20, the output data may be output from the generation AI server 20 in that predetermined output format.

[0057] The output unit 106 then outputs the metadata (e.g., a list of item texts) that will become the formatted output data (step S110). Furthermore, if the inspection in step S108 determines that the output data contains item texts that are not included in the input data, the warning unit 105 generates warning information, such as a warning about the item texts that are not included in the input data. Then, in step S110, the output unit 106 outputs the warning information along with the output data. After the processing in step S110, the metadata acquisition process ends. Outputting such warning information can warn about erroneously generated metadata and prevent the erroneously generated metadata from being used elsewhere. This also leads to improvements in Xpath information and prompts in the information processing device 10, which is expected to reduce errors during metadata acquisition and reduce the occurrence of hallucinations in the generation AI server 20.

[0058] In step S110, the output unit 106 displays and outputs the metadata on, for example, a display device directly connected to the information processing device 10 or a terminal device connected via a communication network. The output unit 106 also outputs the metadata as a file that can be downloaded to, for example, a terminal device connected via a communication network. The output method and output destination of the metadata that serves as output data may be any of those specified by a user or operator of the information processing device 10. The metadata that serves as output data may be any of those that can be used by the user or operator. This makes it possible to extract and output metadata that serves as output data in a way that is suitable for the user or operator.

[0059] Furthermore, if a list of multiple URLs is entered in step S101, steps S102 to S110 are executed for the websites corresponding to each URL. This allows metadata to be automatically collected from multiple structured documents, and the collected metadata can be utilized. This also significantly reduces the human cost and time required to collect metadata from multiple structured documents.

[0060] Next, a display example will be described when a user acquires metadata using the information processing device 10 with a terminal device. When a user connects to the information processing device 10 via a web browser 110 using a communication-capable terminal device such as a personal computer, smartphone, or tablet, a top page of a website for acquiring metadata, such as that shown in Fig. 8(A), is displayed. A sidebar 120 of the top page displays an add URL button 121 for adding the URL of a web page from which metadata will be acquired, an add list button 122 for adding a list of URLs of web pages from which metadata will be acquired, and a start button 123 for starting acquisition of metadata from the web page of the input URL.

[0061] When the add URL button 121 is selected with the pointer 140, the text of an individual URL can be input. When the add list button 122 is selected with the pointer 140, a list of URLs can be uploaded in, for example, CSV or TSV format, thereby allowing the URL list to be input. A list 131 of input URLs is displayed in the main content section 130. The URL list 131 includes, for example, a content ID and a content name for identifying the content included in the input URL. The content ID and content name may be input manually or automatically.

[0062] When the URL of a web page from which metadata is to be acquired is entered and start button 123 is selected with pointer 140, information processing device 10 executes the metadata acquisition process of FIG. 3, thereby starting the process of acquiring metadata from the web page of the entered URL, and as shown in FIG. 8(B), main content section 130 displays status display 132 indicating the progress of the process of acquiring metadata, and preview display 133 for displaying a preview of the acquired metadata. Also, an item indicating the progress of the process of acquiring metadata is added to URL list 131. In FIG. 8(B), the display indicates that acquisition of metadata for content name "name1" has been completed and that metadata for content name "name2" is currently being acquired.

[0063] When the acquisition of metadata from the web page of the input URL is completed, the URL list 131, status display 132, and preview display 133 in the main content section 130 are displayed in accordance with the completion of the acquisition, and a download button 134 for downloading all the acquired metadata is displayed, as shown in Fig. 9(A). When the download button 134 is operated, all the acquired metadata can be downloaded to the user's terminal device.

[0064] When the item with the content name "name1" in the preview display 133 shown in FIG. 9(A) is selected with the pointer 140, a metadata preview 133A for the content name "name1" is displayed, as shown in FIG. 9(B). The metadata preview 133A includes the content name, the URL of the acquisition source, a download button 135 for individually downloading the metadata for the content name "name1", a metadata list 136, and internal text 137 extracted from the web page of the acquisition source URL. When the download button 135 is operated, the acquired metadata for the content name "name1" can be downloaded to the user's terminal device. The metadata list 136 displays the item text extracted from the internal text 137 in association with the item name. Numbers are assigned to the item names in the metadata list 136. The internal text 137 displays the internal text extracted from the web page of the acquisition source, as shown in FIG. 6. In this way, the user can acquire, display, and download metadata from the input web page.

[0065] After obtaining metadata from the web page of the input URL, if the inspection unit 104 determines that the output data (metadata) contains item text that is not included in the input data (internal text), as shown in Fig. 10, the main content unit 130 displays a URL list 131 and a status display 132 indicating that the task has been completed but an error has occurred. Furthermore, warning information 138 is displayed in the metadata preview 133B of the target content name "name3." The warning information 138 may display information about the item text that is not included in the internal text.

[0066] In addition, in the case of Fig. 10, the download button for downloading the metadata for the content name "name3" is not displayed, which prevents inaccurately extracted metadata from being provided to the user.

[0067] 8 to 10 show an example in which a connection is made to the information processing device 10 via a general-purpose web browser 110 and a website screen for acquiring metadata is displayed, but it is also possible to connect to the information processing device 10 using a dedicated application and display a screen for acquiring metadata.

[0068] As described above, the extraction unit 102 of the information processing device 10 extracts multiple elements from a structured document based on multiple query expressions, and the organizing unit 103 provides the internal text for each of the extracted elements as input data to the generation AI server 20 equipped with a generative model, and outputs output data in response to a prompt that instructs the server to organize the data by associating multiple item texts with multiple item names. This improves the accuracy of organizing data using a generative model.

[0069] (Variation) The present invention is not limited to the above-described embodiment, and various modifications and applications are possible. For example, some of the above-described embodiments may be omitted or replaced, or any configuration may be added.

[0070] In the information processing device 10 of the above embodiment, the inspection unit 104 inspects whether the item text associated with each of a plurality of item names is included in the internal text of the structured document, and the warning unit 105 outputs warning information warning about item text determined by the inspection unit 104 not to be included. This is not limited to this, and the generation AI server 20 may be configured to perform the inspection and output the warning information. For example, the organization unit 103 of the information processing device 10 may prompt the generation AI server 20 to inspect whether the item text associated with each of a plurality of item names is included in the internal text of the structured document, and to embed warning information warning about item text determined not to be included by the inspection into a specified warning item name in the output data.

[0071] In the above embodiment, information processing device 10 receives an input of the URL (Uniform Resource Locator) of a webpage that is the source of the structured document, acquires the structured document from the webpage, and acquires metadata (item text) from the structured document. This is not limiting, and information processing device 10 may acquire structured documents from multiple webpages or multiple websites for a given piece of content, and acquire metadata from multiple structured documents. In this case, information processing device 10 may acquire one item of metadata from a first website and another item of metadata from a second website. Furthermore, it may be possible to select from which website specific metadata is acquired.

[0072] In the above embodiment, the information processing device 10 obtained metadata (item text) from a structured document using an external generation AI server 20, but the information processing device 10 may also be provided with a generation model and use the generation model to obtain metadata (item text) from a structured document.

[0073] In the above embodiment, the internal text extracted by the extraction unit 102 is organized by adding element names (frame names) and preparing input data for the generation AI server 20, but the internal text extracted by the extraction unit 102 may also be input directly to the generation AI server 20.

[0074] In the above embodiment, an example has been described in which the information processing device 10 extracts and acquires metadata (item text) from a website that provides movie information, but the source of the structured document may be any. When the source is a website, the target website may be any website that provides various types of information in a specific field (for example, movies, music, sports, games, etc.).

[0075] The information processing device 10 can be realized using a normal computer, not a dedicated device. For example, the information processing device 10 that executes the above-described processes may be configured by installing a program for executing the above-described functions on a computer from a recording medium that stores the program. Also, one information processing device 10 may be configured by multiple computers operating in cooperation with each other.

[0076] Furthermore, the method for supplying the program to the computer is arbitrary. For example, the program may be supplied via a communication line, a communication network, a communication system, etc. For example, a non-transitory computer-readable recording medium storing the program for executing the above-described functions may be distributed.

[0077] Furthermore, if an OS (Operating System) provides some of the above functions, the functions other than those provided by the OS may be provided by a program.

[0078] The above-described embodiments are intended to facilitate understanding of the present invention and are not intended to limit the present invention. The flowcharts, sequences, elements included in the embodiments, and their arrangements are not limited to those illustrated and can be modified as appropriate. Furthermore, configurations shown in different embodiments can be partially substituted or combined with each other.

[0079] (Invention realized by this embodiment) As described above, this embodiment realizes the following inventions.

[0080] (Appendix 1) an extracting unit for extracting a plurality of elements from the structured document based on a plurality of query expressions, respectively; a reorganization unit that provides the internal texts relating to each of the extracted elements as input data to the generative model, and outputs output data in response to a prompt that instructs that the item texts should be organized by associating them with the item names, respectively; An information processing device comprising:

[0081] (Appendix 2) the structured document is an SGML document, an XML document, or an HTML document; each of the plurality of query expressions is expressed by XPath; The textContent attribute or innerText attribute of each of the plurality of elements is set as the inner text of the corresponding element, The generative model is a large-scale language model 2. The information processing device according to claim 1.

[0082] (Appendix 3) an inspection unit that inspects, for each of the plurality of item names, whether or not an item text associated with the item name is included in an internal text of the structured document; a warning unit that outputs warning information that warns of the item text that is determined not to be included by the inspection unit; 3. The information processing device according to claim 1 or 2, further comprising:

[0083] (Appendix 4) The organizing unit causes the generative model to check whether or not item text associated with each of the plurality of item names is included in the internal text of the structured document, and to embed warning information indicating that the item text is not included in the internal text of the structured document, into a predetermined warning item name in the output data. 3. The information processing device according to claim 1 or 2.

[0084] (Appendix 5) an acquisition unit that receives a designation of an acquisition source and acquires the structured document from the designated acquisition source prior to extraction by the extraction unit; Furthermore, The extractor switches among the plurality of query expressions depending on the specified acquisition source. 5. An information processing device according to any one of appendices 1 to 4.

[0085] (Appendix 6) a frame name is associated with each of the plurality of query expressions; The input data includes information that associates the internal texts with frame names that correspond to query expressions used to extract the internal texts. 6. An information processing device according to any one of appendices 1 to 5.

[0086] (Appendix 7) The prompt further defines synonymous and non-synonymous relationships between terms that may be included in the input data. 7. An information processing device according to any one of appendices 1 to 6.

[0087] (Appendix 8) The information processing device extracting a plurality of elements from the structured document based on a plurality of query expressions, respectively; The internal texts relating to each of the extracted elements are given as input data to the generative model, and output data is output in response to a prompt instructing that the item texts should be organized by associating them with the item names. Information processing methods.

[0088] (Appendix 9) Computer, an extracting unit for extracting a plurality of elements from the structured document based on a plurality of query expressions, respectively; a reorganization unit that provides the internal texts relating to each of the extracted elements as input data to the generative model, and outputs output data in response to a prompt that instructs that the item texts should be organized by associating them with the item names, respectively; A program that functions as a [Industrial Applicability]

[0089] According to the present invention, it is possible to provide an information processing device, an information processing method, and a program that improve the accuracy when organizing data using a generative model. [Explanation of symbols]

[0090] 10 Information processing device, 20 Generation AI server, 11 Processor, 12 Storage device, 13 Communication interface, 14 Input device, 15 Output device, 100 Information processing system, 101 Document acquisition unit, 102 Extraction unit, 103 Organization unit, 104 Inspection unit, 105 Warning unit, 106 Output unit

Claims

1. an extracting unit for extracting a plurality of elements from the structured document based on a plurality of query expressions, respectively; a reorganization unit that provides the internal texts relating to each of the extracted elements as input data to the generative model, and outputs output data in response to a prompt that instructs that the item texts should be organized by associating them with the item names, respectively; An information processing device comprising:

2. the structured document is an SGML document, an XML document, or an HTML document; each of the plurality of query expressions is expressed by an XPath; The textContent attribute or innerText attribute associated with each of the plurality of elements is set as the inner text associated with each of the elements; The generative model is a large-scale language model The information processing device according to claim 1 .

3. an inspection unit that inspects, for each of the plurality of item names, whether or not an item text associated with the item name is included in an internal text of the structured document; a warning unit that outputs warning information that warns of the item text that is determined not to be included by the inspection unit; The information processing device according to claim 1 , further comprising:

4. The organizing unit causes the generative model to check whether or not item text associated with each of the plurality of item names is included in the internal text of the structured document, and to embed warning information indicating that the item text is not included in the internal text of the structured document, into a predetermined warning item name in the output data. The information processing device according to claim 1 .

5. an acquisition unit that receives a designation of an acquisition source and acquires the structured document from the designated acquisition source prior to extraction by the extraction unit; Furthermore, The extractor switches among the plurality of query expressions depending on the specified acquisition source. The information processing device according to claim 1 .

6. a frame name is associated with each of the plurality of query expressions; The input data includes information that associates the internal texts with frame names that correspond to query expressions used to extract the internal texts. The information processing device according to claim 1 .

7. The prompt further defines synonymous and non-synonymous relationships between terms that may be included in the input data. The information processing device according to claim 1 .

8. The information processing device extracting a plurality of elements from the structured document based on a plurality of query expressions, respectively; The internal texts relating to each of the extracted elements are given as input data to the generative model, and output data is output in response to a prompt instructing that the item texts should be organized by associating them with the item names. Information processing methods.

9. Computer, an extracting unit for extracting a plurality of elements from the structured document based on a plurality of query expressions, respectively; a reorganization unit that provides the internal texts relating to each of the extracted elements as input data to the generative model, and outputs output data in response to a prompt that instructs that the item texts should be organized by associating them with the item names, respectively; A program that functions as a

Images