Information processing apparatus, analysis method, and analysis program

Abstract

To support use of various contents.SOLUTION: An information processing apparatus includes: an extraction section that uses a language model to extract at least one of a matter described as an antecedent and a matter described as a consequent in a content to be targeted; and an analysis section that associates, on the basis of an extraction result by the extraction section, a first matter described as an antecedent in a content in which an intermediate matter, which is described as an antecedent in a content and as a consequent in another content, is described as a consequent, with a second matter described as a consequent in a content in which the intermediate matter is described as an antecedent. A result of the association by the analysis section can be used for decision making based on matters described in the content to be targeted.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

[0002] There are known analytical techniques for promoting the use of various documents. One example is the document processing device described in Patent Document 1. This document processing device extracts words and phrases from related source documents based on rules corresponding to the type of the related source document, generates search conditions for related destination documents from the extracted words and phrases, and stores the relationships between related destination documents and related source documents that satisfy the search conditions. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2010-108268 Summary of the Invention [Problem to be solved by the invention]

[0004] The document processing device described in Patent Document 1 analyzes relevance by utilizing the fact that words that can be used to search for related documents are written in specific locations in documents classified into predetermined categories such as "daily reports," "weekly reports," "laws," "ordinances," etc. For this reason, the document processing device described in Patent Document 1 has room for improvement in that the objects of analysis are limited.

[0005] For example, suppose document X contains a statement that "when condition A is satisfied, event B occurs," and another document Y contains a statement that "when event B occurs, event C also occurs." These documents can be said to be related in that they both mention event B. However, unless event B is mentioned in a specific location in document X that corresponds to its type, the document processing device described in Patent Document 1 cannot associate these documents. Furthermore, the document processing device described in Patent Document 1 cannot associate content other than documents (e.g., images).

[0006] The present disclosure has been made in view of the above points, and an exemplary purpose thereof is to provide a technology that enables supporting utilization of various contents. [Means for solving the problem]

[0007] An information processing device according to an exemplary aspect of the present disclosure includes an extraction means for extracting, from a plurality of target contents, at least one of an item described as an antecedent in the content and an item described as a consequent in the content using a machine-learned language model; and an analysis means for, when an item described as an antecedent in some of the plurality of contents and as a consequent in other of the contents is considered an intermediate item, associating, based on the extraction result of the extraction means, a first item described as an antecedent in the content in which the intermediate item is described as a consequent and a second item described as a consequent in the content in which the intermediate item is described as an antecedent.

[0008] An analysis method according to an exemplary aspect of the present disclosure includes an extraction process in which at least one processor uses a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the content and an item described as a consequent in the content; and an analysis process in which, when an item described as an antecedent in some of the plurality of contents and described as a consequent in other of the contents is considered an intermediate item, the analysis process associates, based on the extraction results of the extraction process, a first item described as an antecedent in the content in which the intermediate item is described as a consequent and a second item described as a consequent in the content in which the intermediate item is described as an antecedent.

[0009] An analysis program according to an exemplary aspect of the present disclosure causes a computer to function as an extraction means that uses a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the content and an item described as a consequent in the content, and an analysis means that, when an item that is described as an antecedent in some of the plurality of contents and as a consequent in other of the contents is considered an intermediate item, associates, based on the extraction result of the extraction means, a first item described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item described as a consequent in the content in which the intermediate item is described as an antecedent. [Effects of the Invention]

[0010] According to an exemplary aspect of the present disclosure, an exemplary effect is achieved in that it becomes possible to support the utilization of various content. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 2] FIG. 1 is a flow chart showing the flow of an analysis method according to the present disclosure. [Figure 3]FIG. 10 is a block diagram showing a configuration of another information processing device according to the present disclosure. [Figure 4] FIG. 10 is a diagram illustrating an example of extraction and association of document descriptions. [Figure 5] FIG. 1 is a diagram illustrating an example of a logic model. [Figure 6] 4 is a flowchart showing the flow of processing executed by the information processing device shown in FIG. 3. [Figure 7] FIG. 10 is a flowchart showing the process flow for accepting a question and presenting an answer. [Figure 8] FIG. 10 is a diagram illustrating another example of extraction and association of document descriptions. [Figure 9] FIG. 10 is a diagram showing yet another example of extraction and association of document descriptions. [Figure 10] 4 is a flowchart showing another example of the process executed by the information processing device shown in FIG. 3. [Figure 11] FIG. 10 is a diagram showing yet another example of extraction and association of document descriptions. [Figure 12] 10 is a flowchart showing yet another example of the process executed by the information processing device shown in FIG. [Figure 13] FIG. 10 is a diagram showing yet another example of extraction and association of document descriptions. [Figure 14] FIG. 1 is a block diagram illustrating a configuration of a computer that functions as an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technologies (part or all of the products or methods) employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technologies employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, the effects mentioned in the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, embodiments that do not exhibit the effects mentioned in the exemplary embodiments shown below may also be included in the scope of the present invention.

[0013] First Exemplary Embodiment A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is a basic form for each of the exemplary embodiments described below. Note that the scope of application of each technique employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technique employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technique shown in the drawings referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0014] (Configuration of information processing device 1) The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As shown in Fig. 1, the information processing device 1 includes an extraction unit 101 and an analysis unit 102.

[0015] The extraction unit 101 uses a machine-learned language model to extract, from a plurality of target contents, at least one of an item described as an antecedent in the content and an item described as a consequent in the content.

[0016] The "item stated as the antecedent" is a pair of the "item stated as the consequent," and if the "item stated as the antecedent" is true, then the "item stated as the consequent" is also true. The "antecedent" can be rephrased as, for example, a "condition," "premise," or "input," and the "consequent" can be rephrased as, for example, a "result," "consequence," "conclusion," or "output."

[0017] Furthermore, the "content" may be anything that includes both the antecedent and the consequent. For example, the "content" may be a document, i.e., a text-formatted content, an image-formatted content, or a content that includes both text and images.

[0018] Furthermore, the "language model" may be one that has been machine-learned so as to be able to extract, from the content, at least one of an item described as an antecedent and an item described as a consequent. For example, if the target content is text data, the language model may be a model that has been machine-learned to learn the arrangement of components (such as words) in a sentence or the arrangement of sentences in a piece of writing. For example, if the target content is image data, the language model may be a model that has been machine-learned to learn the relationship between the image data and an item corresponding to the antecedent and / or an item corresponding to the consequent in the object represented by the image data. Furthermore, the language model may be a combination of a model that extracts an item corresponding to the antecedent and / or an item corresponding to the consequent from image data and a model that extracts an item corresponding to the antecedent and / or an item corresponding to the consequent from text data.

[0019] Furthermore, when the target content is in a format other than text, the extraction unit 101 may convert the content into a text format and then perform the extraction described above. For example, when the target content is image data, the extraction unit 101 may generate text data using a generative model that generates text indicating an object represented by the image data. The extraction unit 101 may then extract from the text data an item corresponding to the antecedent and / or an item corresponding to the consequent using a language model. For example, when the target content is audio data, the extraction unit 101 may convert the audio data into text data and then extract from the text data an item corresponding to the antecedent and / or an item corresponding to the consequent using a language model. Note that the process of converting content into text format may be performed by a block separate from the extraction unit 101 provided in the information processing device 1, or may be performed by a device other than the information processing device 1.

[0020] The analysis unit 102 associates a first item described as an antecedent in a piece of content in which an intermediary item is described as a consequent, with a second item described as a consequent in a piece of content in which an intermediary item is described as an antecedent, based on the extraction result of the extraction unit 101. Here, an "intermediary item" means an item that is described as an antecedent in one piece of content among a plurality of target contents, and is described as a consequent in another piece of content.

[0021] As described above, the information processing device 1 according to this exemplary embodiment is configured to include an extraction unit 101 that uses a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the content and an item described as a consequent in the content, and an analysis unit 102 that, when an item that is described as an antecedent in some of the plurality of contents and as a consequent in other of the contents is considered an intermediate item, associates, based on the extraction result of the extraction unit 101, a first item that is described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item that is described as a consequent in the content in which the intermediate item is described as an antecedent.

[0022] Here, in the content in which the first item is described as the antecedent, the intermediate item is described as the consequent. In other words, according to the content, if the first item is true, then the intermediate item is true.

[0023] On the other hand, in the content where the second item is described as the consequent, the intermediate item is described as the antecedent. In other words, according to the content, if the intermediate item is true, the second item is true.

[0024] Therefore, according to the two pieces of content, it can be said that there is a relationship in which if the first item is true, the intermediary item is true, and if the intermediary item is true, the second item is true. From this relationship, it can also be said that there is a relationship in which if the first item is true, the second item is true. The analysis unit 102 can extract first items and second items that have such a relationship. In this way, associating items described in different pieces of content creates new knowledge and promotes the use of content.

[0025] Furthermore, since a language model is used for extraction by the extraction unit 101, the target content is not limited to a specific type of document, but various types of content can be targeted. As described above, the information processing device 1 has the effect of being able to support the utilization of various types of content.

[0026] The analysis result by the analysis unit 102, i.e., the result of associating the first item with the second item, can be used for various purposes. For example, the information processing device 1 may present the result of associating the first item with the second item to the user of the information processing device 1, thereby providing the user with new knowledge. The result of associating the first item with the second item can also be used for decision-making based on the items described in the target content. For example, suppose the first item is "consuming 50g or more of food A daily" and the second item is "increasing lifetime income." In this case, by associating these items, a decision can be made to consume 50g or more of food A daily based on the items described in each content.

[0027] Furthermore, the information processing device 1 may present to the user at least one of content in which the first item is described and content in which the second item is described, based on the result of associating the first item with the second item. This allows the user to make more accurate decisions based on the context in which the first item / second item is described in the content. Furthermore, as will be described in detail in exemplary embodiment 2, the result of associating the first item with the second item can also be used to generate an answer to a question from the user that is appropriate for the content.

[0028] (Analysis Program) The functions of the information processing device 1 described above can also be realized by a program. The analysis program according to this exemplary embodiment causes a computer to function as: extraction means for extracting, from a plurality of target contents, at least one of an antecedent and a consequent from the target contents using a machine-learned language model; and analysis means for associating, when an intermediate item is described as an antecedent in one of the plurality of contents and as a consequent in another of the plurality of contents, a first item described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item described as a consequent in the content in which the intermediate item is described as an antecedent, based on the extraction result of the extraction means. This analysis program advantageously supports the utilization of various content.

[0029] (Analysis method flow) The flow of the analysis method according to this exemplary embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the analysis method. Note that the execution entity of each step in this analysis method may be a processor provided in the information processing device 1, or a processor provided in another device, or each step may be executed by a processor provided in a different device.

[0030] In S1 (extraction process), at least one processor uses a machine-learned language model to extract from multiple target contents at least one of the items described as antecedents in the contents and the items described as consequents.

[0031] In S2 (analysis processing), at least one processor associates a first item described as an antecedent in a content in which an intermediary item is described as a consequent, with a second item described as a consequent in a content in which the intermediary item is described as an antecedent, based on the extraction result in S1. As described above, an intermediary item is an item that is described as an antecedent in some content among the plurality of contents, and as a consequent in other content.

[0032] As described above, the analysis method according to this exemplary embodiment includes an extraction process in which at least one processor uses a machine-learned language model to extract, from a plurality of target content items, at least one of an item described as an antecedent in the content item and an item described as a consequent in the content item, and an analysis process in which, when an item described as an antecedent in one of the plurality of content items and as a consequent in another of the plurality of content items is an intermediate item, the extraction process associates, based on the extraction result of the extraction process, a first item described as an antecedent in the content item in which the intermediate item is described as a consequent with a second item described as a consequent in the content item in which the intermediate item is described as an antecedent. Therefore, the analysis method according to this embodiment has the advantage of being able to support the utilization of various content items.

[0033] Second Exemplary Embodiment A second exemplary embodiment, which is one example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be assigned the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technology employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. Furthermore, each technology shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs.

[0034] (Configuration of information processing device 1A) The configuration of an information processing device 1A according to this exemplary embodiment will be described with reference to FIG. 3. FIG. 3 is a block diagram showing the configuration of the information processing device 1A. The information processing device 1A is a device having a function of supporting content utilization. The information processing device 1A may be a device whose main function is to support content utilization, or may be a general-purpose device having other functions as well. Furthermore, the information processing device 1A may be a stationary device or a portable device.

[0035] As shown in the figure, information processing device 1A includes a control unit 10A that controls each unit of information processing device 1A and a storage unit 11A that stores various data used by information processing device 1A. Information processing device 1A also includes a communication unit 12A that enables information processing device 1A to communicate with other devices, an input unit 13A that accepts input to information processing device 1A, and an output unit 14A that enables information processing device 1A to output data. Control unit 10A includes an extraction unit 101A, an analysis unit 102A, a model generation unit 103A, an inference unit 104A, a reception unit 105A, a response generation unit 106A, and a presentation unit 107A. Storage unit 11A stores a language model 111A and a logic model 112A. Details of model generation unit 103A, inference unit 104A, reception unit 105A, response generation unit 106A, and logic model 112A will be described later.

[0036] Similar to the extraction unit 101 in exemplary embodiment 1, the extraction unit 101A uses the language model 111A, which is a machine-learned language model, to extract from multiple target contents at least one of the items described as antecedents in the contents and the items described as consequents.

[0037] The following describes an example in which the target content is a text document. The document may be, for example, an academic paper, text extracted from a webpage introducing a product or service, user reviews, or a message posted on a social networking service (SNS). The target content may also be limited to content in a specific field. For example, by limiting the target content to medical papers, specialized technical knowledge in the medical field can be analyzed. Furthermore, by limiting the target content to healthcare-related documents, the information processing device 1A can be used in healthcare. As described in the first exemplary embodiment, the target content is not limited to text documents, and any content in any format can be analyzed. Therefore, the term "document" in the following description can be replaced with "content" in any format.

[0038] The language model 111A may be one that has been machine-learned so as to be able to extract, from the content to be analyzed, at least one of an item described as an antecedent and an item described as a consequent, similar to the language model described in exemplary embodiment 1. As described above, the content to be analyzed in this exemplary embodiment is a text document, and therefore, as the language model 111A, a model that has been machine-learned to learn the arrangement of components (such as words) in a sentence or the arrangement of sentences in a piece of writing may be applied.

[0039] The information processing device 1A does not necessarily need to include the language model 111A, and may use a language model 111A stored in a device external to the information processing device 1A. In this case, the extraction unit 101A instructs the external device including the language model 111A to extract matters described as antecedents and / or consequents of a document, and obtains from the device the matters extracted by the device using the language model 111A.

[0040] Similar to the analysis unit 102 of exemplary embodiment 1, when an item that is described as an antecedent in some of the multiple contents (documents in this exemplary embodiment) to be analyzed and described as a consequent in other contents is considered to be an intermediate item, the analysis unit 102A associates, based on the extraction results of the extraction unit 101, a first item that is described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item that is described as a consequent in the content in which the intermediate item is described as an antecedent.

[0041] The presentation unit 107A presents various pieces of information for supporting content utilization to the user of the information processing device 1A. For example, the presentation unit 107A presents a logic model 112A to the user. For example, the presentation unit 107A may present the information by causing the output unit 14A to output the information, or by causing a terminal device or the like carried by the user to output the information. The manner of presentation is not particularly limited. For example, when presenting information that is preferably presented as an image, such as the logic model 112A, the presentation unit 107A may simply display and output the information. When presenting other information, the presentation unit 107A may display and output the information, output it as audio, or print it out.

[0042] As described above, the information processing device 1A according to this exemplary embodiment includes an extraction unit 101A that extracts, from a plurality of target contents (documents in this exemplary embodiment) using a machine-learned language model, at least one of an item described as an antecedent in the content and an item described as a consequent in the content, and an analysis unit 102A that, when an item that is described as an antecedent in one of the plurality of contents and as a consequent in another of the plurality of contents is considered an intermediate item, associates, based on the extraction result of the extraction unit 101, a first item described as an antecedent in the content in which the intermediate item is described as a consequent with a second item described as a consequent in the content in which the intermediate item is described as an antecedent. This provides the effect of enabling support for the utilization of various content.

[0043] (Extraction and Association Example 1: Start by extracting the antecedent) Extraction by extraction unit 101A and association by analysis unit 102A will be described based on the specific example shown in Fig. 4. Fig. 4 is a diagram showing an example of extraction and association of document descriptions by extraction unit 101A and analysis unit 102A. Note that the description of each document shown in Fig. 4 is intended to explain extraction and association by extraction unit 101A and analysis unit 102A, and whether the content is correct or not is not an issue here. The same applies to other examples described below.

[0044] In the example of Figure 4, documents D1 to D4 recorded in a DB (Data Base) are the content to be analyzed. Note that for simplicity, only four target documents are shown in the figure, but the number of target documents is not particularly limited as long as there is more than one. Furthermore, the target documents do not necessarily have to be recorded in a single DB; documents recorded in multiple DBs or in multiple storage devices can also be the subject of analysis.

[0045] In the example of Fig. 4, the extraction unit 101A reads out documents D1 to D4 recorded in the DB one by one, and extracts from the read out documents the items described as antecedents in the documents. Fig. 4 shows extraction from document D1. As shown in the figure, document D1 describes an item M11, "increasing the number of steps taken per day," as the antecedent, and an item M12, "increasing healthy life expectancy," as the consequent corresponding to this antecedent.

[0046] The extraction unit 101A inputs a prompt, which instructs the language model 111A to extract an antecedent described in a document, together with the document read from the DB, thereby extracting the antecedent described in the document from the document. For example, as shown in FIG. 4, the extraction unit 101A may input a standard prompt P11, such as "Please extract an antecedent described in this document," and the document D1 to the language model 111A. This allows the extraction of an antecedent M11 from the document D1, as shown in the figure. The extracted antecedent M11 is text data. The extraction unit 101A may also extract multiple antecedents described in a single document from the document.

[0047] Next, the extraction unit 101A extracts documents in which the item extracted as described above is described as a consequent. For example, as shown in FIG. 4, the extraction unit 101A may input a standard prompt P12, such as "Please extract documents in which this item is described as a consequent," and an item M11 to the language model 111A. Furthermore, the extraction unit 101A may specify documents D1 to D4 recorded in the DB as extraction candidates. As a result, in the example of FIG. 4, document D4 is extracted. Document D4 contains an item M41, "record the number of steps taken each day," as the antecedent, and an item M42, "increase the number of steps taken per day," as the consequent corresponding to this antecedent.

[0048] Here, item M42 and item M11 in the example of FIG. 4 are the same. However, because extraction unit 101A performs extraction using language model 111A, even if there is a difference in expression, if there is a description that matches item M11 in content, it is possible to extract a document containing that description. Note that if multiple items described as antecedents are extracted from document D1, extraction unit 101A attempts to extract documents in which the items are described as consequents for each extracted item. Furthermore, if extraction unit 101A does not extract a corresponding document, it reads other documents from the DB and extracts from those other documents the items described as antecedents in those documents.

[0049] Next, extraction unit 101A inputs document D4 extracted as described above and prompt P13, which instructs language model 111A to extract the matter described as an antecedent in document D4, to extract the matter described as an antecedent in document D4. As a result, as shown in the figure, it is possible to extract matter M41 from document D4.

[0050] The extraction unit 101A also inputs document D1 and a prompt P14, which instructs the language model 111A to extract the matter described as the consequent in document D1, to extract the matter described as the consequent in document D1. As a result, it is possible to extract matter M12 from document D1 as shown in the figure.

[0051] The analysis unit 102A associates the item M41 extracted from document D4 with the item M12 extracted from document D1 in the above manner. This leads to the finding that "recording the number of steps taken each day" "extends healthy life expectancy." Note that the "intermediate items" in the example of Figure 4 are items M11 and M42.

[0052] As described above, extraction unit 101A may use language model 111A to (1) extract an item (item M11 in the example of FIG. 4) described as an antecedent in a first content (document D1 in the example of FIG. 4), which is one of the multiple contents to be analyzed, (2) extract a second content (document D4 in the example of FIG. 4) from the multiple contents in which the extracted item is described as a consequent, (3) extract an item (item M41 in the example of FIG. 4) described as an antecedent in the second content as a first item, and (4) extract an item (M12 in the example of FIG. 4) described as a consequent in the first content as a second item. This provides the effect of logically associating the items described in the multiple contents to derive new knowledge.

[0053] (About the logic model) Based on the above-described association results by the analysis unit 102A, the model generation unit 103A generates a logic model that is a model showing the logical relationships between the items described in the multiple pieces of content to be analyzed (documents in this exemplary embodiment). By including the model generation unit 103A, the information processing device 1A can obtain, in addition to the effects of the information processing device 1, the effect of being able to model the items described in the multiple pieces of content to be analyzed based on their logical relationships.

[0054] The generated logic model is stored in storage unit 11A as logic model 112A. Logic model 112A will be described below with reference to Fig. 5. Fig. 5 is a diagram showing an example of logic model 112A. Note that, although logic model 112A is shown in a graph format in Fig. 5, logic model 112A does not necessarily have to be in a graph format as long as it shows the logical relationships between the items.

[0055] Logic model 112A-1 shown in FIG. 5 is an example of logic model 112A generated by model generation unit 103A. In logic model 112A-1, first and second items extracted from multiple pieces of content (documents in this exemplary embodiment) to be analyzed are shown as nodes. Logic model 112A-1 indicates the logical relationship between each node and can also be expressed as a logic graph. As described above, the first item is the item described as the antecedent in the content in which the intermediary item is described as the consequent, and the second item is the item described as the consequent in the content in which the intermediary item is described as the antecedent.

[0056] In logic model 112A-1, the relationships between nodes, i.e., the relationships between the items described in the content, are represented by arrow edges and dashed edges. Nodes connected by arrow edges are in an antecedent-consequent relationship, with the node at the base of the arrow corresponding to the antecedent and the node at the tip of the arrow corresponding to the consequent. For example, in logic model 112A-1, the node corresponding to item M41 and the node corresponding to item M42 shown in FIG. 4 are connected by an arrow edge. The node corresponding to item M41 is at the base of the arrow, and the node corresponding to item M42 is at the tip of the arrow. This indicates that item M41 is the antecedent and item M42 is the consequent. The relationship between these nodes is identified from document 4D (see FIG. 4) describing items M41 and M42. Although FIG. 4 does not describe the extraction of item M42 from document D4, extraction unit 101A can extract item M42 by inputting prompt P14 and document D4 shown in FIG. 4 into language model 111A. That is, the extraction unit 101A may extract intermediate items in order to generate the logic model 112A-1.

[0057] On the other hand, the dashed edges are based on the associations made by the analysis unit 102A, and each node connected by a dashed edge represents an intermediate item. Specifically, the nodes corresponding to items M42, M11, and M21 are intermediate items. Note that items M21 and M22 are items extracted from document D2 shown in Figure 4 (the former is the antecedent and the latter is the consequent).

[0058] In this way, logic model 112A-1 shows the logical relationships between the items described in different documents via the nodes of intermediate items. Therefore, by presenting logic model 112A to the user of information processing device 1A, the user can recognize the relationships. Presentation of logic model 112A-1 may be performed by presentation unit 107A. Furthermore, when presenting logic model 112A-1, presentation unit 107A may also present the documents from which the items were extracted, in association with the items described in each document. For example, in the example of FIG. 5, presentation unit 107A may present information indicating that the items M41 and M42 were extracted from document D4, in association with the nodes of items M41 and M42 or the edges connecting these nodes.

[0059] The model generation unit 103A can generate the logic model 112A-1 by representing each item that has been associated by the analysis unit 102A, i.e., the first item and the second item, as a node and expressing the relationship between them as an edge. Furthermore, as shown in Fig. 5, the model generation unit 103A may also include intermediate items in the nodes of the logic model 112A-1.

[0060] Furthermore, the model generation unit 103A can also associate items described in three or more documents in the logic model 112A. For example, in the example of Figures 4 and 5, in document D3, "healthy life expectancy will be extended" (hereinafter referred to as item M31) is described as the antecedent, and "the foundations of the local community will be strengthened" (hereinafter referred to as item M32) is described as the consequent corresponding to this antecedent.

[0061] In this case, the extraction unit 101A extracts the item M31 from the document D3 and attempts to extract a document in which the item M31 is written as a consequent from among the documents D1 to D4. This attempt results in the extraction of the document D1. The extraction unit 101A then extracts the item M11 written as an antecedent in the document D1 and the item M32 written as a consequent in the document D3. This allows the analysis unit 102A to associate the items M11 and M32. The item M11 is also associated with the item M41. Therefore, based on the associations made by the analysis unit 102A, the model generation unit 103A generates a logic model 112A that indicates a logical relationship in which, if the item M41 holds, then the items M42 and M11 (and M21) hold; if the item M11 holds, then the items M12 and M31 hold; and if the item M31 holds, then the item M32 holds.

[0062] The model generation unit 103A can also update the logic model 112A. For example, after generating the logic model 112A-1 shown in FIG. 5, a new document D7 is analyzed, and item M21 is associated with item M72 described in document D7 via item M22 as an intermediate item. In this case, the model generation unit 103A adds to the logic model 112A-1 a node corresponding to item M72 and a node corresponding to item M71 described in document D7 as an antecedent for item M72. The model generation unit 103A then connects the node corresponding to item M71 and the node corresponding to item M22 with a dashed edge, and also connects the node corresponding to item M71 and the node corresponding to item M72 with an arrow edge. This updates the logic model 112A-1.

[0063] (Regarding relationship information) Furthermore, the presentation unit 107A may present relationship information indicating the relationship between items described in each document in association with those items. The relationship information may be extracted by the extraction unit 101A. For example, when extracting an item described as an antecedent from a document, the extraction unit 101A may input a prompt to the language model 111A requesting extraction of a description indicating the relationship between the item to be extracted and the item described as a consequent corresponding to that item. The description extracted in this manner can be used as relationship information.

[0064] For example, if document D1 in the example of Figure 4 contains a formula that indicates the relationship between the increase in the number of steps taken per day and the degree to which healthy life expectancy is extended, extraction unit 101A can extract this formula as relationship information. Then, presentation unit 107A can present the extracted formula as relationship information. For example, presentation unit 107A may display the extracted formula in association with an edge connecting a node corresponding to item M11 and a node corresponding to item M12.

[0065] The inference unit 104A uses the above-mentioned relationship information to infer the relationship between the first item and the second item. This will be described based on the logic model 112A-2 shown in FIG. 5. The logic model 112A-2 includes nodes corresponding to an item M51 listed as an antecedent in document D5, an item M52 listed as a consequent in document D5, an item M61 listed as an antecedent in document D6, and an item M62 listed as a consequent in document D6. Of these, items M52 and M61 are intermediate items, and items M51 and M62 are related via these items.

[0066] In logic model 112A-2, the edge connecting the node corresponding to item M51 and the node corresponding to item M52 is associated with relevance information RI1 indicating the relationship between these items. Similarly, the edge connecting the node corresponding to item M61 and the node corresponding to item M62 is associated with relevance information RI2 indicating the relationship between these items.

[0067] Relevance information RI1 is information extracted from document D5 and is a mathematical formula showing the relationship between the period of time a healthy weight is maintained and the amount of investment for weight management. According to this formula, the period of time a healthy weight is maintained is proportional to the amount of investment for weight management, with the proportionality constant being a. Relevance information RI2 is information extracted from document D6 and is a mathematical formula showing the relationship between the rate of reduction in medical expenses and the period of time a healthy weight is maintained. According to this formula, the rate of reduction in medical expenses is proportional to the period of time a healthy weight is maintained, with the proportionality constant being b.

[0068] The inference unit 104A uses this relevance information to infer the relationship between item M51 (corresponding to the first item described above) and item M62 (corresponding to the second item described above). Specifically, from the relevance information RI1 and the relevance information RI2, the inference unit 104A infers that the relationship between the investment amount for weight management and the reduction rate of medical expenses is expressed by the formula "(investment amount for weight management) = (reduction rate of medical expenses) / a·b".

[0069] Furthermore, based on the above-described inference results, the inference unit 104A can infer the rate of reduction in medical expenses when the amount of investment for weight management is a certain value, or can infer the amount of change in the rate of reduction in medical expenses when the amount of investment for weight management changes by a certain value. In this way, the inference unit 104A can perform various simulations based on the above-described inference results. The inference results by the inference unit 104A may be recorded in association with the logic model 112A.

[0070] As described above, the information processing device 1A includes an inference unit 104A that infers the relationship between a first item and a second item using first relationship information (RI1 in the example of logic model 112A-2) that indicates the relationship between a first item and an intermediary item, extracted from content in which the first item (M51 in the example of logic model 112A-2) is described as an antecedent and the intermediary item is described as a consequent, and second relationship information (RI2 in the example of logic model 112A-2) that indicates the relationship between the intermediary item and the second item, extracted from content in which the intermediary item is described as an antecedent and the second item (M62 in the example of logic model 112A-2) is described as a consequent. This provides the effect of being able to obtain analysis results on the relationship between the first item and the second item, in addition to the effects provided by the information processing device 1. Furthermore, by utilizing the analysis results, various simulations can be performed based on the inference results.

[0071] The inference unit 104A can also infer the relationship between items described in three or more documents in a similar manner. Furthermore, when the relationship information is a mathematical formula, the inference unit 104A can estimate the relationship by modifying the formula. In this case, the extraction unit 101A may include a statement explicitly instructing the extraction of a mathematical formula in a prompt when extracting relationship information from a document.

[0072] Furthermore, the relationship information is not limited to mathematical formulas and may be, for example, text. In this case, the inference unit 104A may input the first item, the second item, and each piece of relationship information (the first relationship information and the second relationship information described above) into the language model 111A and infer the relationship between the first item and the second item derived from the relationship information. For example, suppose that the text extracted as the first relationship information is "The period for which a healthy weight is maintained is proportional to the amount of investment for weight management," and the text extracted as the second relationship information is "The rate of reduction in medical expenses is proportional to the period for which a healthy weight is maintained." In this case, the extraction unit 101A inputs these texts into the language model 111A, thereby obtaining an inference result such as "The period for which a healthy weight is maintained is proportional to the rate of reduction in medical expenses."

[0073] (About accepting questions and generating answers) The following describes the receiving unit 105A and the answer generating unit 106A. The receiving unit 105A receives a question input from a user. For example, the receiving unit 105A may receive the question input via the communication unit 12A, or may receive the question input via the input unit 13A. The question may be input as text or as voice. In the latter case, the receiving unit 105A may convert the input voice into text and then process it.

[0074] The answer generation unit 106A generates an answer to the question using the logic model 112A. The method of generating the answer is not particularly limited. For example, the answer generation unit 106A may generate an answer using the language model 111A. In this case, the answer generation unit 106A may generate a prompt that includes the question received by the reception unit 105A and the logic model 112A and instructs the unit 106A to generate an answer to the question based on the logic model 112A. The answer generation unit 106A can generate an answer based on the logic model 112A by inputting such a prompt into the language model 111A.

[0075] Furthermore, instead of inputting logic model 112A itself into language model 111A, answer generation unit 106A may detect items related to the user's question from among the items included in logic model 112A, and input the detected items into language model 111A. This example will be described later with reference to FIG. 7.

[0076] The answer generated by answer generation unit 106A is presented to the user by presentation unit 107A. Note that the person who inputs the question and the person to whom the answer is presented may be the same person or different people. The answer may be presented in the form of a text display output or a voice output format.

[0077] As described above, information processing device 1A includes receiving unit 105A that receives input of a question, answer generating unit 106A that generates an answer to the question using logic model 112A, and presentation unit 107A that presents the generated answer. In addition to the effects of information processing device 1, this provides the effect of being able to return an answer to a question from a user based on the logical relationship between the descriptions in each piece of content that is the target of analysis.

[0078] For example, suppose that logic model 112A-1 of FIG. 5 has been generated and stored in memory unit 11A, and reception unit 105A receives the question, "What should I do to extend my healthy lifespan?" Here, logic model 112A-1 associates item M41 with item M12 related to "healthy lifespan." Therefore, by using logic model 112A-1, answer generation unit 106A can generate an answer that includes item M41 (for example, "It is effective to record the number of steps you take every day").

[0079] 5 has been generated and stored in the memory unit 11A, and the receiving unit 105A receives the question, "What effect can I expect if I decide to go to a fitness club every week to manage my weight?" Here, in the logic model 112A-2, the item M62 is associated with the item M51 related to "weight management." Therefore, the answer generating unit 106A can generate an answer including the item M62 (for example, "You can expect a reduction in medical expenses") by using the logic model 112A-2.

[0080] Furthermore, the answer generation unit 106A may generate an answer taking into consideration the result of the inference by the inference unit 104A. In this case, for example, the answer generation unit 106A can generate an answer such as, "If the investment amount for going to a fitness club for one year is x yen, the expected reduction in medical expenses is y yen."

[0081] The answer generation unit 106A may also generate an answer taking into consideration attribute information indicating the user's attributes. Here, the "user" refers to the asker of the question or the person to whom the answer is to be presented. Any attribute may be taken into consideration, and for example, attribute information indicating the user's age, sex, occupation, personality, past behavior history, etc. may be used. The user's attribute information may be input in advance, or part of the question accepted by the acceptance unit 105A may be used as the attribute information.

[0082] For example, answer generation unit 106A may use attribute information indicating the user's occupation to generate an answer with content appropriate to the user's occupation. To cite a specific example, when a question such as "What side jobs do you recommend?" is input, answer generation unit 106A may generate a prompt indicating the user's occupation, including the input question and logic model 112A, and input the generated prompt to language model 111A. This makes it possible to generate an answer indicating a side job appropriate to the user's occupation, based on logic model 112A.

[0083] (Processing flow: Analysis method) The flow of processing executed by the information processing device 1A will be described with reference to Fig. 6. Fig. 6 is a flow diagram showing the flow of processing executed by the information processing device 1A. The flow of Fig. 6 includes each step of the analysis method according to this exemplary embodiment.

[0084] In S11, the extraction unit 101A selects one document from among the multiple documents to be analyzed. The selection method is arbitrary, but a document that has already been selected should not be selected again. Note that documents to be selected may be designated in advance. For example, as in the example of FIG. 4, a specific DB may be designated. In this case, in S11, the extraction unit 101A selects one document from the documents recorded in the designated DB.

[0085] In S12 (extraction process), the extraction unit 101A extracts, from the document acquired in S11 (corresponding to the first content described above), matters described as antecedents in the document, using the language model 111A.

[0086] In S13, the extraction unit 101A extracts documents (corresponding to the above-mentioned second content) in which the matter extracted in S12 is described as a consequent from among the multiple documents to be analyzed, using the language model 111A. Note that if there is no document in which the matter extracted in S12 is described as a consequent, i.e., no second content, among the multiple documents to be analyzed, the process proceeds to S17 without performing the processes of S13 to S16.

[0087] In S14 (extraction process), the extraction unit 101A extracts, as a first item, an item described as an antecedent in the document extracted in S13, i.e., the second content. As described above, the language model 111A is also used to extract the first item.

[0088] In S15 (extraction process), the extraction unit 101A extracts the matter described as a consequent in the document selected in S11, i.e., the first content, as a second matter. As described above, the language model 111A is also used to extract the second matter. Note that the process of S15 may be performed before S14, or may be performed in parallel with S14.

[0089] In S16 (analysis process), the analysis unit 102A associates the items described in the document. Specifically, the analysis unit 102A associates a first item (extracted in S14) described as an antecedent in a document in which an intermediary item is described as a consequent, i.e., a second content, with a second item (extracted in S15) described as a consequent in a document in which the intermediary item is described as an antecedent, i.e., a first content. Note that the process of S16 may be performed after S17, which will be described later.

[0090] In S17, the extraction unit 101A determines whether the processes from S11 onward have been performed on all of the documents to be analyzed. If the determination in S17 is NO, the process returns to S11, where the extraction unit 101A selects the next document to be processed. On the other hand, if the determination in S17 is YES, the process proceeds to S18.

[0091] In S18, model generation unit 103A generates logic model 112A that indicates the logical relationships between the items described in the multiple documents to be analyzed, based on the association results of S16. If logic model 112A has already been generated, model generation unit 103A updates the existing logic model 112A.

[0092] In S19, the presentation unit 107A displays the logic model 112A generated or updated in S18 on a display device (which may be included in the information processing device 1A or another device), thereby completing the processing of FIG.

[0093] (Process flow: Accepting questions and providing answers) The flow of processing when information processing device 1A accepts a question and presents an answer will be described with reference to Fig. 7. Fig. 7 is a flow diagram showing the flow of processing when accepting a question and presenting an answer.

[0094] In S21, the reception unit 105A receives an input of a question from a user. There is no particular limitation on how the question is received. For example, the reception unit 105A may receive the input of the question via the communication unit 12A or via the input unit 13A.

[0095] In S22, the answer generation unit 106A detects, from the logic model 112A, an item related to the question received in S21. For example, the answer generation unit 106A may generate a prompt that includes the received question and the logic model 112A and instructs the user to select an item related to the question from the logic model 112A, and input the generated prompt to the language model 111A. This allows the answer generation unit 106A to detect an item related to the question received in S21 based on the output of the language model 111A. The answer generation unit 106A also detects, in the logic model 112A, an item associated with the item detected as described above as an item related to the question.

[0096] For example, suppose that logic model 112A-1 shown in FIG. 5 is stored in storage unit 11A and a question such as "I've been diagnosed with high blood pressure and am thinking about changing my lifestyle habits. Are there any good ways to do this?" is received. In this case, answer generation unit 106A can detect item M22 from logic model 112A-1 as an item related to the question by the above-described search. Then, answer generation unit 106A can also detect items M21, M11, M42, and M41 associated with item M22 in logic model 112A-1 as items related to the question. Note that intermediate items may be excluded from detection, in which case only item M41 is detected as an item related to item M21.

[0097] In S23, the answer generation unit 106A acquires the inference result of the inference unit 104A related to the item detected in S22. The inference by the inference unit 104A may be performed in advance (for example, after S16 in the flow of FIG. 6), or may be performed after S22 and before S23. Furthermore, the answer generation unit 106A may acquire relevance information related to the item detected in S22 in addition to the inference result of the inference unit 104A. Extraction of the relevance information may also be performed in advance, similar to the inference by the inference unit 104A, or may be performed after S22 and before S23.

[0098] In S24, the answer generation unit 106A generates an answer to the question received in S21. More specifically, the answer generation unit 106A may generate a prompt that includes the question received in S21, the items detected in S22, and the inference result acquired in S23, and instructs the unit 106A to generate an answer to the question based on the items and the inference result. The answer generation unit 106A can then generate an answer to the question by inputting such a prompt into the language model 111A. Furthermore, if relevance information has been acquired in S23, the answer generation unit 106A may also include the acquired relevance information in the prompt.

[0099] In S25, the presenting unit 107A presents the answer generated in S24 to the user, which ends the processing in Fig. 7. After S25, the processing may return to S21 to accept an additional question.

[0100] Alternatively, extraction and association of items from the document to be analyzed may be performed after the question is received. In this case, extraction unit 101A searches the document to be analyzed for the items asked in the received question, defines the items obtained by the search as the first or second items, and extracts the second or first items related to the first or second items. This allows answer generation unit 106A to generate an answer based on the extracted second or first items.

[0101] For example, suppose a question is received that asks, "I've been diagnosed with high blood pressure, so I'm thinking about changing my lifestyle. Are there any good ways to do this?" When a question asking about an antecedent, such as a measure or condition, is received, the extraction unit 101A attempts to extract a document in which "high blood pressure" is described as a consequent, thereby extracting document D2. The extraction unit 101A then extracts item M22, "high blood pressure improved," which is described as a consequent in document D2, and sets this as the second item. Next, the extraction unit 101A attempts to extract a document in which item M21, which is described as an antecedent corresponding to item M22 in document D2, is described as a consequent, thereby detecting document D4. The extraction unit 101A then extracts item M41, "record the number of steps taken each day," which is described as an antecedent in document D4, as the first item. This allows the analysis unit 102A to associate the item M41 with the item M22, and the answer generation unit 106A to generate an answer based on this association (for example, an answer recommending "record the number of steps you take each day").

[0102] Also, for example, suppose a question such as "What effect can I expect if I decide to go to a fitness club every week to manage my weight?" is received. When a question asking about a "consequent" such as an effect or result is received, the extraction unit 101A attempts to extract a document in which "weight management" is described as an antecedent, thereby extracting document D5. Then, the extraction unit 101A extracts item M51, "amount of investment for weight management," which is described as an antecedent in document D5, and sets this as the first item. Next, the extraction unit 101A attempts to extract a document in which item M52, which is described as a consequent corresponding to item M51 in document D5, is described as an antecedent, thereby detecting document D6. Then, the extraction unit 101A extracts item M62, "rate of reduction in medical expenses," which is described as a consequent in document D6, as the second item. This allows the analysis unit 102A to associate the item M41 with the item M22, and the answer generation unit 106A to generate an answer based on this association (for example, an answer that "reduction in medical expenses" can be expected).

[0103] (Extraction and Association Example 2: Starting with Extracting the Consequence) In the example of Fig. 4, after selecting one document from documents D1 to D4 recorded in the DB, the matter described as an antecedent in the selected document is extracted, but it is also possible to extract the matter described as a consequent in the selected document. This will be explained with reference to Fig. 8. Fig. 8 is a diagram showing another example of extraction and association of document description matters. Note that the description matters of documents D1 to D4 are the same as those in the example of Fig. 4, and therefore the description of the description matters will not be repeated.

[0104] In the example of FIG. 8, the extraction unit 101A reads out documents D1 to D4 recorded in the DB one by one, and extracts from the read out documents items described as consequents in the documents. For example, as shown in FIG. 8, the extraction unit 101A may input a standard prompt P21, such as "Please extract the items described as consequents in this document," and document D4 into the language model 111A. This allows the extraction of item M42 from document D4 as shown. Note that the extraction unit 101A may extract multiple items described as consequents from one document.

[0105] Next, the extraction unit 101A extracts documents in which the extracted matter is described as an antecedent. For example, as shown in FIG. 8, the extraction unit 101A may input a standard prompt P22, such as "Please extract a document in which this matter is described as an antecedent," and a matter M42 to the language model 111A. The extraction unit 101A may also specify documents D1 to D4 stored in the DB as extraction candidates. As a result, in the example shown in FIG. 8, document D1 is extracted. If multiple matters described as antecedents are extracted from document D4, the extraction unit 101A attempts to extract documents in which the extracted matters are described as antecedents for each extracted matter. If the extraction unit 101A fails to extract a corresponding document, it reads other documents from the DB and extracts matters described as consequents from the other documents.

[0106] Next, the extraction unit 101A inputs the document D1 extracted as described above and a prompt P23 instructing the language model 111A to extract the matter described as the consequent in the document D1, thereby extracting the matter described as the consequent in the document D1. As a result, the matter M12 can be extracted from the document D1 as shown in the figure.

[0107] Furthermore, the extraction unit 101A inputs document D4 and a prompt P24, which instructs the language model 111A to extract the matter described as an antecedent in document D4, to extract the matter described as an antecedent in document D4. As a result, it is possible to extract matter M41 from document D4, as shown in the figure.

[0108] The analysis unit 102A associates the item M41 extracted from document D4 in the above manner with the item M12 extracted from document D1. As a result, the knowledge that "recording the number of steps taken each day" "prolongs healthy life expectancy" is derived, as in the example of FIG.

[0109] As described above, extraction unit 101A uses language model 111A to (1) extract an item (item M42 in the example of FIG. 8) described as a consequent in a first content (document D4 in the example of FIG. 8), which is one of the multiple contents to be analyzed, (2) extract a second content (document D1 in the example of FIG. 8) from the multiple contents in which the extracted item is described as an antecedent, (3) extract the item described as a consequent in the second content as a second item (item M12 in the example of FIG. 8), and (4) extract the item described as an antecedent in the first content (item M11 in the example of FIG. 8) as a first item. This provides the effect of making it possible to logically associate the items described in each of the multiple contents and derive new knowledge.

[0110] When starting from the extraction of the consequent as in Figure 8, the processing flow is the same as in Figure 6. However, when starting from the extraction of the consequent, in S12, the matter described as the "consequent" in the selected document is extracted, and in S13, documents in which the matter extracted in S12 is described as the "antecedent" are extracted. Then, in S14, the matter described as the "consequent" in the document extracted in S13 is extracted as the "second matter," and in S15, the matter described as the "antecedent" in the document selected in S11 is extracted as the "first matter."

[0111] (Extraction and Association Example 3: Extracting antecedents and consequents from each document) Alternatively, the extraction unit 101A may use the language model 111A to extract, from each of the plurality of pieces of content to be analyzed, an item described as an antecedent in the content and an item described as a consequent corresponding to the item described as the antecedent. In this case, the analysis unit 102A may use, as an intermediate item, an item that is described as an antecedent in one piece of content and as a consequent in another piece of content, to identify a first item and a second item from the extracted items, and associate the identified first item and second item. This process also has the effect of logically associating the items described in each piece of content and deriving new knowledge.

[0112] The above process will be described with reference to FIG. 9. FIG. 9 is a diagram illustrating yet another example of extraction and association of document descriptions. In the example of FIG. 9, the extraction unit 101A reads documents D1 to D4 recorded in the DB one by one, and extracts from the read documents the items described as antecedents and the items described as consequents. For example, as shown in FIG. 9, the extraction unit 101A may input a standard prompt P31, such as "Please extract the items described as antecedents in this document and the corresponding items described as consequents," and the read documents into the language model 111A. This allows the items described as antecedents and the items described as consequents in each of the documents D1 to D4 to be extracted, as illustrated.

[0113] Then, the analysis unit 102A identifies a first item and a second item from the items extracted as described above, and associates the identified first item with the second item. As described above, the first item is the item described as the antecedent in the content in which the intermediary item is described as the consequent, and the second item is the item described as the consequent in the content in which the intermediary item is described as the antecedent. For this reason, the analysis unit 102A may identify an intermediary item from the items extracted by the extraction unit 101A, and associate the item described as the antecedent in the content in which the identified intermediary item is described as the consequent as the first item and the item described as the consequent in the content in which the identified intermediary item is described as the antecedent as the second item.

[0114] The method for identifying intermediary matters is not particularly limited, and may be performed by the language model 111A, for example. In this case, the analysis unit 102A generates a prompt to instruct the language model 111A to extract matters having similar contents from the matters extracted by the extraction unit 101A, and inputs the generated prompt to the language model 111A. Then, the analysis unit 102A can identify, among the extracted matters, a matter that is described as an antecedent in one content and as a consequent in another content, as an intermediary matter.

[0115] In the example of Figure 9, item M42 and item M11 have the same content, so they are identified as intermediary items. Then, through this intermediary item, item M41 extracted from document D4 and item M12 extracted from document D1 are associated. As a result, as in the example of Figure 4, the knowledge that "recording the number of steps taken each day" "extends healthy life expectancy" is derived.

[0116] (Processing flow: Analysis method corresponding to Figure 9) As explained with reference to Fig. 9, from each of a plurality of pieces of content to be analyzed, an item described as an antecedent in the content and an item described as a consequent corresponding to the item may be extracted, and then the items may be associated with each other. The flow of processing executed by the information processing device 1A in this case will be explained with reference to Fig. 10. Fig. 10 is a flow diagram showing another example of processing executed by the information processing device 1A. Like the flow of Fig. 6, the flow of Fig. 10 also includes each step of the analysis method according to this exemplary embodiment.

[0117] Note that the processes of S31 and S33 in Fig. 10 are similar to the processes of S11 and S17 in Fig. 6, respectively, and therefore will not be described repeatedly here. Also, although the flow in Fig. 10 does not include the processes of generating and presenting logic model 112A, the flow in Fig. 10 may also include the processes of generating and presenting logic model 112A.

[0118] In S32 (extraction process), the extraction unit 101A uses the language model 111A to extract, from the document acquired in S31, an item described as an antecedent in the document and an item described as a consequent corresponding to the item described in the document. Note that the extraction unit 101A may extract multiple pairs of an item described as an antecedent and an item described as a consequent from one document.

[0119] In S34, the analysis unit 102A identifies an intermediary matter from among the descriptions of the multiple documents to be analyzed that have been extracted by repeating the processes of S31 to S33. Note that the analysis unit 102A may identify multiple intermediary matters. If no intermediary matter is present among the extracted descriptions, the process of FIG. 10 ends.

[0120] In S35 (analysis process), the analysis unit 102A associates the items described in the document. Specifically, the analysis unit 102A associates a first item described as an antecedent in the document in which the intermediary item identified in S34 is described as a consequent, with a second item described as a consequent in the document in which the intermediary item is described as an antecedent. This completes the process in FIG. 10. Note that if multiple intermediary items are identified in S34, the association is performed for each intermediary item in S35.

[0121] (Example 4 of extraction and association: Example 1 of extracting documents that describe intermediary matters) Furthermore, using the language model 111A, the extraction unit 101A may (1) extract, from among the plurality of pieces of content, content in which an item described as an antecedent in another piece of content is described as a consequent, i.e., a second piece of content in which an intermediate item is described, (2) extract, as a first item, the item described as an antecedent in the extracted content, and (3) extract, as a second item, content in which an item described as a consequent in the extracted content is described as an antecedent, i.e., the item described as a consequent in the first piece of content in which an intermediate item is described. This type of processing also has the effect of making it possible to logically associate the items described in each of the plurality of pieces of content and derive new knowledge.

[0122] The above process will be described with reference to FIG. 11. FIG. 11 illustrates another example of document description extraction and association. In the example of FIG. 11, the extraction unit 101A reads documents D1 to D4 stored in the DB one by one and extracts documents in which an antecedent described in the read document is described as a consequent. For example, as shown in FIG. 11, the extraction unit 101A may input a standard prompt P41, such as "Please extract documents in which an antecedent described in this document is described as a consequent from the documents stored in the DB," along with the read document to the language model 111A. As a result, as illustrated, if the read document is document D1, document D4 can be extracted in which an antecedent item M42, having the same content as an antecedent item M11 in document D1, is described as a consequent. Since items M11 and M42 are intermediate items, document D4 in this example corresponds to the second content described above, and document D1 corresponds to the first content described above.

[0123] Next, the extraction unit 101A extracts the matter described as the antecedent in the extracted document D4, i.e., the second content, as the first matter, and extracts the matter described as the consequent in the document D1, i.e., the first content, read from the DB, as the second matter.

[0124] More specifically, as shown in the figure, the extraction unit 101A inputs the extracted document D4 and a prompt P42, which instructs the language model 111A to extract the matter described as an antecedent in document D4, to extract the matter described as an antecedent in document D4. As a result, as shown in the figure, it is possible to extract matter M41 described in document D4 as the first matter.

[0125] In addition, as shown in the figure, the extraction unit 101A inputs the read document D1 and a prompt P43, which instructs the language model 111A to extract the matter described as the consequent in document D1, thereby extracting the matter described as the consequent in document D1. As a result, as shown in the figure, it is possible to extract the matter M12 described in document D1 as the second matter.

[0126] The analysis unit 102A associates the item M41 extracted from document D4 in the above manner with the item M12 extracted from document D1. As a result, the knowledge that "recording the number of steps taken each day" "prolongs healthy life expectancy" is derived, as in the example of FIG. 4 etc.

[0127] (Processing flow: Analysis method corresponding to Figure 11) As explained with reference to Fig. 11, it is also possible to extract documents in which intermediary matters are described using language model 111A, extract first and second matters, and associate the respective matters. The flow of processing executed by information processing device 1A in this case will be explained with reference to Fig. 12. Fig. 12 is a flow diagram showing another example of processing executed by information processing device 1A. Like the flow of Fig. 6, the flow of Fig. 12 also includes each step of the analysis method according to this exemplary embodiment.

[0128] 12 and the flow of Fig. 6 are similar except that the processes of S12 and S13 in the latter are replaced by the process of S42, so the following description will focus on S42. Also, although the process of generating and presenting the logic model 112A is not included in the flow of Fig. 12, the process of generating and presenting the logic model 112A may also be included in the flow of Fig. 12.

[0129] In S42 (extraction process), the extraction unit 101A uses the language model 111A to extract, from the plurality of documents to be analyzed, a document in which the matter described as the antecedent in the document selected in S41 is described as the consequent, i.e., a second content. Note that the extraction unit 101A may extract a plurality of documents in S42.

[0130] In S43 (extraction process), extraction unit 101A extracts, as a first matter, the matter described as an antecedent in the document extracted in S42, i.e., the second content. In S44 (extraction process), extraction unit 101A extracts, as a second matter, the matter described as a consequent in the document selected in S41, i.e., the first content. Then, in S45 (analysis process), analysis unit 102A associates the first matter extracted in S43 with the second matter extracted in S44.

[0131] (Example 5 of extraction and association: Example 2 of extracting documents that describe intermediary matters) Furthermore, using the language model 111A, the extraction unit 101A may (1) extract, from among the plurality of pieces of content, a piece of content in which an item described as a consequent in another piece of content is described as an antecedent, i.e., a first piece of content in which an intermediate item is described, (2) extract, as a second item, the item described as a consequent in the extracted content, and (3) extract, as a first item, the item described as an antecedent in the content in which an item described as an antecedent in the extracted content is described as a consequent, i.e., the item described as an antecedent in the second piece of content in which an intermediate item is described. This type of processing also has the effect of making it possible to logically associate the items described in each of the plurality of pieces of content and derive new knowledge.

[0132] The above process will be described with reference to FIG. 13. FIG. 13 is a diagram illustrating yet another example of extraction and association of document descriptions. In the example of FIG. 13, the extraction unit 101A reads documents D1 to D4 stored in the DB one by one and extracts documents in which the item described as the consequent in the read document is described as the antecedent. For example, as shown in FIG. 13, the extraction unit 101A may input a standard prompt P51, such as "Please extract documents in which the item described as the consequent in this document is described as the antecedent from the documents stored in the DB," along with the read document to the language model 111A. As a result, as shown in the figure, if the read document is document D4, document D1 can be extracted in which item M11, which has the same content as item M42 described as the consequent in document D4, is described as the antecedent. Since items M42 and M11 are intermediate items, document D1 in this example corresponds to the first content described above, and document D4 corresponds to the second content described above.

[0133] Next, the extraction unit 101A extracts the matter described as the consequent in the extracted document D1, i.e., the first content, as the second matter, and also extracts the matter described as the antecedent in the document D4, i.e., the second content, read from the DB, as the first matter.

[0134] More specifically, as shown in the figure, the extraction unit 101A inputs the extracted document D1 and a prompt P52, which instructs the language model 111A to extract the matter described as the consequent in document D1, thereby extracting the matter described as the consequent in document D1. As a result, as shown in the figure, it is possible to extract the matter M12 described in document D1 as the second matter.

[0135] In addition, as shown in the figure, the extraction unit 101A inputs the read document D4 and a prompt P53 instructing to extract the matter described as an antecedent in document D4 to the language model 111A, thereby extracting the matter described as an antecedent in document D4. As a result, as shown in the figure, it is possible to extract matter M41 described in document D4 as the first matter.

[0136] The analysis unit 102A associates the item M41 extracted from document D4 with the item M12 extracted from document D1 in the above manner. As a result, the knowledge that "recording the number of steps taken each day" "prolongs healthy life expectancy" is derived, as in the example of FIG. 4 etc.

[0137] The flow of the processing (analysis method) in the example of Fig. 13 is the same as the flow of processing in the example of Fig. 11 (see Fig. 12). However, in the processing in the example of Fig. 13, in S42, a document in which an item described as a "consequent" in a selected document is described as an "antecedent" is extracted, in S43, the item described as a "consequent" in the document extracted in S42 is extracted as a "second item," and in S44, the item described as an "antecedent" in the document selected in S41 is extracted as a "first item."

[0138] [Use as a document search device] The information processing device 1A can also be used as a document search device. In this case, a user inputs information indicating a search target into the information processing device 1A. The information indicating the search target may be, for example, a document including a description related to the document the user wants to search for, or a sentence (text data) indicating the search target.

[0139] For example, suppose a user inputs a document as information indicating a search target. In this case, the extraction unit 101A considers the items described as antecedents or consequents in the input document as intermediate items and extracts documents that describe the intermediate items from multiple documents to be searched. Then, the presentation unit 107A presents the documents extracted by the extraction unit 101A to the user.

[0140] The method shown in Fig. 4, Fig. 8, Fig. 9, Fig. 11, or Fig. 13 can be applied to extract documents. For example, when the method shown in Fig. 4 is applied, extraction unit 101A extracts matters described as antecedents in the input documents, and extracts documents in which the extracted matters are described as consequents. When the method shown in Fig. 8 is applied, extraction unit 101A extracts matters described as consequents in the input documents, and extracts documents in which the extracted matters are described as antecedents.

[0141] 9, the extraction unit 101A extracts an item described as an antecedent and an item described as a consequent in the input document. The extraction unit 101A also extracts an item described as an antecedent and an item described as a consequent from multiple documents to be searched. The extraction unit 101A then extracts, from the multiple documents to be searched, at least one of a document in which an item described as an antecedent in the input document is described as a consequent, and a document in which an item described as a consequent in the input document is described as an antecedent.

[0142] 11, extraction unit 101A extracts documents in which an item described as an antecedent in the input document is described as a consequent from among the documents to be searched. Similarly, when the method described in FIG. 13 is applied, extraction unit 101A extracts documents in which an item described as a consequent in the input document is described as an antecedent from among the documents to be searched.

[0143] Also, suppose that the user inputs text indicating a search target. In this case, the extraction unit 101A extracts documents in which the matter indicated in the input text is described as an antecedent or consequent from among multiple documents to be searched. Then, the extraction unit 101A regards the matter described as an antecedent or consequent corresponding to the matter indicated in the text in the extracted documents as an intermediate matter, and extracts documents in which the intermediate matter is described.

[0144] For example, suppose the above text represents a user's question, "What effects can be expected from taking supplement s?" When a question asking about a "consequent" such as an effect or result is received, the extraction unit 101A extracts documents in which the item "taking supplement s" is described as an antecedent. Then, the extraction unit 101A uses an item (e.g., "sleep quality improves") described as a consequent corresponding to the item in the extracted documents as an intermediate item, and extracts documents in which the intermediate item is described as an antecedent. This extracts documents that state, for example, that improving sleep quality will improve work efficiency. Therefore, by presenting such documents to the user by the presentation unit 107A, it is possible to provide the user with the knowledge that taking supplement s is expected to improve work efficiency.

[0145] Furthermore, for example, when a question about an "antecedent" such as a measure or a condition is received, the extraction unit 101A may extract documents in which the matter indicated in the text is described as a consequent. Then, the extraction unit 101A may extract documents in which the matter described as an antecedent corresponding to the matter in the extracted documents is described as an intermediate matter.

[0146] As described above, when an item that is described as an antecedent in some content among a plurality of target contents and as a consequent in other content is set as an intermediate item, the content in which the intermediate item is described as a consequent is set as a first content, and the content in which the intermediate item is described as an antecedent is set as a second content, the information processing device 1A includes an extraction unit 101A that extracts at least one of the first content and the second content using a machine-learned language model 111A, and a presentation unit 107A that presents the content extracted by the extraction unit 101A. This provides the effect of enabling support for utilizing a variety of content.

[0147] Note that the "plurality of target contents" in the above configuration may include not only the content to be searched but also the content input by the user for the search. In this case, the content input by the user for the search becomes the first content or the second content. If the content input by the user for the search is the first content, the extraction unit 101A extracts the second content from the content to be searched. On the other hand, if the content input by the user for the search is the second content, the extraction unit 101A extracts the first content from the content to be searched.

[0148] [Modification] The execution entity of each process described in the above exemplary embodiment is arbitrary and is not limited to the above example. For example, a system having the same functions as the information processing devices 1 and 1A can be constructed using multiple devices that can communicate with each other. Furthermore, the execution entity of each process shown in the flowcharts of Figures 6, 7, 10, and 12 may be one device (which can also be called a processor) or multiple devices (which can also be called processors).

[0149] [Software implementation example] Some or all of the functions of the information processing device 1, 1A may be realized by hardware such as an integrated circuit (IC chip), or may be realized by software.

[0150] In the latter case, the information processing device 1 or 1A is realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in Fig. 14. Fig. 14 is a block diagram showing the hardware configuration of the computer C that functions as the information processing device 1 or 1A.

[0151] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program (analysis program) P for causing the computer C to operate as the information processing device 1 or 1A. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing each function of the information processing device 1 or 1A.

[0152] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0153] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, mouse, display, and printer.

[0154] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0155] Furthermore, each of the above functions of the information processing device 1 or 1A may be realized by a single processor provided in a single computer, by multiple processors provided in a single computer working together, or by multiple processors provided in each of multiple computers working together. Furthermore, a program for causing the information processing device 1 or 1A to realize each of the above functions may be stored in a single memory provided in a single computer, or may be distributed and stored in multiple memories provided in a single computer, or may be distributed and stored in multiple memories provided in each of multiple computers.

[0156] [Appendix 1] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0157] [Appendix A] (Appendix A1) An information processing device comprising: an extraction unit that uses a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the content and an item described as a consequent in the content; and an analysis unit that, when an item that is described as an antecedent in some of the plurality of contents and as a consequent in other of the plurality of contents is considered to be an intermediate item, associates, based on the extraction result of the extraction unit, a first item described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item described as a consequent in the content in which the intermediate item is described as an antecedent.

[0158] (Appendix A2) The information processing device described in Appendix A1, wherein the extraction unit uses the language model to extract an item described as an antecedent in a first content that is one of the plurality of contents, extract a second content from the plurality of contents in which the extracted item is described as a consequent, extract the item described as an antecedent in the second content as the first item, and extract the item described as a consequent in the first content as the second item.

[0159] (Appendix A3) The information processing device described in Appendix A1, wherein the extraction unit uses the language model to extract an item described as a consequent in a first content that is one of the plurality of contents, extracts a second content from the plurality of contents in which the extracted item is described as an antecedent, extracts the item described as a consequent in the second content as the second item, and extracts the item described as an antecedent in the first content as the first item.

[0160] (Appendix A4) The information processing device described in Appendix A1, wherein the extraction unit uses the language model to extract from each of the plurality of contents an item described as an antecedent in the content and an item described as a consequent corresponding to the item described in the content, and the analysis unit uses an item among the extracted items that is described as an antecedent in some content and as a consequent in other content as an intermediate item, identifies the first item and the second item from the extracted items, and associates the identified first item and the second item.

[0161] (Appendix A5) The information processing device described in Appendix A1, wherein the extraction unit uses the language model to extract, from the plurality of contents, content in which an item described as an antecedent in another content is described as a consequent, extracts the item described as an antecedent in the extracted content as the first item, and extracts the item described as a consequent in the content in which an item described as a consequent in the extracted content is described as an antecedent as the second item.

[0162] (Appendix A6) The information processing device described in Appendix A1, wherein the extraction unit uses the language model to extract content from the plurality of contents in which an item described as a consequent in another content is described as an antecedent, extracts the item described as a consequent in the extracted content as the second item, and extracts the item described as an antecedent in the content in which an item described as an antecedent in the extracted content is described as a consequent as the first item.

[0163] (Appendix A7) An information processing device according to any one of appendices A1 to A6, comprising an inference unit that infers the relationship between the first item and the second item using first relationship information that indicates the relationship between the first item and the intermediary item, extracted from content in which the first item is described as an antecedent and the intermediary item is described as a consequent, and second relationship information that indicates the relationship between the intermediary item and the second item, extracted from content in which the intermediary item is described as an antecedent and the second item is described as a consequent.

[0164] (Appendix A8) An information processing device according to any one of appendices A1 to A7, comprising a model generation unit that generates a logic model showing the logical relationship between each item described in the plurality of contents based on the association results by the analysis unit.

[0165] (Appendix A9) The information processing device according to Appendix A8, comprising: a receiving unit that receives an input of a question; an answer generating unit that generates an answer to the question using the logic model; and a presenting unit that presents the answer.

[0166] [Appendix B] (Appendix B1) An analysis method including: an extraction process in which at least one processor uses a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the contents and an item described as a consequent in the contents; and an analysis process in which, when an item described as an antecedent in some of the plurality of contents and described as a consequent in other contents is considered an intermediate item, associates, based on the extraction result of the extraction process, a first item described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item described as a consequent in the content in which the intermediate item is described as an antecedent.

[0167] (Appendix B2) The analysis method described in Appendix B1, wherein in the extraction process, the at least one processor uses the language model to extract an item described as an antecedent in a first content that is one of the plurality of contents, extract a second content from the plurality of contents in which the extracted item is described as a consequent, extract the item described as an antecedent in the second content as the first item, and extract the item described as a consequent in the first content as the second item.

[0168] (Appendix B3) The analysis method described in Appendix B1, wherein in the extraction process, the at least one processor uses the language model to extract an item described as a consequent in a first content that is one of the plurality of contents, extract a second content from the plurality of contents in which the extracted item is described as an antecedent, extract the item described as a consequent in the second content as the second item, and extract the item described as an antecedent in the first content as the first item.

[0169] (Appendix B4) The analysis method described in Appendix B1, wherein in the extraction process, the at least one processor uses the language model to extract from each of the plurality of contents an item described as an antecedent in the content and an item described as a consequent corresponding to the item described in the content; and in the analysis process, the at least one processor identifies the first item and the second item from the extracted items, using an item among the extracted items that is described as an antecedent in some content and as a consequent in other content as an intermediate item, and associates the identified first item with the second item.

[0170] (Appendix B5) The analysis method described in Appendix B1, wherein in the extraction process, the at least one processor uses the language model to extract, from the plurality of contents, content in which an item described as an antecedent in another content is described as a consequent, extracts the item described as an antecedent in the extracted content as the first item, and extracts the item described as a consequent in the content in which the item described as a consequent in the extracted content is described as an antecedent as the second item.

[0171] (Appendix B6) The analysis method described in Appendix B1, wherein in the extraction process, the at least one processor uses the language model to extract, from the plurality of contents, content in which an item described as a consequent in other content is described as an antecedent, extracts the item described as a consequent in the extracted content as the second item, and extracts the item described as an antecedent in the content in which an item described as an antecedent in the extracted content is described as a consequent as the first item.

[0172] (Appendix B7) The analysis method of any one of Appendices B1 to B6, including an inference process in which the at least one processor infers the relationship between the first item and the second item using first relationship information indicating the relationship between the first item and the intermediary item, extracted from content in which the first item is described as an antecedent and the intermediary item is described as a consequent, and second relationship information indicating the relationship between the intermediary item and the second item, extracted from content in which the intermediary item is described as an antecedent and the second item is described as a consequent.

[0173] (Appendix B8) An analysis method described in any of Appendices B1 to B7, including a model generation process in which the at least one processor generates a logic model showing the logical relationship between each item described in the multiple contents based on the association results of the analysis process.

[0174] (Appendix B9) An analysis method according to Appendix B8, comprising: a reception process in which the at least one processor receives input of a question; an answer generation process in which the at least one processor generates an answer to the question using the logic model; and a presentation process in which the at least one processor presents the answer.

[0175] [Appendix C] (Appendix C1) An analysis program that causes a computer to function as: an extraction means that uses a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the content and an item described as a consequent in the content; and an analysis means that, when an item that is described as an antecedent in one of the plurality of contents and as a consequent in another of the plurality of contents is an intermediate item, associates, based on the extraction result of the extraction means, a first item that is described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item that is described as a consequent in the content in which the intermediate item is described as an antecedent.

[0176] (Appendix C2) The analysis program described in Appendix C1, wherein the extraction means uses the language model to extract an item described as an antecedent in a first content that is one of the plurality of contents, extract a second content from the plurality of contents in which the extracted item is described as a consequent, extract the item described as an antecedent in the second content as the first item, and extract the item described as a consequent in the first content as the second item.

[0177] (Appendix C3) The analysis program described in Appendix C1, wherein the extraction means uses the language model to extract an item described as a consequent in a first content that is one of the plurality of contents, extracts a second content from the plurality of contents in which the extracted item is described as an antecedent, extracts the item described as a consequent in the second content as the second item, and extracts the item described as an antecedent in the first content as the first item.

[0178] (Appendix C4) The analysis program described in Appendix C1, wherein the extraction means uses the language model to extract from each of the plurality of contents an item described as an antecedent in the content and an item described as a consequent corresponding to the item described in the content, and the analysis means identifies the first item and the second item from the extracted items, using an item among the extracted items that is described as an antecedent in some content and as a consequent in other content as an intermediate item, and associates the identified first item with the second item.

[0179] (Appendix C5) The analysis program described in Appendix C1, wherein the extraction means uses the language model to extract, from the plurality of contents, content in which an item described as an antecedent in another content is described as a consequent, extracts the item described as an antecedent in the extracted content as the first item, and extracts the item described as a consequent in the content in which an item described as a consequent in the extracted content is described as an antecedent as the second item.

[0180] (Appendix C6) The analysis program described in Appendix C1, wherein the extraction means uses the language model to extract, from the plurality of contents, content in which an item described as a consequent in another content is described as an antecedent, extracts the item described as a consequent in the extracted content as the second item, and extracts the item described as an antecedent in the content in which an item described as an antecedent in the extracted content is described as a consequent as the first item.

[0181] (Appendix C7) The analysis program of any one of appendices C1 to C6, which causes the computer to function as an inference means for inferring the relationship between the first item and the second item using first relationship information indicating the relationship between the first item and the intermediary item, extracted from content in which the first item is described as an antecedent and the intermediary item is described as a consequent, and second relationship information indicating the relationship between the intermediary item and the second item, extracted from content in which the intermediary item is described as an antecedent and the second item is described as a consequent.

[0182] (Appendix C8) An analysis program according to any one of appendices C1 to C7, which causes the computer to function as a model generation means for generating a logic model showing the logical relationship between each item described in the plurality of contents based on the association results by the analysis means.

[0183] (Appendix C9) The analysis program according to appendix C8, which causes the computer to function as a receiving means for receiving input of a question, an answer generating means for generating an answer to the question using the logic model, and a presentation means for presenting the answer.

[0184] [Appendix D] (Appendix D1) An information processing device comprising at least one processor, the at least one processor executing an extraction process using a machine-learned language model to extract from a plurality of target contents at least one of an item described as an antecedent in the contents and an item described as a consequent in the contents; and an analysis process, when an item described as an antecedent in some of the plurality of contents and described as a consequent in other contents is considered an intermediate item, to associate, based on the extraction result of the extraction process, a first item described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item described as a consequent in the content in which the intermediate item is described as an antecedent.

[0185] The information processing device may further include a memory, and the memory may store a program for causing the at least one processor to execute each of the processes.

[0186] (Appendix D2) The information processing device described in Appendix D1, wherein in the extraction process, the at least one processor uses the language model to extract an item described as an antecedent in a first content that is one of the plurality of contents, extract a second content from the plurality of contents in which the extracted item is described as a consequent, extract the item described as an antecedent in the second content as the first item, and extract the item described as a consequent in the first content as the second item.

[0187] (Appendix D3) The information processing device described in Appendix D1, wherein in the extraction process, the at least one processor uses the language model to extract an item described as a consequent in a first content that is one of the plurality of contents, extracts a second content from the plurality of contents in which the extracted item is described as an antecedent, extracts the item described as a consequent in the second content as the second item, and extracts the item described as an antecedent in the first content as the first item.

[0188] (Appendix D4) The information processing device described in Appendix D1, wherein in the extraction process, the at least one processor uses the language model to extract from each of the plurality of contents an item described as an antecedent in the content and an item described as a consequent corresponding to the item, and in the analysis process, the at least one processor identifies the first item and the second item from the extracted items, using an item among the extracted items that is described as an antecedent in some content and as a consequent in other content as an intermediate item, and associates the identified first item and the second item.

[0189] (Appendix D5) The information processing device described in Appendix D1, wherein in the extraction process, the at least one processor uses the language model to extract content from the plurality of contents in which an item described as an antecedent in other content is described as a consequent, extracts the item described as an antecedent in the extracted content as the first item, and extracts the item described as a consequent in the content in which an item described as a consequent in the extracted content is described as an antecedent as the second item.

[0190] (Appendix D6) The information processing device described in Appendix D1, wherein in the extraction process, the at least one processor uses the language model to extract content from the plurality of contents in which an item described as a consequent in other content is described as an antecedent, extracts the item described as a consequent in the extracted content as the second item, and extracts the item described as an antecedent in the content in which an item described as an antecedent in the extracted content is described as a consequent as the first item.

[0191] (Appendix D7) An information processing device described in any of Appendices D1 to D6, wherein the at least one processor performs an inference process to infer the relationship between the first item and the second item using first relationship information indicating the relationship between the first item and the intermediary item, extracted from content in which the first item is described as an antecedent and the intermediary item is described as a consequent, and second relationship information indicating the relationship between the intermediary item and the second item, extracted from content in which the intermediary item is described as an antecedent and the second item is described as a consequent.

[0192] (Appendix D8) An information processing device described in any of Appendices D1 to D7, wherein the at least one processor executes a model generation process to generate a logic model showing the logical relationship between each item described in the multiple contents based on the association results from the analysis process.

[0193] (Appendix D9) The information processing device described in Appendix D8, wherein the at least one processor executes a reception process for receiving input of a question, an answer generation process for generating an answer to the question using the logic model, and a presentation process for presenting the answer.

[0194] [Appendix E] A non-transient recording medium having recorded thereon an analysis program that causes a computer to execute the following steps: an extraction process that uses a machine-learned language model to extract from multiple target contents at least one of an item described as an antecedent in the content and an item described as a consequent in the content; and an analysis process that, when an item that is described as an antecedent in some of the multiple contents and as a consequent in other of the multiple contents is an intermediate item, associates, based on the extraction results of the extraction process, a first item described as an antecedent in the content in which the intermediate item is described as a consequent with a second item described as a consequent in the content in which the intermediate item is described as an antecedent. [Explanation of symbols]

[0195] 1. Information processing equipment 101 Extraction part (extraction means) 102 Analysis Department (Analysis Means) 1A Information processing equipment 101A Extraction part (extraction means) 102A Analysis Department (Analysis Means) 103A model generation unit (model generation means) 104A Reasoning part (reasoning means) 105A Reception section (reception means) 106A Answer generation unit (answer generation means) 107A Presentation unit (presentation means) 111A Language Model 112A Logic Model

Claims

1. an extraction means for extracting, from a plurality of target contents, at least one of an item described as an antecedent and an item described as a consequent in the content, using a machine-learned language model; An information processing device comprising: an analysis means for associating, based on the extraction result of the extraction means, a first item described as an antecedent in the content in which the intermediary item is described as a consequent, with a second item described as a consequent in the content in which the intermediary item is described as an antecedent, when the intermediary item is described as an antecedent in some of the plurality of contents and as a consequent in other of the contents, as an intermediary item.

2. The extraction means uses the language model to extracting a matter described as an antecedent in a first content that is one of the plurality of contents; extracting a second content from the plurality of contents in which the extracted item is described as a consequent; extracting, as the first item, an item described as an antecedent in the second content; The information processing apparatus according to claim 1 , wherein an item described as a consequent in the first content is extracted as the second item.

3. The extraction means uses the language model to extracting a matter described as a consequent in a first content that is one of the plurality of contents; extracting a second content from the plurality of contents in which the extracted matter is described as an antecedent; extracting, as the second item, an item described as a consequent in the second content; The information processing apparatus according to claim 1 , wherein an item described as an antecedent in the first content is extracted as the first item.

4. the extraction means extracts, from each of the plurality of contents, an item described as an antecedent in the content and an item described as a consequent corresponding to the item described in the antecedent, and 2. The information processing device according to claim 1, wherein the analysis means identifies the first item and the second item from among the extracted items, using an item among the extracted items that is described as an antecedent in some content and as a consequent in other content as the intermediate item, and associates the identified first item with the identified second item.

5. The extraction means uses the language model to extracting, from the plurality of contents, a content in which an item described as an antecedent in another content is described as a consequent; extracting, as the first item, an item described as an antecedent in the extracted content; The information processing apparatus according to claim 1 , wherein an item written as a consequent in the extracted content is written as an antecedent, and an item written as a consequent in the content is extracted as the second item.

6. The extraction means uses the language model to extracting, from the plurality of contents, a content in which a matter described as a consequent in another content is described as an antecedent; extracting, as the second item, an item described as a consequent in the extracted content; The information processing apparatus according to claim 1 , wherein an item described as an antecedent in the extracted content is described as a consequent, and the item described as an antecedent in the extracted content is extracted as the first item.

7. 7. The information processing device according to claim 1, further comprising an inference means for inferring the relationship between the first item and the second item using first relationship information indicating the relationship between the first item and the intermediary item, extracted from content in which the first item is described as an antecedent and the intermediary item is described as a consequent, and second relationship information indicating the relationship between the intermediary item and the second item, extracted from content in which the intermediary item is described as an antecedent and the second item is described as a consequent.

8. 7. The information processing device according to claim 1, further comprising: a model generation unit that generates a logic model indicating a logical relationship between each item described in the plurality of contents based on the association result by the analysis unit.

9. A receiving means for receiving an input of a question; an answer generation means for generating an answer to the question using the logic model; The information processing device according to claim 8 , further comprising: a presentation unit that presents the answer.

10. At least one processor an extraction process for extracting, from a plurality of target contents, at least one of an item described as an antecedent and an item described as a consequent in the content, using a machine-learned language model; an analysis method including: when an item that is described as an antecedent in some of the plurality of contents and as a consequent in other contents is considered to be an intermediate item, an analysis process that associates, based on the extraction results of the extraction process, a first item that is described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item that is described as a consequent in the content in which the intermediate item is described as an antecedent.

11. Computer, An extraction means for extracting, from a plurality of target contents, at least one of an item described as an antecedent and an item described as a consequent in the content, using a machine-learned language model; and When an item that is described as an antecedent in one of the plurality of contents and as a consequent in another of the contents is regarded as an intermediate item, the analysis program functions as an analysis means that associates, based on the extraction result of the extraction means, a first item that is described as an antecedent in the content in which the intermediate item is described as a consequent, with a second item that is described as a consequent in the content in which the intermediate item is described as an antecedent.

Images