Information processing apparatus, information processing method, and information processing program

Abstract

To provide an information processing apparatus, an information processing method, and an information processing program capable of effectively making a summary of a news article.SOLUTION: An information processing apparatus includes: a selection / rearrangement unit which causes a generative AI to select and rearrange a plurality of news articles, using first instruction information including information on selecting a plurality of news articles and information on constraint conditions for the selection, and second instruction information including information on rearranging the news articles and information on constraint conditions for the rearrangement; and a summary unit which causes the generative AI to make summaries of the news articles, using third instruction information including information for making summaries of the news articles selected and rearranged by the selection / rearrangement unit.SELECTED DRAWING: Figure 3

Description

[Technical Field]

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

[0002] In recent years, technologies using AI have been proposed. For example, Patent Document 1 proposes generating summaries of news articles from news articles using a trained neural network model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-087353 Summary of the Invention [Problem to be solved by the invention]

[0004] However, when there are many news articles, for example, generating summaries for all news articles increases the processing cost for the AI, and there is also a risk that the AI ​​will confuse information if it summarizes all news articles at once. Therefore, there are challenges in providing effective summaries of news articles.

[0005] The present application has been made in consideration of the above, and aims to provide an information processing device, an information processing method, and an information processing program that can improve the convenience of users by enabling them to effectively summarize news articles. [Means for solving the problem]

[0006] The information processing device of the present application includes a selection and rearrangement unit that causes a generation AI to select and rearrange multiple news articles using first instruction information that includes information instructing the selection of multiple news articles and information indicating selection constraints, and second instruction information that includes information instructing the rearrangement of the multiple news articles and information indicating rearrangement constraints, and a summarization unit that causes the generation AI to summarize each of the multiple news articles using third instruction information that includes information instructing the summary of each of the multiple news articles selected and rearranged by the selection and rearrangement unit. [Effects of the Invention]

[0007] According to one aspect of the embodiment, it is possible to effectively summarize news articles. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram illustrating an example of information processing according to an embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the configuration of an information processing system according to the embodiment. [Figure 3] FIG. 3 is a diagram illustrating an example of the configuration of the information processing device according to the embodiment. [Figure 4] FIG. 4 is a diagram showing an example of a news article group information table stored in the news article group information storage unit of the information processing device according to the embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of a summary information table stored in the summary information storage unit of the information processing device according to the embodiment. [Figure 6] FIG. 6 is a diagram illustrating an example of the configuration of a deletion target detection unit in a preprocessing unit in a processing unit of an information processing device according to an embodiment. [Figure 7] FIG. 7 is a diagram showing an example of an application screen that is distributed by a distribution unit in the processing unit of the information processing device according to the embodiment and displayed on the terminal device. [Figure 8] FIG. 8 is a flowchart showing an example of information processing by the processing unit of the information processing device according to the embodiment. [Figure 9] FIG. 9 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the information processing device according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, modes for implementing an information processing device, an information processing method, and an information processing program according to the present application (hereinafter referred to as "embodiments") will be described in detail with reference to the drawings. Note that the information processing device, the information processing method, and the information processing program according to the present application are not limited to these embodiments. Furthermore, the respective embodiments can be appropriately combined within the scope of not causing any contradiction in the processing content. Furthermore, the same components in the following embodiments will be assigned the same reference numerals, and redundant explanations will be omitted.

[0010] [1. An example of information processing] First, an example of information processing according to the embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of information processing according to the embodiment.

[0011] 1, the information processing device 1 accepts submission of multiple news articles from multiple submission user terminals 3 (step S1). The multiple submission user terminals 3 are terminals of different submission users. The submission users may be, for example, employees of a news organization, journalists, etc., but are not limited to such examples.

[0012] Furthermore, the news article submitted by the submitter includes, but is not limited to, information such as the headline, text, text snippet, image, category, submission date and time (or creation date and time), keywords, submitter, and links to related articles. A text snippet is a snippet of the text of a news article.

[0013] Next, the information processing device 1 performs preprocessing on the news article group including the multiple news articles submitted in step S1 using a generation AI (Artificial Intelligence) (step S2). The information on the news article group used in the preprocessing is represented by a news article list that lists combinations of headlines, body snippets, and categories of the news articles included in the news article group, but is not limited to this example. In the preprocessing using the generation AI, the news article group is represented in JSON (JavaScript (registered trademark) Object Notation) format, but is not limited to this example.

[0014] The generative AI is a text generation AI, which is, for example, a large-scale language model trained to estimate and output the next token from an input token sequence, such as a transformer-based model or an RNN (Recurrent Neural Network)-based model, or may be a mixed model of these.

[0015] Examples of the transformer-based model include, but are not limited to, GPT (Generative Pre-trained Transformer) (registered trademark), PaLM2 (Pathways Language Model Version 2), and LLaMA (Large Language Model Meta AI). Examples of the RNN-based model include, but are not limited to, RWKV (Receptance Weighted Key Value).

[0016] It is desirable that the generation AI is trained so as not to include personal information in the generated results. The generation AI is placed in an external information processing device, and the information processing device 1 uses the generation AI via an API (Application Programming Interface), but the generation AI may also be placed within the information processing device 1.

[0017] The pre-processing in step S2 includes, for example, a process for detecting duplicates, a process for detecting items to be deleted, and a weighting process. First, the deletion process will be described. The deletion process is a process for deleting a specific news article from a group of news articles using a generation AI.

[0018] The duplication detection process is described below. The duplication detection process is a process of detecting duplication of news articles in a group of news articles using a generation AI. The information processing device 1 inputs, as input information, a prompt to the generation AI, information including instruction information that includes information instructing detection of duplication of news articles in the group of news articles and information about the news article group, and causes the generation AI to detect duplication of news articles.

[0019] The instruction information includes information indicating an instruction to output topic summaries from news articles and a news article set associated with the summaries. For example, the instruction information includes information such as a string "You are a capable assistant whose role is to detect duplicate news articles in a given news article collection. Please detect duplicate news articles by following the tasks below.\n\n# Tasks\nStarting from the beginning of the news article collection, extract the topic of each news article and output news articles with the same topic according to the following output format.", input format information, and output format information.

[0020] The input format information is information about the format of the information on the group of news articles to be input, for example, information expressed in JSON format, and information indicating the format for listing combinations of headlines, body snippets, and categories of news articles included in the information on the group of news articles, but is not limited to such examples.

[0021] The output format information is information about the format of the information to be output by the generation AI. For example, the output format information is information expressed in JSON format, and is information indicating a format for listing combinations of topic summaries and news article sets, but is not limited to such an example.

[0022] A news article set includes two or more news articles as overlapping news articles if there are two or more news articles with the same topic summary, and includes one news article as a non-overlapping news article if there are no news articles with the same topic summary.

[0023] As described above, the information processing device 1 causes the generation AI to output topic summaries, enabling the generation AI to detect overlapping news articles with greater accuracy than when overlapping news articles are output without outputting topic summaries.

[0024] The information processing device 1 can also improve the accuracy of detecting duplication of news articles by, for example, including in the instruction information information information indicating an instruction to output the reason for the duplication detection. The information processing device 1 can also improve the detection accuracy by, for example, including in the instruction information information information indicating an example of input and output (few-shot information).

[0025] The information processing device 1 deletes duplicate news articles from the group of news articles based on the information output from the generation AI. For example, the information processing device 1 selects one news article from the multiple news articles included in the article set for each topic summary according to a predetermined rule or randomly, and deletes the remaining news articles from the group of news articles.

[0026] The information processing device 1 can also cause the generation AI to execute a process of selecting one news article from overlapping news articles, for example. In this case, the information processing device 1 inputs, as input information to the generation AI, information including instruction information including information indicating an instruction to output one news article that is deemed most appropriate from the overlapping news articles together with a topic summary, and information including information listing combinations of topic summaries and news article sets. The information processing device 1 deletes from the group of news articles the remaining news articles other than the one news article selected by the generation AI from among the multiple news articles included in the article set in each topic summary.

[0027] Next, the deletion target detection process is a process for detecting specific news articles as deletion targets. Specific news articles include news articles with low breaking news potential, news articles in a predetermined exclusion category, and the like.

[0028] The deletion target detection process includes a first detection process for detecting news articles with low breaking news potential as first deletion targets, and a second detection process for detecting news articles with content that falls within predetermined exclusion targets as second deletion targets.

[0029] In the first detection process, a news article with low breaking news potential is a news article in a specific category for which breaking news potential is important and that was published before the threshold time Tth. For example, a news article in a specific category for which breaking news potential is important is a category in which the more time passes since the occurrence of the event indicated in the news article, the less newsworthy it becomes. In the first detection process, if there are multiple categories for which breaking news potential is important, the threshold time Tth can be set to a different value for each category for which breaking news potential is important.

[0030] In the first detection process, the information processing device 1 inputs, as input information, a prompt to the generation AI, information including instruction information that includes information instructing the generation AI to detect whether a news article in a news article group is a news article with low breaking news potential and information about the news article group, and causes the generation AI to detect information indicating whether the news article is a news article with low breaking news potential.

[0031] The instruction information in the first detection process includes information indicating an instruction to output a summary of the news article indicating whether or not the news article is likely to be a news article with low breaking news potential and information indicating whether or not the news article is likely to be a news article with low breaking news potential (for example, information on whether or not the news article is true), and information specifying the content of the news article with low breaking news potential. The information specifying the content of the news article with low breaking news potential includes, for example, information indicating the specific category described above and information indicating the threshold time Tth, but is not limited to such examples.

[0032] For example, the instruction information in the first detection process includes the string "You are an excellent assistant whose role is to determine whether a news article included in a given group of news articles is a news article with low breaking news potential. As a determination result, output a summary of the news article that clearly indicates whether the news article is a news article with low breaking news potential and truth information indicating whether the news article is a news article with low breaking news potential. The truth information should be true if the news article is a news article with low breaking news potential, and false if not." as well as information specifying the content of a news article with low breaking news potential.

[0033] In the above example, the information processing device 1 causes the generation AI to determine whether or not a news article included in a news article group is a news article with low breaking news potential, but the information processing device 1 can also cause the generation AI to determine whether or not a news article included in a news article group is a news article with high breaking news potential. A news article with high breaking news potential is a news article that is not low breaking news potential, and in this case, the instruction information includes information that specifies the content of a news article with high breaking news potential.

[0034] The prompt in the first detection process includes, in addition to the instruction information described above, input format information and output format information. The input format information in the first detection process is information on the format of the news article group, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article group, but is not limited to this example.

[0035] The output format information in the first detection process is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, and is information indicating a format for listing combinations of news article summaries and truth information, but is not limited to such an example.

[0036] In the first detection process, the information processing device 1 causes the generation AI to output a summary of the news article that clearly indicates whether the news article is not breaking news, thereby improving detection accuracy compared to when the AI ​​outputs true / false information without outputting such a summary.

[0037] The information processing device 1 can also improve the detection accuracy by, for example, further including in the instruction information information information indicating an instruction to output a reason as to whether or not the news article is of low breaking nature. The information processing device 1 can also improve the detection accuracy by, for example, including in the instruction information information information indicating an example of input and output (few-shot information).

[0038] In the first detection process, the information processing device 1 deletes news articles with low breaking news potential from the group of news articles based on the information output from the generation AI. For example, the information processing device 1 deletes news articles whose truth information indicates falsehood (e.g., false) from the group of news articles.

[0039] Next, the second detection process will be described. In the second detection process, news articles with predetermined excluded content include news articles distributed at regular intervals in standard text, news articles distributed in standard text, news articles distributed at regular intervals in standard text, and other news articles with low news value, but are not limited to such examples.

[0040] In the second detection process, the information processing device 1 inputs information containing instruction information including information instructing the generation AI to detect whether a news article in the news article group is a news article with content that is a predetermined exclusion target, and information containing information about the news article group, as input information, as a prompt, and causes the generation AI to detect news articles with content that is a predetermined exclusion target.

[0041] The instruction information in the second detection process includes, for example, information indicating an instruction to output a summary of the news article that clearly indicates whether or not the news article has content that is a predetermined exclusion target, and information indicating whether or not the news article has content that is a predetermined exclusion target (for example, truth or falsehood information), and information indicating the content of the predetermined exclusion target.

[0042] For example, the instruction information in the second detection process includes the string "You are an excellent assistant whose role is to determine whether or not a news article included in a given group of news articles is a news article with the excluded content shown below. As a determination result, output a summary of the news article that clearly indicates whether or not it is a news article with the excluded content, and truth information indicating whether or not it is a news article with the excluded content. The truth information should be true if the news article is a news article with the excluded content, and false if not." as well as information indicating the content to be excluded.

[0043] The prompt in the second detection process includes, in addition to the instruction information described above, input format information and output format information. The input format information in the second detection process is information on the format of the news article group, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article group, but is not limited to this example.

[0044] The output format information in the second detection process is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, and is information indicating a format for listing combinations of news article summaries and truth information, but is not limited to such an example.

[0045] In the second detection process, the information processing device 1 causes the generation AI to output summaries of news articles that indicate whether or not the news article has content to be excluded, thereby improving detection accuracy compared to when the AI ​​outputs truth information without outputting such summaries. Note that the information processing device 1 can also improve detection accuracy by, for example, further including in the instruction information information information indicating an instruction to output the reason for whether or not the news article has content to be excluded.

[0046] In the second detection process, the information processing device 1 deletes news articles with content to be excluded from the group of news articles based on the information output from the generation AI. For example, the information processing device 1 deletes news articles whose truth information indicates falsehood (for example, true) from the group of news articles.

[0047] Next, the weighting process will be described. The weighting process is a process for weighting news articles in a group of news articles from which duplicate news articles have been removed by the duplicate detection process and specific news articles have been deleted by the deletion target detection process.

[0048] In the weighting process, the information processing device 1 inputs instruction information including information instructing weighting of news articles in a news article group and information including information on the news article group as input information to the generation AI as a prompt, and causes the generation AI to perform weighting of the news articles in the news article group.

[0049] The instruction information in the weighting process includes information indicating an instruction to output a summary of a news article and its weight, and information specifying the weight of the news article. The information specifying the weight of a news article includes, for example, the character string "Please assign a higher weight to news that has a more direct impact on people's lives." However, the information processing device 1 is not limited to this example, and can change the content to be taken into account in the weighting process depending on the purpose. The instruction information in the weighting process may also include, for example, information specifying the weight of a news article for each category of news article.

[0050] In addition to the instruction information described above, the prompt in the weighting process includes input format information and output format information. The input format information in the weighting process is information on the format of the news article group, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article group, but is not limited to such an example.

[0051] The output format information in the weighting process is information about the format of the information to be output by the generation AI. For example, the output format information is information expressed in JSON format, and is information indicating a format for listing combinations of news article summaries and weights, but is not limited to such an example.

[0052] In the weighting process, the information processing device 1 causes the generation AI to output summaries of news articles, which can improve detection accuracy compared to when information indicating weights is output without outputting such summaries. Note that the information processing device 1 can also improve detection accuracy by, for example, further including information indicating an instruction to output the reason for the weights in the instruction information.

[0053] The weighting process is performed to reduce a characteristic, for example, when using a generation AI that has a tendency for news articles closer to the beginning of a group of news articles to be selected as news articles that meet the conditions specified in the prompt.

[0054] In the weighting process, the information processing device 1 rearranges the news articles in the group of news articles from which the news articles detected in each of the duplicate detection process and the deletion target detection process have been deleted, based on weight information indicating the weight output by the generation AI, so that the higher the weight, the higher (closer to the beginning) the article is positioned.

[0055] When the information of the news article group includes date and time information indicating the date and time of the news article (for example, the submission date and time or the creation date and time), the information processing device 1 rearranges the news articles in the news article group based on the weight information and the date and time information so that the higher the weight, the higher (closer to the beginning) the article is positioned, and if the weights are the same, the newer the date and time, the higher (closer to the beginning) the article is positioned.

[0056] In the above-mentioned pre-processing, duplicate news articles are deleted in the duplicate detection process, the first news article to be deleted is deleted in the first detection process, and the second news article to be deleted is deleted in the second detection process, but this example is not limited to this.

[0057] For example, after the duplication detection process, the first detection process, and the second detection process are completed, the information processing device 1 can also delete duplicate news articles, news articles to be deleted first, and news articles to be deleted second based on the duplication detection process, the first detection process, and the second detection process.

[0058] Next, the information processing device 1 causes the generation AI to perform a selection and rearrangement process to select and rearrange multiple news articles from the group of news articles preprocessed in step S2 (step S3). In step S3, the information processing device 1 causes the generation AI to select and rearrange multiple news articles using instruction information (an example of first instruction information) that includes information instructing the selection of multiple news articles from the group of news articles and information indicating constraints on the selection, and instruction information (an example of second instruction information) that includes information instructing the rearrangement of the multiple news articles and information indicating constraints on the rearrangement.

[0059] The selection and sorting process includes, for example, a selection process for selecting a plurality of news articles from the group of news articles preprocessed in step S2, and a sorting process for sorting the plurality of news articles selected in the selection process.

[0060] First, the selection process will be described. In the selection process, the information processing device 1 inputs, as a prompt, input information to the generation AI, which includes instruction information (an example of first instruction information) that includes information instructing the selection of multiple news articles from the group of news articles preprocessed in step S2 and information indicating selection constraints, and information including the group of news articles preprocessed in step S2, and causes the generation AI to output information on the multiple news articles selected from the group of news articles.

[0061] If the number of news articles to be selected is 15, the instruction information for the selection process includes, for example, the character string "Please select and output 15 news articles that meet the following selection criteria" and information indicating the selection criteria as information indicating constraints. News articles that meet the selection criteria are, for example, important news articles, news articles that interest many people, news articles that are highly desired to be shared, etc.

[0062] In this case, the information indicating the selection criteria is, for example, the string "Selection criteria\n\nImportant news articles\nNews articles that interest many people\nNews articles that people want to share highly," but the information is not limited to such an example, as long as the generation AI selects important news articles, news articles that interest many people, and news articles that people want to share highly.

[0063] The instruction information for the selection process may also include information specifying the number of important news articles to select, the number of news articles that are of high interest to many people to select, and the number of news articles that are highly desired to be shared to the public.

[0064] Furthermore, the information indicating the selection criteria may include, for example, information such as a character string "Please give priority to selecting news articles related to Japan" so that news articles related to Japan are given priority. Note that the information for giving priority to selecting news articles related to Japan is not limited to the above-mentioned example.

[0065] In addition to the instruction information described above, the prompt in the selection process includes input format information and output format information. The input format information in the selection process is information on the format of the news article collection, such as, but not limited to, information expressed in JSON format, which indicates a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection.

[0066] The output format information in the selection process is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, such as information indicating a format for listing the headlines of the selected news articles, but is not limited to such an example.

[0067] Next, the sorting process will be described. In the sorting process, the information processing device 1 inputs, as input information, a prompt to the generation AI, instruction information (an example of second instruction information) including information instructing the sorting of the multiple news articles selected in the selection process and information indicating constraints on the sorting, and information including the multiple news articles selected in the selection process, and causes the generation AI to output sorting information for the multiple news articles selected in the selection process.

[0068] The instruction information for the sorting process includes, for example, the string "Please sort the given multiple news articles according to the following sorting criteria." and information indicating the sorting criteria as information indicating constraints on the sorting. The information indicating the sorting criteria includes, for example, the string "Sorting criteria\n\nSort the news articles in order of importance according to the following editorial policy.\n\nThe editorial policy is based on the following elements, and the following additional criteria should be considered:\nRapidity, public interest, reliability, novelty, interest, and recognition\n\nAdditional criteria\n·Preferentially give priority to news articles in Japan\n·Preferentially give priority to news that is close to home\n·News articles published in the past few hours," but is not limited to such examples.

[0069] In addition to the instruction information described above, the prompt for the sorting process includes input format information and output format information. The input format information is information about the format of the news article collection, such as, but not limited to, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection.

[0070] The output format information in the sorting process is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, and is information indicating a format in which m combinations of news article headlines and news article summaries are listed as one set, but is not limited to such an example. m is an integer greater than or equal to 1.

[0071] In the sorting process, the information processing device 1 causes the generation AI to output summaries of the news articles, which can improve sorting accuracy compared to sorting news articles without outputting such summaries. Note that the information processing device 1 can also improve sorting accuracy by, for example, further including information indicating an instruction to output the reason for sorting in the instruction information.

[0072] Next, the information processing device 1 executes summarization processing (step S4) to have the generation AI execute summarization of each of the multiple news articles selected and rearranged in step S3. The summarization processing is processing to have the generation AI execute summarization of each of the multiple news articles rearranged in step S3 using instruction information (an example of third instruction information) including information instructing the generation AI to summarize each of the multiple news articles selected and rearranged in step S3 and a prompt including information on the multiple news articles rearranged in step S3.

[0073] The summary of each of the plurality of news articles sorted in step S3 is, for example, a summary of the headline and body snippet of each of the plurality of news articles sorted in step S3, but is not limited to this example.

[0074] The instruction information for the summarization process includes, for example, the information of the character string "Please summarize the headline and text of each of the multiple news articles provided." The instruction information for the summarization process also includes the information of the character string "Please summarize the headline and text of each of the multiple news articles provided, using the following input / output examples as reference," and information indicating the input / output examples. The information indicating the input / output examples is information in which the headline and text snippets of news articles and the summarized headline and text snippets are used as input examples, and includes information of multiple input / output examples.

[0075] The instruction information for the summarization process may include information specifying the number of characters in the headline summary and information specifying the number of characters in the body snippet summary. For example, the instruction information for the summarization process may include, but is not limited to, a string such as "The headline summary of a news article should be about 15 characters long, and the body summary of a news article should be about 30 characters long."

[0076] In addition to the instruction information described above, the prompt in the summarization process includes input format information and output format information. The input format information in the summarization process is information on the format of the news article group, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article group, but is not limited to such an example.

[0077] For example, the input format information may include news article body information instead of news article body snippet information, in which case the news article information included in the prompt for summarization includes news article body information instead of news article body snippet information.

[0078] The output format information in the summarization process is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, such as information indicating a format for listing combinations of headlines and body snippets of summarized news articles, but is not limited to such an example.

[0079] Next, the information processing device 1 distributes the plurality of news articles summarized in step S4 (step S5). For example, when the information processing device 1 receives a distribution request from a dedicated application installed on the terminal device 2, it transmits the plurality of news articles summarized in step S4 to the terminal device 2. The terminal device 2 receives the plurality of news articles transmitted from the information processing device 1 and displays the received plurality of news articles.

[0080] In the example shown in Figure 1, the first three news articles among the multiple news articles summarized in step S4 are displayed on the terminal device 2. The user of the terminal device 2 can select the "Next News" button, in which case the terminal device 2 displays the next three news articles.

[0081] In this way, the information processing device 1 causes the generation AI to select and rearrange multiple news articles using instruction information (an example of first instruction information) that includes information instructing the selection of multiple news articles and information indicating the constraints on the selection, and instruction information (an example of second instruction information) that includes information instructing the rearrangement of multiple news articles and information indicating the constraints on the rearrangement, and then causes the generation AI to summarize each of the multiple news articles using instruction information (an example of second instruction information) that includes information instructing the summary of each of the rearranged multiple news articles.

[0082] This allows the information processing device 1 to appropriately narrow down the news articles to be summarized compared to when summarizing all news articles in the news article group to be distributed, and to effectively summarize the news articles.In addition, the information processing device 1 rearranges the news articles before summarizing, so that the information processing device 1 can appropriately summarize the news articles.

[0083] The configuration of an information processing system including the information processing device 1, the terminal device 2, and the uploader terminal 3 that perform such processing will be described in detail below.

[0084] [2. Information Processing System Configuration] 2 is a diagram illustrating an example of the configuration of an information processing system 100 according to the embodiment. As illustrated in FIG. 2, the information processing system 100 according to the embodiment includes an information processing device 1, a plurality of terminal devices 2, and a plurality of uploader terminals 3.

[0085] The multiple terminal devices 2 are used by different users U. The multiple uploader terminals 3 are used by different uploaders. Each terminal device 2 and each uploader terminal 3 is, for example, a notebook PC (Personal Computer), a desktop PC, a smartphone, a tablet PC, or a wearable device. The wearable device is, for example, smart glasses or a smart watch, but is not limited to these examples.

[0086] The information processing device 1, the terminal device 2, and the uploader terminal 3 are connected to each other via a network N so as to be able to communicate with each other via a wired or wireless connection. Note that the information processing system 100 shown in FIG. 2 may include a plurality of information processing devices 1 and the like.

[0087] The network N includes, for example, a WAN (Wide Area Network) such as the Internet and a mobile communication network such as LTE (Long Term Evolution), 4G (4th Generation), and 5G (5th Generation: 5th generation mobile communication system).

[0088] Each terminal device 2 and each uploader terminal 3 can connect to the network N via short-range wireless communication such as a mobile communication network, Bluetooth (registered trademark), or wireless LAN (Local Area Network), and communicate with the information processing device 1.

[0089] 3. Configuration of Information Processing Device 1 3 is a diagram showing an example of the configuration of the information processing device 1 according to the embodiment. As shown in FIG. 3, the information processing device 1 includes a communication unit 10, a storage unit 11, and a processing unit 12.

[0090] [3.1. Communication Unit 10] The communication unit 10 is realized by, for example, a communication module or a network interface card (NIC). The communication unit 10 is connected to a network N by wire or wirelessly, and transmits and receives information to and from various other devices. For example, the communication unit 10 transmits and receives information to and from the terminal device 2 via the network N.

[0091] [3.2. Storage section 11] The storage unit 11 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 11 has a news article group information storage unit 20 and a summary information storage unit 21.

[0092] 3.2.1. News Article Group Information Storage Unit 20 The news article group information storage unit 20 stores various information related to news articles included in a news article group. Fig. 4 is a diagram showing an example of a news article group information table stored in the news article group information storage unit 20 of the information processing device 1 according to the embodiment.

[0093] 4, the news article group information table stored in the news article group information storage unit 20 includes information items such as “Article ID (Identifier),” “Submitter ID,” and “News Article Information.” The “Article ID” is an identifier that identifies a news article included in a news article group, and is information assigned to each news article.

[0094] The "Submitter ID" is information about the submitter who submitted the news article corresponding to the "Article ID." The "News Article Information" includes, but is not limited to, information about the news article corresponding to the "Article ID," such as the headline, text, text snippet, image, category, submission date and time (or creation date and time), keywords, submitter, and links to related articles.

[0095] 3.2.2. Summary Information Storage Unit 21 The summary information storage unit 21 stores various types of information about news articles summarized by the processing unit 12. Fig. 5 is a diagram showing an example of a summary information table stored in the summary information storage unit 21 of the information processing device 1 according to the embodiment.

[0096] In the example shown in Figure 5, the summary information table stored in the summary information storage unit 21 includes information under the items "Summary ID," "Article ID," and "Summary Information." The "Summary ID" is an identifier that identifies a summary and is information assigned to each summary. The "Article ID" is the article ID of the news article corresponding to the summary indicated by the "Summary ID."

[0097] The "summary information" is information about the summary indicated by the "summary ID," and may include, for example, information about a summary of the headline of a news article and a summary of a body snippet. The "summary information" may also include information about the date and time when the summary indicated by the "summary ID" was created.

[0098] [3.3. Processing Unit 12] The processing unit 12 is a controller, and is realized, for example, by a CPU (Central Processing Unit) or an MPU (Micro Processing Unit) executing various programs stored in a storage device inside the information processing device 1 using RAM as a working area.

[0099] The processing unit 12 may be partially or entirely realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0100] 3, the processing unit 12 has an acquisition unit 30, a reception unit 31, a preprocessing unit 32, a selection and sorting unit 33, a summarization unit 34, and a distribution unit 35, and realizes or executes the functions and actions of information processing described below. Note that the internal configuration of the processing unit 12 is not limited to the configuration shown in FIG. 3, and may be any other configuration that performs the information processing described below.

[0101] [3.3.1. Acquisition part 30] The acquisition unit 30 acquires various pieces of information from an external information processing device or the terminal device 2 via the communication unit 10, and stores the acquired information in the storage unit 11.

[0102] The acquisition unit 30 also acquires various types of information from the storage unit 11. For example, the acquisition unit 30 acquires information about news article groups from the news article group information storage unit 20, etc. The acquisition unit 30 also acquires summary information from the summary information storage unit 21, etc.

[0103] [3.3.2. Reception Unit 31] The receiving unit 31 receives various requests and information from the terminal device 2 and the uploader terminal 3 via the communication unit 10. For example, the receiving unit 31 receives a summary delivery request transmitted from the terminal device 2.

[0104] The reception unit 31 also receives a manuscript submission request transmitted from the manuscript submitter terminal 3. The manuscript submission request includes information such as news articles. When the reception unit 31 receives a manuscript submission request, the reception unit 31 stores the information such as news articles included in the manuscript submission request in the news article group information storage unit 20.

[0105] The news article included in the submission request may include information such as the news article headline, body text, body text snippets, images, category, submission date and time (or creation date and time), keywords, submitter, and links to related articles, but is not limited to these examples.

[0106] 3.3.3. Pre-processing section 32 The preprocessing unit 32 uses a generation AI to perform preprocessing on the group of news articles acquired by the acquisition unit 30. For example, the preprocessing unit 32 uses a generation AI to perform preprocessing on the group of news articles from which multiple news articles are selected by the selection and sorting unit 33.

[0107] The information about the news articles used for preprocessing is represented by, but is not limited to, a list of news articles that lists the combinations of headlines, body snippets, and categories of the news articles included in the news article group. The news article group is represented in, but is not limited to, JSON format.

[0108] The generative AI is a text generation AI, which is, for example, a large-scale language model trained to estimate and output the next token from an input token sequence, such as a transformer-based model or an RNN-based model, or may be a hybrid model of these.

[0109] Examples of the transformer-based model include, but are not limited to, GPT, PaLM2, or LLaMA. Examples of the RNN-based model include, but are not limited to, RWKV. It is desirable that the generation AI be trained so as not to include personal information in the generated results. The generation AI is located in an external information processing device, and the information processing device 1 uses the generation AI via an API, but the generation AI may also be located within the information processing device 1.

[0110] The pre-processing unit 32 includes a duplication detection unit 40 that detects duplication of news articles in the news article group, a deletion target detection unit 41 that detects news articles to be deleted in the news article group, and a weighting unit 42 that weights the news articles in the news article group. The duplication detection unit 40, deletion target detection unit 41, and weighting unit 42 will be described below in that order.

[0111] [3.3.3.1. Duplication detection unit 40] The duplication detection unit 40 uses the generation AI to detect duplication of news articles in the group of news articles. The duplication detection unit 40 uses the generation AI to perform the detection process of duplication of news articles, for example, by using instruction information including an instruction to detect duplication of news articles in the group of news articles.

[0112] For example, the duplication detection unit 40 inputs instruction information including information instructing the detection of duplication of news articles in a group of news articles and information including information on the group of news articles as input information, which is a prompt, to the generation AI, and causes the generation AI to detect duplication of news articles.

[0113] The instruction information includes information indicating an instruction to output topic summaries from news articles and a news article set associated with the summaries. For example, the instruction information includes information such as a string "You are a capable assistant whose role is to detect duplicate news articles in a given news article collection. Please detect duplicate news articles by following the tasks below.\n\n# Tasks\nStarting from the beginning of the news article collection, extract the topic of each news article and output news articles with the same topic according to the following output format.", input format information, and output format information.

[0114] The input format information is information about the format of the information on the group of news articles to be input, for example, information expressed in JSON format, and information indicating the format for listing combinations of headlines, body snippets, and categories of news articles included in the information on the group of news articles, but is not limited to such examples.

[0115] The output format information is information about the format of the information to be output by the generation AI. For example, the output format information is information expressed in JSON format, and is information indicating a format for listing combinations of topic summaries and news article sets, but is not limited to such an example.

[0116] A news article set includes two or more news articles as overlapping news articles if there are two or more news articles with the same topic summary, and includes one news article as a non-overlapping news article if there are no news articles with the same topic summary.

[0117] As described above, the duplication detection unit 40 causes the generation AI to output topic summaries, enabling the generation AI to detect overlapping news articles with greater accuracy than when the generation AI is caused to output overlapping news articles without outputting topic summaries.

[0118] The duplication detection unit 40 can improve the accuracy of detecting duplication of news articles by, for example, including in the instruction information information indicating an instruction to output the reason for the duplication detection. The duplication detection unit 40 can also improve the detection accuracy by, for example, including in the instruction information information indicating an example of input and output (few-shot information).

[0119] The duplication detection unit 40 deletes duplicate news articles from the group of news articles based on the information output from the generation AI. For example, the duplication detection unit 40 selects one news article from the multiple news articles included in the article set for each topic summary according to a predetermined rule or randomly, and deletes the remaining news articles from the group of news articles.

[0120] The duplication detection unit 40 can also cause the generation AI to execute a process of selecting one news article from overlapping news articles, for example. In this case, the duplication detection unit 40 inputs, as input information to the generation AI, instruction information including information indicating an instruction to output one news article that is deemed most appropriate from the overlapping news articles together with a topic summary, and information including information listing combinations of topic summaries and news article sets.

[0121] The duplication detection unit 40 deletes from the group of news articles all the news articles other than the one news article selected by the generation AI from among the multiple news articles included in the article set in the summary of each topic. Note that the duplication detection unit 40 can also prevent the generation AI from outputting the topic summary.

[0122] [3.3.3.2. Deletion target detection unit 41] The deletion target detection unit 41 is a process for detecting specific news articles as deletion targets. For example, the deletion target detection unit 41 uses instruction information including an instruction to detect specific news articles to be deleted from a group of news articles, and performs a process for detecting specific news articles using a generation AI. The specific news articles include news articles that do not require breaking news, news articles in a predetermined exclusion category, etc.

[0123] 6 is a diagram showing an example of the configuration of the deletion target detection unit 41 of the pre-processing unit 32 in the processing unit 12 of the information processing device 1 according to the embodiment. As shown in FIG. 6, the deletion target detection unit 41 includes a first detection unit 50 that detects news articles with low breaking news potential as first deletion targets, and a second detection process that detects news articles with content that is a predetermined exclusion target as second deletion targets.

[0124] 3.3.3.2.1. First detection unit 50 A first detection unit 50 detects news articles with low breaking news potential as first deletion targets. News articles with low breaking news potential are news articles in a specific category for which breaking news potential is important, and are news articles published before a threshold time Tth.

[0125] For example, news articles in a specific category for which breaking news is important are those in which the more time passes since the occurrence of an event described in the news article, the less newsworthy the article becomes. In the first detection unit 50, when there are multiple categories for which breaking news is important, the threshold time Tth can be set to a different value for each category for which breaking news is important.

[0126] The first detection unit 50 inputs information including instruction information including information instructing the generation AI to detect whether a news article in the news article group is a news article with low breaking news potential and information about the news article group as input information, which is a prompt, and causes the generation AI to detect information indicating whether the news article is a news article with low breaking news potential.

[0127] The instruction information includes information indicating an instruction to output a summary of the news article, which indicates whether the news article is likely to be a news article with low breaking news potential, and information indicating whether the news article is likely to be a news article with low breaking news potential (for example, true / false information), and information specifying the content of the news article with low breaking news potential. The information specifying the content of the news article with low breaking news potential includes, for example, information indicating the specific category described above and information indicating the threshold time Tth, but is not limited to such examples.

[0128] The instruction information includes, for example, the string "You are an excellent assistant whose role is to determine whether a news article included in a given group of news articles is a news article with low breaking news potential. As a determination result, output a summary of the news article that clearly indicates whether the news article is a news article with low breaking news potential, and truth information indicating whether the news article is a news article with low breaking news potential. The truth information should be true if the news article is a news article with low breaking news potential, and false if not." as well as information specifying the content of a news article with low breaking news potential.

[0129] In the example described above, the first detection unit 50 causes the generation AI to determine whether or not a news article included in the news article group is a news article with low breaking news potential, but the generation AI can also cause the AI ​​to determine whether or not a news article included in the news article group is a news article with high breaking news potential. A news article with high breaking news potential is a news article that is not low breaking news potential, and in this case, the instruction information includes information that specifies the content of a news article with high breaking news potential.

[0130] In addition to the instruction information described above, the prompt includes input format information and output format information. The input format information is information about the format of the news article collection, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection, but is not limited to such an example.

[0131] The output format information is information about the format of the information to be output by the generation AI. For example, the output format information is information expressed in JSON format, and is information indicating a format for listing combinations of news article summaries and truth information, but is not limited to such an example.

[0132] The first detection unit 50 causes the generation AI to output a summary of the news article that clearly indicates whether the news article is not breaking news, thereby improving detection accuracy compared to when the AI ​​outputs true / false information without outputting such a summary.

[0133] The first detection unit 50 can also improve the detection accuracy by, for example, including in the instruction information information that further includes information indicating an instruction to output a reason as to whether or not the news article is of low breaking quality. The first detection unit 50 can also improve the detection accuracy by, for example, including in the instruction information information that information indicating input / output examples (few-shot information).

[0134] The first detection unit 50 deletes news articles with low breaking news potential from the group of news articles based on the information output from the generation AI. For example, the first detection unit 50 deletes news articles whose truth information indicates falsehood (e.g., false) from the group of news articles.

[0135] [3.3.3.2.2. Second detection unit 51] The second detection unit 51 detects news articles with predetermined content to be excluded as second deletion targets. In the second detection unit 51, news articles with predetermined content to be excluded are, for example, news articles distributed at regular times, news articles distributed in standard phrases, news articles distributed at regular times in standard phrases, etc., and are, for example, news articles with low news value, but are not limited to such examples.

[0136] The second detection unit 51 inputs information containing instruction information including information instructing the generation AI to detect whether a news article in the news article group is a news article with content that is a predetermined exclusion target, and information containing information about the news article group, as input information, as a prompt, and causes the generation AI to detect news articles with content that is a predetermined exclusion target.

[0137] The instruction information includes, for example, information indicating an instruction to output a summary of a news article that clearly indicates whether or not the news article has content that is a predetermined exclusion target, and information indicating whether or not the news article has content that is a predetermined exclusion target (for example, truth information), and information indicating the content of the predetermined exclusion target.

[0138] The instruction information includes, for example, the string "You are an excellent assistant whose role is to determine whether or not a news article included in a given group of news articles is a news article with the excluded content shown below. As a determination result, output a summary of the news article that clearly indicates whether or not it is a news article with the excluded content, and truth information indicating whether or not it is a news article with the excluded content. The truth information should be true if the news article is a news article with the excluded content, and false if not." as well as information indicating the content to be excluded.

[0139] In addition to the instruction information described above, the prompt includes input format information and output format information. The input format information is information about the format of the news article collection, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection, but is not limited to such an example.

[0140] The output format information is information about the format of the information to be output by the generation AI. For example, the output format information is information expressed in JSON format, and is information indicating a format for listing combinations of news article summaries and truth information, but is not limited to such an example.

[0141] In the second detection process, the second detection unit 51 causes the generation AI to output summaries of news articles that indicate whether or not the news article has content to be excluded, thereby improving detection accuracy compared to when the AI ​​outputs truth information without outputting such summaries. Note that the second detection unit 51 can also improve detection accuracy by, for example, further including in the instruction information information information indicating an instruction to output the reason for whether or not the news article has content to be excluded.

[0142] In the second detection process, the second detection unit 51 deletes news articles with content to be excluded from the group of news articles based on the information output from the generation AI. For example, the second detection unit 51 deletes news articles whose truth information indicates falsehood (for example, true) from the group of news articles.

[0143] In the above-mentioned pre-processing, duplicate news articles are deleted in the duplicate detection process in the duplicate detection unit 40, the first news article to be deleted is deleted in the first detection process in the first detection unit 50, and the second news article to be deleted is deleted in the second detection process in the second detection unit 51, but this example is not limited to this.

[0144] For example, after the duplication detection process in the duplication detection unit 40, the first detection process in the first detection unit 50, and the second detection process in the second detection unit 51 are completed, the pre-processing unit 32 can also delete duplicate news articles, news articles to be deleted first, and news articles to be deleted second based on the duplication detection process, the first detection process, and the second detection process.

[0145] [3.3.3.3. Weighting Unit 42] The weighting unit 42 weights the news articles in the group of news articles from which the overlapping news articles have been deleted by the overlap detection unit 40 and the specific news articles have been deleted by the deletion target detection unit 41.

[0146] The weighting unit 42 uses the generation AI to weight the news articles using instruction information including an instruction to weight the news articles in the news article group, for example. For example, the weighting unit 42 inputs information including instruction information including information instructing weighting of the news articles in the news article group and information about the news article group to the generation AI as a prompt, which is input information, and causes the generation AI to weight the news articles in the news article group.

[0147] The instruction information includes information indicating an instruction to output a summary of a news article and a weight of the news article, and information specifying the weight of the news article. The information specifying the weight of a news article includes, for example, a character string such as "Please assign a higher weight to news that has a more direct impact on people's lives." However, the information is not limited to this example, and the weighting unit 42 can change the content to be taken into account in the weighting depending on the purpose. The instruction information may also include, for example, information specifying the weight of a news article for each category of news article.

[0148] In addition to the instruction information described above, the prompt includes input format information and output format information. The input format information is information about the format of the news article collection, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection, but is not limited to such an example.

[0149] The output format information is information about the format of the information to be output by the generation AI. For example, the output format information is information expressed in JSON format, and is information indicating a format for listing combinations of news article summaries and weights, but is not limited to such an example.

[0150] The weighting unit 42 causes the generation AI to output summaries of news articles, which can improve detection accuracy compared to when information indicating weights is output without outputting such summaries. Note that the weighting unit 42 can also improve detection accuracy by, for example, further including information indicating an instruction to output the reason for the weights in the instruction information.

[0151] The weighting unit 42 is used to reduce a characteristic, for example, when using a generation AI that has a tendency for news articles closer to the beginning of a group of news articles to be selected as news articles that meet the conditions specified in the prompt.

[0152] The weighting unit 42 causes the generation AI to output the weights of the news articles in the news article group, and then rearranges the news articles in the news article group based on the weights. For example, the weighting unit 42 rearranges the news articles in the news article group from which the news articles detected by each of the duplication detection unit 40 and the deletion target detection unit 41 have been deleted, based on the weight information indicating the weights output by the generation AI, so that the news articles with larger weights are positioned higher (closer to the beginning).

[0153] If the information in the news article group includes date and time information indicating the date and time of the news article (for example, the submission date and time or the creation date and time), the weighting unit 42 rearranges the news articles in the news article group based on the weight information and the date and time information so that the greater the weight, the higher (closer to the beginning) the article will be positioned, and if the weights are the same, the newer the date and time will be positioned higher (closer to the beginning).

[0154] [3.3.4. Selection and sorting unit 33] The selection and sorting unit 33 causes the generation AI to select and sort multiple news articles from the group of news articles that have been preprocessed by the preprocessing unit 32. For example, the selection and sorting unit 33 uses the generation AI to select multiple news articles, and then uses the generation AI to sort the multiple news articles.

[0155] The selection and rearrangement unit 33 causes the generation AI to select and rearrange multiple news articles using, for example, instruction information (an example of first instruction information) that includes information instructing the selection of multiple news articles from a group of news articles and information indicating constraints on the selection, and instruction information (an example of second instruction information) that includes information instructing the rearrangement of multiple news articles and information indicating constraints on the rearrangement.

[0156] The selection and sorting unit 33 includes, for example, a selection unit 45 that selects multiple news articles from the group of news articles that have been preprocessed by the preprocessing unit 32, and a sorting unit 36 ​​that sorts the multiple news articles selected by the selection unit 45.

[0157] [3.3.4.1. Selection unit 45] The selection unit 45 inputs instruction information (an example of first instruction information) that includes information instructing the selection of multiple news articles from the group of news articles that have been preprocessed by the preprocessing unit 32 and information indicating selection constraints, and information that includes the group of news articles that have been preprocessed by the preprocessing unit 32, as input information, a prompt, to the generation AI, and causes the generation AI to output information on the multiple news articles selected from the group of news articles.

[0158] If the number of news articles to be selected is 15, the instruction information includes, for example, a character string "Please select and output 15 news articles that satisfy the following selection criteria" and information indicating the selection criteria as information indicating constraints. News articles that satisfy the selection criteria are, for example, important news articles, news articles that interest many people, news articles that are highly desired to be shared, etc.

[0159] In this case, the information indicating the selection criteria is, for example, the string "Selection criteria\n\nImportant news articles\nNews articles that interest many people\nNews articles that people want to share highly," but the information is not limited to such an example, as long as the generation AI selects important news articles, news articles that interest many people, and news articles that people want to share highly.

[0160] The instruction information may also include information specifying the number of important news articles to select, the number of news articles that are of high interest to many people to select, and the number of news articles that are highly desired to be reported to the public.

[0161] Furthermore, the information indicating the selection criteria may include, for example, information such as a character string "Please give priority to selecting news articles related to Japan" so that news articles related to Japan are given priority. Note that the information for giving priority to selecting news articles related to Japan is not limited to the above-mentioned example.

[0162] In addition to the instruction information described above, the prompt includes input format information and output format information. The input format information is information about the format of the news article group, such as information expressed in JSON format, and is information indicating a format that lists combinations of headlines, body snippets, and categories of news articles included in the news article group, but is not limited to this example. Note that the input format information may also include information about weights assigned to the news articles by the weighting unit 42.

[0163] The output format information is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, such as information indicating a format for listing the headlines of selected news articles, but is not limited to such an example.

[0164] [3.3.4.2. Sorting Unit 46] The sorting unit 46 inputs instruction information (an example of second instruction information) that includes information instructing the sorting of the multiple news articles selected by the selection unit 45 and information indicating constraints on the sorting, and information that includes the multiple news articles selected by the selection unit 45, as input information, a prompt, to the generation AI, and causes the generation AI to output sorting information for the multiple news articles selected by the selection unit 45.

[0165] The instruction information includes, for example, the string "Please sort the given multiple news articles according to the following sorting criteria." and information indicating the sorting criteria as information indicating constraints on the sorting. The information indicating the sorting criteria includes, for example, the string "Sorting criteria\n\nSort the news articles in order of importance according to the following editorial policy.\n\nThe editorial policy is based on the following elements, and the following additional criteria should be considered:\nRapidity, public interest, reliability, novelty, interest, and recognition\n\nAdditional criteria\n·Priority on news articles in Japan\n·Priority on familiar news articles\n·News articles published in the past few hours," but is not limited to such examples.

[0166] In addition to the instruction information described above, the prompt includes input format information and output format information. The input format information is information about the format of the news article collection, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection, but is not limited to such an example.

[0167] The output format information is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, and is information indicating a format in which m combinations of news article headlines and news article summaries are listed as one set, but is not limited to such an example. m is an integer greater than or equal to 1.

[0168] The sorting unit 46 causes the generation AI to output summaries of the news articles, which can improve the accuracy of sorting compared to when the news articles are sorted without outputting such summaries. Note that the sorting unit 46 can also improve the accuracy of sorting by, for example, further including information indicating an instruction to output the reason for sorting in the instruction information.

[0169] [3.3.5. Summary Section 34] The summarizing unit 34 causes the generating AI to perform summarization of each of the plurality of news articles selected and sorted by the selecting and sorting unit 33.

[0170] The summarization unit 34 is a process that causes the generation AI to perform a summary of each of the multiple news articles by the selection and sorting unit 33 using, for example, instruction information (an example of third instruction information) that includes information instructing a summary of each of the multiple news articles sorted by the selection and sorting unit 33 and a prompt that includes information on the multiple news articles sorted by the selection and sorting unit 33.

[0171] The summary of each of the multiple news articles sorted by the selection sorting unit 33 is, for example, a summary of the headline and body text snippet of each of the multiple news articles sorted by the selection sorting unit 33, but is not limited to such an example.

[0172] The instruction information includes, for example, the information of a character string "Please summarize the headline and text of each of the multiple news articles provided." The instruction information also includes the information of a character string "Please summarize the headline and text of each of the multiple news articles provided, using the following input / output examples as reference," and information indicating the input / output examples. The information indicating the input / output examples is information in which the headline and text snippets of news articles and the summarized headline and text snippets are used as input examples, and includes information of multiple input / output examples.

[0173] The instruction information may also include information specifying the number of characters in a headline summary and information specifying the number of characters in a body snippet summary. For example, the instruction information may include, but is not limited to, a string such as "The headline summary of a news article should be about 15 characters long, and the body summary of a news article should be about 30 characters long."

[0174] In addition to the instruction information described above, the prompt includes input format information and output format information. The input format information is information about the format of the news article collection, for example, information expressed in JSON format, and information indicating a format for listing combinations of headlines, body snippets, and categories of the news articles included in the news article collection, but is not limited to such an example.

[0175] For example, the input format information may include news article body information instead of news article body snippet information, in which case the news article information included in the prompt includes news article body information instead of news article body snippet information.

[0176] The output format information is information about the format of the information to be output by the generation AI. The output format information is, for example, information expressed in JSON format, such as information indicating a format for listing combinations of headlines and body snippets of summarized news articles, but is not limited to such an example.

[0177] In the above example, the selection and sorting unit 33 separately executes a selection process for selecting multiple news articles from a group of news articles and a sorting process for sorting the multiple news articles, but the selection process and the sorting process may be performed at the same time.

[0178] For example, the selection and rearrangement unit 33 can cause the generation AI to select and rearrange multiple news articles using a prompt including information on the news article group, and information including instruction information (an example of first instruction information) that includes information instructing the selection of multiple news articles from a group of news articles and information indicating constraints on the selection, and instruction information (an example of second instruction information) that includes information instructing the rearrangement of multiple news articles and information indicating constraints on the rearrangement.

[0179] [3.3.6. Distribution Unit 35] The distribution unit 35 distributes various pieces of information. The distribution unit 35 transmits the various pieces of information to the terminal device 2 via the communication unit 10 and the network N, thereby distributing the various pieces of information.

[0180] For example, the distribution unit 35 distributes a plurality of news articles summarized by the summarization unit 34. For example, the distribution unit 35 distributes the plurality of news articles summarized by the summarization unit 34 by transmitting the plurality of news articles summarized by the summarization unit 34 to the terminal device 2 via the communication unit 10 and the network N.

[0181] When the distribution unit 35 receives a distribution request from a dedicated application installed on the terminal device 2, it transmits the plurality of news articles summarized by the summarization unit 34 to the terminal device 2. The terminal device 2 receives the plurality of news articles transmitted from the information processing device 1 and displays the received plurality of news articles.

[0182] 7 is a diagram showing an example of an application screen distributed by the distribution unit 35 in the processing unit 12 of the information processing device 1 according to the embodiment and displayed on the terminal device 2. The application screen 60 shown in FIG. 7 is a screen displayed by a dedicated application, and includes a search box 61, a search button 62, a tab row 63, and tab contents 64 and 70.

[0183] A search keyword is input into the search box 61 by the user U operating the terminal device 2. When the user U operates the terminal device 2 to select a search button 62 with the search keyword input into the search box 61, the terminal device 2 transmits a search query including the search keyword input into the search box 61 to the information processing device 1. The distribution unit 35 transmits search results according to the search keyword included in the search query to the terminal device 2. The search results are displayed by the terminal device 2.

[0184] The tab column 63 includes a plurality of tabs 63a, 63b, 63c, 63d, and 63e that can be selected by the user U. The tab 63a is the "All" tab, the tab 63b is the "News" tab, the tab 63c is the "Sports" tab, the tab 63e is the "Topics" tab, and the tab 63e is the "IT" tab. In the example shown in FIG. 7, the "News" tab is selected and highlighted.

[0185] Tab contents 64, 70 are arranged with a plurality of contents corresponding to the tab selected in tab row 63. In the example shown in Fig. 7, tab content 64 includes a title 65, summary information 66a, 66b, 66c, explanation information 67, and a next button 68. Title 65 includes the information of the character string "AI Digest Updated at 1 PM", which indicates that summary information 66a, 66b, 66c are summaries by AI and were updated at 1 PM.

[0186] Each of the summary information 66a, 66b, and 66c includes a summary of a headline and a summary of a body snippet of different news articles that are information about the news articles summarized by the summarizing unit 34. Each of the summary information 66a, 66b, and 66c also includes link information to the corresponding news article, and the user U can select one of the summary information 66a, 66b, and 66c, causing the news article corresponding to the selected summary information to be displayed on the terminal device 2.

[0187] The explanatory information 67 includes information explaining the creator of the summary information, and in the example shown in Fig. 7, it includes the string "This function uses AI, and accuracy is not guaranteed." This allows the user U to easily understand that the summary information 66a, 66b, and 66c were generated by AI.

[0188] The next button 68 is a GUI (Graphical User Interface) button for displaying on the terminal device 2 news articles other than the news article displayed on the terminal device 2, among the multiple news articles summarized by the summarizing unit 34. In the example shown in FIG. 7, the first three news articles among the multiple news articles summarized by the summarizing unit 34 are displayed on the terminal device 2. By selecting the next button 68, the user U causes the next three news articles to be displayed on the terminal device 2.

[0189] The tab content 70 includes article contents 71a and 71b. Each of the article contents 71a and 71b includes a headline and a body snippet of a news article. Each of the article contents 71a and 71b includes link information to the corresponding news article, and the user U can select either of the article contents 71a and 71b, causing the news article corresponding to the selected article content to be displayed on the terminal device 2.

[0190] The article content included in tab content 70 includes headlines and body snippets of news articles included in the group of news articles, and headlines and body snippets of news articles included in the group of news articles that have not been summarized by summarization unit 34. Furthermore, the article content included in tab content 70 includes headlines and body snippets of news articles summarized by summarization unit 34, but does not necessarily have to include headlines and body snippets of news articles summarized by summarization unit 34.

[0191] [4. Processing Procedure] Next, a procedure of information processing by the processing unit 12 of the information processing device 1 according to the embodiment will be described. Fig. 8 is a flowchart showing an example of information processing by the processing unit 12 of the information processing device 1 according to the embodiment.

[0192] 8, the processing unit 12 of the information processing device 1 determines whether or not a news article has been submitted (step S10). If the processing unit 12 determines that a news article has been submitted (step S10: Yes), the processing unit 12 stores information about the submitted news article in the storage unit 11 (step S11).

[0193] When the processing of step S11 is completed or when it is determined that no news article has been submitted (step S10: No), the processing unit 12 determines whether or not it is time to summarize (step S12). The summarization timing is a timing that occurs at a predetermined cycle or a timing designated by the administrator of the information processing device 1, but is not limited to such examples.

[0194] When the processing unit 12 determines that it is time to summarize (step S12: Yes), it acquires information on a group of news articles including multiple news articles up to a predetermined period ago from the storage unit 11 or the like (step S13).Then, the processing unit 12 detects and deletes a specific news article from the group of news articles about which information was acquired in step S13 (step S14).

[0195] Next, the processing unit 12 detects overlapping news articles in the group of news articles and deletes all but one of the detected news articles to remove the overlapping news articles (step S15).Then, the processing unit 12 weights the news articles in the group of news articles (step S16).

[0196] Next, the processing unit 12 selects a plurality of news articles from the group of news articles (step S17), rearranges the selected plurality of news articles (step S18), and then summarizes the selected plurality of news articles (step S19), and delivers the summarized plurality of news articles (step S20).

[0197] When the processing of step S20 is completed or when it is determined that the timing for summarizing has not come (step S12: No), the processing unit 12 determines whether the timing for ending the operation has come (step S21). The processing unit 12 determines that the timing for ending the operation has come when, for example, the power of the information processing device 1 is turned off.

[0198] If the processing unit 12 determines that the operation end time has not yet arrived (step S21: No), it proceeds to step S10, and if it determines that the operation end time has arrived (step S21: Yes), it terminates the processing shown in Figure 8.

[0199] [5. Modifications] In the example described above, the summarizing unit 34 collectively includes information about multiple news articles selected and rearranged by the selecting and rearranging unit 33 in the input information and inputs it to the generation AI, but this is not a limitation. For example, the summarizing unit 34 can include information about each of the multiple news articles selected and rearranged by the selecting and rearranging unit 33 individually in the input information, thereby causing the generation AI to summarize each of the multiple selected and rearranged news articles individually. This allows the summarizing unit 34 to avoid confusion between the information in the multiple news articles compared to, for example, summarizing multiple news articles selected by the selecting and rearranging unit 33 collectively, thereby increasing the likelihood of improving the accuracy of the summary and enabling the generation of summaries of multiple news articles in a more optimal order without including incorrect information.

[0200] The summarizing unit 34 can also divide the information on multiple news articles selected and rearranged by the selecting and rearranging unit 33 into information on p (p is an integer greater than or equal to 2) news articles, and include information on the p news articles in the input information for each of the p divided news articles, thereby causing the generation AI to summarize each of the p news articles for each of the p news articles. This also allows the summarizing unit 34 to avoid confusion between the information on multiple news articles compared to, for example, summarizing multiple news articles selected by the selecting and rearranging unit 33 all at once, thereby increasing the possibility of improving the accuracy of the summary and enabling the generation of summaries of multiple news articles in a more optimal order without including incorrect information.

[0201] The deletion target detection process and weighting process in the preprocessing unit 32 can be performed using instruction information according to information on the attributes (for example, a combination of gender and age) of the user U to whom the news article summaries are provided. Also, the selection process and sorting process in the selecting and sorting unit 33 can be performed using instruction information according to information on the attributes (for example, a combination of gender and age) of the user U to whom the news article summaries are provided.

[0202] Furthermore, the summarization process in the summarization unit 34 can be performed using instruction information according to information on the attributes (for example, a combination of gender and age) of the user U to whom the summaries of the news articles are provided. For example, the summarization unit 34 can summarize the news articles using instruction information including information specifying the number of summaries and the summarization method according to the attributes of the user U to whom the summaries of the news articles are provided.

[0203] In addition, in the selection process in the selection and sorting unit 33, news articles can also be selected using instruction information including information indicating an instruction to preferentially select news articles that include trend words obtained from an external device.

[0204] In addition, in the sorting process in the selection and sorting unit 33, news articles can also be selected using instruction information including information indicating an instruction to prioritize news articles containing trend words obtained from an external device.

[0205] The processing unit 12 can also process summaries of news articles in real time based on the information of the news article group. For example, the processing unit 12 can process summaries of news articles in real time based on the information of the news article group using instruction information based on the context of the user U to whom the summaries of the news articles are provided. The processing unit 12 can perform processing using instruction information based on the context of the user U as instruction information for one or more of a deletion target detection process, a weighting process, a selection process, a sorting process, and a summarization process.

[0206] The generation AI may be a multimodal generation AI that includes the functionality of a text generation AI. The multimodal generation AI is, for example, a generation AI that can generate text and images from text and images. Examples of the multimodal generation AI include, but are not limited to, GPT-4 Turbo with vision, Gemini, and CM3Leon (Chameleon Multimodal Model).

[0207] The information processing device 1 uses the same generation AI for the duplicate detection process, first detection process, second detection process, weighting process, selection process, sorting process, and summarization process, but it is also possible to use different generation AIs for some or all of these processes.

[0208] Also, in the above example, the information of the group of news articles input into the generation AI is not the main text of the news articles but snippets of the main text of the news articles, but this is not limited to such an example, and the information of the group of news articles input into the generation AI may include the main text of the news articles instead of snippets of the main text of the news articles, and may also include images of the news articles.

[0209] [6. Hardware Configuration] The information processing device 1 according to the embodiment described above is realized by, for example, a computer 80 configured as shown in Fig. 9. Fig. 9 is a hardware configuration diagram showing an example of the computer 80 that realizes the functions of the information processing device 1 according to the embodiment. The computer 80 has a CPU 81, a RAM 82, a ROM (Read Only Memory) 83, an HDD (Hard Disk Drive) 84, a communication interface (I / F) 85, an input / output interface (I / F) 86, and a media interface (I / F) 87.

[0210] The CPU 81 operates and controls each part based on programs stored in the ROM 83 or the HDD 84. The ROM 83 stores a boot program executed by the CPU 81 when the computer 80 starts up, programs that depend on the hardware of the computer 80, and the like.

[0211] The HDD 84 stores programs executed by the CPU 81, data used by such programs, etc. The communication interface 85 receives data from other devices via the network N (see FIG. 2) and sends it to the CPU 81, and transmits data generated by the CPU 81 to other devices via the network N.

[0212] The CPU 81 controls output devices such as a display and a printer, and input devices such as a keyboard or a mouse, via the input / output interface 86. The CPU 81 acquires data from the input devices via the input / output interface 86. The CPU 81 also outputs generated data to the output devices via the input / output interface 86.

[0213] The media interface 87 reads a program or data stored in a recording medium 88 and provides it to the CPU 81 via the RAM 82. The CPU 81 loads the program or data from the recording medium 88 onto the RAM 82 via the media interface 87 and executes the loaded program. The recording medium 88 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase Change Rewritable Disc), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

[0214] For example, when the computer 80 functions as the information processing device 1 according to the embodiment, the CPU 81 of the computer 80 executes programs loaded onto the RAM 82 to realize the functions of the processing unit 12. In addition, the HDD 84 stores data in the storage unit 11. The CPU 81 of the computer 80 reads and executes these programs from a recording medium 88, but as another example, the CPU 81 may obtain these programs from another device via the network N.

[0215] [7. Other] Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using known methods. In addition, the information including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

[0216] Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0217] For example, the information processing device 1 described above may be realized by a terminal device and a server computer, or may be realized by multiple server computers. Furthermore, depending on the function, the configuration can be flexibly changed, such as by calling an external platform using an API or network computing.

[0218] Furthermore, the above-described embodiments and modifications can be combined as appropriate within the scope of not causing any contradiction in the processing content.

[0219] [8. Effects] As described above, the information processing device 1 according to the embodiment includes a selection and sorting unit 33 that causes a generation AI to select and sort multiple news articles using instruction information (an example of first instruction information) including information instructing the selection of multiple news articles and information indicating selection constraints, and instruction information (an example of second instruction information) including information instructing the rearrangement of multiple news articles and information indicating rearrangement constraints, and a summarization unit 34 that causes the generation AI to summarize each of the multiple news articles using instruction information (an example of third instruction information) including information instructing the generation AI to summarize each of the multiple news articles selected and rearranged by the selection and sorting unit 33. This allows the information processing device 1 to appropriately narrow down the news articles to be summarized compared to summarizing all news articles in a group of news articles to be distributed, thereby enabling more effective summarization of news articles. Furthermore, because the information processing device 1 rearranges the news articles before summarization, it is able to appropriately summarize the news articles.

[0220] Furthermore, the selecting and sorting unit 33 uses the generation AI to select multiple news articles, and then sorts the multiple news articles using the generation AI, thereby enabling the information processing device 1 to appropriately summarize the news articles.

[0221] The information processing device 1 also includes a preprocessing unit 32 that uses a generation AI to perform preprocessing on a group of news articles from which multiple news articles are selected by the selection and sorting unit 33. This allows the information processing device 1 to appropriately select news articles.

[0222] The preprocessing unit 32 also includes a duplication detection unit 40 that uses a generation AI to perform a process of detecting duplication of news articles using instruction information including an instruction to detect duplication of news articles in a group of news articles, thereby enabling the information processing device 1 to appropriately select news articles.

[0223] The preprocessing unit 32 also includes a deletion target detection unit 41 that uses instruction information including an instruction to detect specific news articles to be deleted from a group of news articles, and performs a process of detecting specific news articles using a generation AI. This allows the information processing device 1 to appropriately select news articles.

[0224] The preprocessing unit 32 also includes a weighting unit 42 that uses a generation AI to perform weighting processing of news articles using instruction information including an instruction to weight news articles in a group of news articles, thereby enabling the information processing device 1 to appropriately select news articles.

[0225] Furthermore, the weighting unit 42 causes the generation AI to output the weights of the news articles in the group of news articles, and then rearranges the news articles in the group of news articles based on the weights, thereby enabling the information processing device 1 to appropriately select news articles.

[0226] The above describes the embodiments of the present application in detail based on the drawings, but this is merely an example, and the present invention can be implemented in other forms that include the embodiments described in the Disclosure of the Invention section and that have been modified and improved in various ways based on the knowledge of those skilled in the art.

[0227] Furthermore, the above-mentioned "section, module, unit" can be read as "means" or "circuit," etc. For example, an acquisition unit can be read as an acquisition means or an acquisition circuit. [Explanation of symbols]

[0228] 1. Information processing equipment 2. Terminal Device 3. Uploader's device 10. Communications Department 11 Storage section 12 Processing section 20 News article group information storage unit 21 Summary information storage unit 30 Acquisition Department 31 Reception 32 Pre-processing section 33 Selection sorting section 34 Summary 35 Distribution Department 36 Sorting section 40 Duplicate detection unit 41 Deletion target detection unit 42 Weighting section 45 Selection section 50 First detection unit 51 Second detection unit 100 Information Processing Systems N Network

Claims

1. a selection and rearrangement unit that causes a generation AI to select and rearrange the plurality of news articles using first instruction information that includes information instructing the selection of a plurality of news articles and information indicating constraints on the selection, and second instruction information that includes information instructing the rearrangement of the plurality of news articles and information indicating constraints on the rearrangement; a summarization unit that causes a generation AI to summarize each of the plurality of news articles using third instruction information that includes information instructing a summary of each of the plurality of news articles selected and sorted by the selection and sorting unit.

1. An information processing device comprising:

2. The selection and sorting unit After selecting the plurality of news articles using a generation AI, sorting the plurality of news articles using the generation AI.

2. The information processing apparatus according to claim 1, wherein:

3. a preprocessing unit that performs preprocessing on a group of news articles from which the plurality of news articles are selected by the selection and sorting unit, using a generation AI; 2. The information processing apparatus according to claim 1, wherein:

4. The pre-treatment unit The news article group includes a duplication detection unit that uses instruction information including an instruction to detect duplication of news articles in the news article group to perform a process of detecting duplication of the news articles using a generation AI.

4. The information processing apparatus according to claim 3,

5. The pre-treatment unit The system includes a deletion target detection unit that uses a generation AI to perform a process of detecting a specific news article to be deleted from the group of news articles by using instruction information including an instruction to detect the specific news article to be deleted from the group of news articles.

5. The information processing apparatus according to claim 3, wherein the information processing apparatus is a computer.

6. The pre-treatment unit A weighting unit performs a weighting process of the news articles using a generation AI by using instruction information including an instruction to weight the news articles in the group of news articles.

5. The information processing apparatus according to claim 3, wherein the information processing apparatus is a computer.

7. The pre-treatment unit a duplication detection unit that uses a generation AI to perform a process of detecting duplication of news articles using instruction information including an instruction to detect duplication of news articles in the group of news articles; a deletion target detection unit that uses a generation AI to perform a process of detecting a specific news article to be deleted from the group of news articles using instruction information including an instruction to detect the specific news article to be deleted; a weighting unit that uses a generation AI to perform a weighting process on the news articles using instruction information including an instruction to weight the news articles in the group of news articles.

4. The information processing apparatus according to claim 3,

8. The weighting unit After outputting the weights of the news articles in the group of news articles to a generation AI, the news articles in the group of news articles are sorted based on the weights.

8. The information processing apparatus according to claim 7,

9. 1. A computer-implemented information processing method, comprising: a selection and rearrangement process for causing a generation AI to select and rearrange the plurality of news articles using first instruction information including information instructing the selection of a plurality of news articles and information indicating constraints on the selection, and second instruction information including information instructing the rearrangement of the plurality of news articles and information indicating constraints on the rearrangement; and a summarizing step of causing a generating AI to generate a summary of each of the plurality of news articles using third instruction information including information instructing a summary of each of the plurality of news articles selected and sorted by the selecting and sorting step.

1. An information processing method comprising:

10. a selection and rearrangement procedure for causing a generation AI to select and rearrange the plurality of news articles using first instruction information including information instructing the selection of a plurality of news articles and information indicating constraints on the selection, and second instruction information including information instructing the rearrangement of the plurality of news articles and information indicating constraints on the rearrangement; a summarization procedure for causing a generation AI to generate summaries of each of the plurality of news articles using third instruction information including information instructing a summary of each of the plurality of news articles selected and sorted by the selection and sorting procedure. An information processing program characterized by:

Images