System, method, and program for generating and evaluating news releases

Abstract

Easily generate, evaluate, and score news releases with richly expressive sentences [Solution] The present invention provides a system for generating a news release, the system comprising: a receiving means for receiving input information including a topic for the news release; an acquiring means for acquiring a plurality of expressions based on the input information, the plurality of expressions being included in words generated by a plurality of language models; and a generating means for generating at least one phrase for the news release using at least one of the plurality of expressions. The system of the present invention may further comprise an evaluating means for evaluating the generated news release.

Description

[Technical Field]

[0001] The present invention relates to a system, method, and program for generating, evaluating, and scoring news releases. [Background technology]

[0002] Technologies for generating sentences using language models have been developed (Patent Document 1). One example is ChatGPT by OpenAI. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2024 / 236619 Summary of the Invention [Problem to be solved by the invention]

[0004] The inventors of the present invention believed that the summary text generated by ChatGPT and other tools was not suitable for news releases. The inventors of the present invention believed that news releases required text that was rich in expression, as if written by a human. They also believed that news releases needed to be made more effective by appropriately evaluating and scoring the generated news releases (for example, by evaluating and scoring the likelihood of the news release being picked up by the mass media).

[0005] The present invention aims to easily generate news releases using sentences with rich expressions, and also aims to evaluate and score the generated news releases. [Means for solving the problem]

[0006] The present invention utilizes a plurality of language models to generate at least one phrase for the news release from the plurality of captured expressions. The present invention also utilizes the plurality of language models to evaluate and score the generated news release. The present invention provides, for example, the following items. (Item 1) 1. A system for generating a news release, comprising: receiving means for receiving input information including topics for a news release; an acquiring means for acquiring a plurality of expressions based on the input information, the plurality of expressions being included in words generated by a plurality of language models; generating means for generating at least one phrase for a news release using at least one expression from the plurality of expressions; A system comprising: (Item 2) the generating means further generates a news release including the generated at least one phrase; The system comprises: The system described in the preceding item further comprises an evaluation means for evaluating the generated news release. (Item 3) The system of any of the preceding items, wherein the evaluation means evaluates the generated news release by determining the likelihood of the generated news release being covered by mass media. (Item 4) The evaluation means evaluates expressions included in the generated news release by: determining whether the expression satisfies criteria for a plurality of items, thereby calculating a score for each of the plurality of items; determining whether the total score obtained by adding up the scores of each item exceeds a threshold; 2. The system according to claim 1, further comprising: (Item 5) A system described in any of the above items, wherein the plurality of items include at least one of the following: the topicality of the news release, the appeal of the news release from a consumer perspective, the data and performance evidence contained in the news release, the appeal of the introductory part of the news release, the market impact of the news release, or the uniqueness of the news release. (Item 6) The system according to any of the preceding items, wherein the acquisition means instructs the plurality of language models to engage in casual conversation, and the plurality of expressions include expressions used in casual conversation between the plurality of language models. (Item 7) The system described in any of the preceding items, wherein the acquisition means instructs at least one of the plurality of language models to generate a diary or journal, and the plurality of expressions include expressions in the diary or journal. (Item 8) The system according to any of the preceding items, wherein the plurality of expressions include onomatopoeia or mimetic words, or expressions in hiragana. (Item 9) The system of any of the preceding items, wherein the acquisition means instructs at least one of the plurality of language models to generate a poem or a verse, and the expression includes an expression in a poem or a verse. (Item 10) The system according to any of the preceding items, wherein the generating means generates a title of the news using at least one expression from the plurality of expressions. (Item 11) The system according to any one of the preceding items, wherein the generation means generates a summary of the news using at least one expression from the plurality of expressions. (Item 12) The system according to any one of the preceding items, wherein the generating means extracts, from the plurality of expressions, an expression that expresses emotion such as excitement, amazement, empathy, or doubt as the at least one expression. (Item 13) 1. A method for generating a news release, comprising: receiving input regarding a particular news item; obtaining a plurality of expressions based on the input, the plurality of expressions being included in vocabulary generated by a plurality of language models; generating text for a news release using at least one expression from the plurality of expressions; A method comprising: (Item 13A) Item 14. The method according to item 13, comprising the features according to any one of the preceding items. (Item 14) 1. A program for generating a news release, the program running on a computer having a processor, the program comprising: receiving input regarding a particular news item; obtaining a plurality of expressions based on the input, the plurality of expressions being included in vocabulary generated by a plurality of language models; generating text for a news release using at least one expression from the plurality of expressions; A program causing the processor to perform processing including the steps of: (Item 14A) Item 15. A program according to item 14, comprising the features according to any one of the preceding items. (Item 14B) A computer-readable storage medium storing the program according to item 14 or item 14A. [Effects of the Invention]

[0007] The present invention provides a system that uses multiple language models to generate news releases with rich, expressive sentences that appear as if they were written by a human. The system can also evaluate and score the likelihood of news releases being picked up by the mass media, enabling the generation of effective news releases. The present invention can bring about improvements in fields such as advertising and journalism. [Brief explanation of the drawings]

[0008] [Figure 1] A diagram showing an example of the flow of a new service that allows for the easy generation of news releases. [Figure 2] FIG. 1 shows an example of the configuration of a system 100 for generating news releases. [Figure 3] FIG. 1 is a diagram showing an example of a specific configuration of a system 100 for generating news releases. [Figure 4A] FIG. 4A is a diagram showing an example of the configuration of the processor unit 120. [Figure 4B] FIG. 10 is a diagram showing an example of the configuration of a processor unit 120 according to another embodiment. [Figure 5] Data flow of processing in system 100 for generating news releases DETAILED DESCRIPTION OF THE INVENTION

[0009] (definition) As used herein, the term "news release" refers to information provided to the mass media regarding a particular topic, and in particular to a written summary of a particular topic. A news release may be provided or distributed as part of a public relations campaign. For example, a company may provide or distribute a news release to advertise its new product in order to be featured in the mass media (e.g., newspapers, television programs, internet news, etc.).

[0010] As used herein, a "language model" refers to a model that has learned a language dataset and is capable of performing language processing. In particular, a language model can generate words (e.g., words, phrases, clauses, sentences, paragraphs, etc.). Language processing using a language model includes the generation of phrases.

[0011] As used herein, a "phrase" refers to one or more words or symbols, or a combination of multiple words or symbols. For example, a "phrase" may be a single word, a single clause, a single sentence, or a single paragraph.

[0012] In this specification, a "sentence" refers to a sequence of one or more sentences, and is included in the above-mentioned "phrase."

[0013] As used herein, "expression" refers to something that is expressed in a way that can be perceived. An "expression" may typically represent the internal state of a person, such as psychological, emotional, or spiritual. An "expression" may be, for example, a word, phrase, sentence, or paragraph in casual conversation, a word, phrase, sentence, or paragraph in a diary or journal, or a word, phrase, sentence, or paragraph in a poem or poem.

[0014] In this specification, "chat" refers to a conversation between people (including virtual people, personas), and the conversation may be expressed in spoken language.

[0015] (1. News release generation service) The inventors of the present invention have developed a new service that allows users to easily generate news releases. This service uses AI that utilizes language models (so-called generative AI) to provide phrases for news releases that sound as if they were written by a human copywriter. Users can complete their news releases using the provided phrases, or they can select phrases they want to use, and the service outputs a news release containing the selected phrases.

[0016] So-called generative AI excels at generating summaries. When so-called generative AI is used to generate phrases for news releases, it only outputs summary expressions and is unable to provide phrases that sound as if they were created by a human copywriter. After extensive research, the inventors of the present invention discovered a system for providing phrases that sound as if they were created by a human copywriter. This system uses multiple language models (generative AI) to generate words that are not directly related to news releases, and then generates phrases for news releases from the generated words. Words that are not directly related to news releases include, for example, casual conversation, diary entries, and poetry. The phrases generated from these words sound as if they were created by a human copywriter and include phrases that can be used in news releases. This was unexpected.

[0017] Furthermore, this service also evaluates and scores the generated phrases to determine whether they are likely to be picked up by the mass media. News releases containing phrases with good evaluation results are likely to be picked up by the mass media and are therefore effective news releases. Users can refer to the scoring or evaluation results for each generated phrase to complete their news releases.

[0018] Figure 1 shows an example of the flow of a new service that allows for the easy generation of news releases.

[0019] 1, it will be explained that a user U uses this service to obtain a news release R. This service is provided using a system 100.

[0020] In step S1, a user U provides input information. The input information includes at least a topic for the news release, i.e., an outline of the subject for which the user wants to generate a news release. The input information may also include content or points that the user wants to explicitly emphasize in the news release (e.g., the implementation of a campaign, etc.).

[0021] For example, the user U can provide input information to the system 100 to generate a news release about a new product, including the name of the new product, the features of the new product, what the user most wants to communicate, and the like.

[0022] The input information may include a public relations framework. The public relations framework specifies expressions to be included in phrases for a news release or perspectives to be considered, and may be related to IMPAKT, for example. IMPAKT refers to six elements: Inverse, Most, Public, Actor, Keyword, and Trend. Inverse may include paradoxical expressions or expressions of opposing structures, such as "XXX is ending its low prices" or "Hybrid cars vs. electric cars!" Most may include the largest or first expressions, such as "The world's largest XXX" or "Japan's first XXX." Public may include social or regional expressions, such as "Regional revitalization" or "Opening XXX in response to an aging society." Actor may include actor-like or humanistic expressions, such as "The president himself cleans customer toilets" or "XXX robot gives special lecture at elementary school." Keyword may include keywords or numerical expressions, such as "XXX activity" or "the third XXX." Trend may include expressions of current events, social conditions, or seasonality, such as "Football XX Cup," "Christmas," and "Hay fever."

[0023] When the input information is provided, the system 100 performs processing to generate a news release R. The system 100 can use multiple generation AIs to generate at least one phrase for the news release R. To this end, the system 100 inputs the input information to the multiple generation AIs. At this time, instructions (so-called prompts) for causing the multiple generation AIs to generate words can also be input.

[0024] The instruction to make the multiple generation AIs generate words is, for example, an instruction to make the multiple generation AIs engage in casual conversation. In accordance with this instruction, the multiple generation AIs can engage in casual conversation and output expressions C in the casual conversation. For example, each of the multiple generation AIs becomes a virtual person (a so-called persona), and the virtual people converse with each other based on the input information provided. It is preferable that expressions containing impressions or emotions or subjective expressions appear in the casual conversation. It is also preferable that topics related to the challenges that the virtual people face on a daily basis appear in the casual conversation. The conversational exchanges between the virtual people can include, for example, 10 or more exchanges, and preferably 40 or more exchanges.

[0025] The prompt may indicate the personality or setting of the virtual person, including, but not limited to, gender, age, occupation or position, etc. For example, multiple virtual people may have different personalities or settings, or may have the same or similar personalities or settings.

[0026] For example, instead of or in addition to chatting, a diary may be generated.

[0027] The instruction to have the multiple generation AIs generate words is, for example, an instruction to have at least one of the multiple generation AIs generate a diary. In accordance with this instruction, at least one of the multiple generation AIs generates a diary and can generate an expression D in the diary. The diary is, for example, a diary written by a virtual person about the provided input information. The virtual person (so-called persona) can be, for example, a businessman, an elementary school student, etc., and this persona can be arbitrarily set by the user U via a prompt.

[0028] For example, a diary entry by a fictitious businessman may describe his experiences from waking up, commuting to work, working at the office, coming home, and going to bed. The businessman's diary entry may include, for example, pretentious, affectational, or formal language, or may be written in idiomatic or proverbial terms.

[0029] For example, a diary written by a fictitious elementary school student may be a diary that describes the experiences of waking up, going to school, studying at the elementary school, coming home from school, playing, and going to bed. A diary written by an elementary school student may include, for example, childish expressions, soft expressions, humorous expressions, expressions using onomatopoeia or mimetic words, or expressions using hiragana.

[0030] For example, instead of or in addition to a diary, a poem may be created.

[0031] The instruction to have the multiple generation AIs generate words is, for example, an instruction to have at least one of the multiple generation AIs generate a poem. In accordance with this instruction, at least one of the multiple generation AIs can generate a poem and generate an expression P in the poem.

[0032] The poem is, for example, a diary written by a fictitious person about the input information provided. The fictitious person (so-called persona) can be, for example, a professional poet, an amateur poet, a businessman, a university student, an elementary school student, etc., and this persona can be arbitrarily set by the user U via prompts.

[0033] Poetry may include, for example, literary devices such as metaphors, scene descriptions, psychological descriptions, onomatopoeia or mimetic language, or techniques such as nominative sentences, inversion, or repetition.

[0034] System 100 extracts at least one expression from multiple expressions contained in chats, diaries, poems, etc. generated by multiple generation AIs, and generates a phrase for a news release. The phrase is preferably one that can interest mass media recipients (e.g., readers, viewers).

[0035] Next, the system 100 generates a news release R that includes the generated phrase. The generated phrase can form, for example, a title, an abstract, a lead sentence, or a body sentence in the news release R. At this time, the user U may select a phrase that he or she likes from among the generated phrases, and the system 100 may generate a news release R that includes the selected phrase. Alternatively, the system 100 may automatically select a specific phrase from among the generated phrases, and generate a news release R that includes the selected phrase.

[0036] Once the news release R is generated by the system 100, in step S2 the news release R is output from the system 100. The news release R can be provided electronically from the system 100 to the user U's terminal device, for example.

[0037] System 100 may evaluate the likelihood of generated news release R being picked up by the mass media (e.g., likelihood of being published in a newspaper) and output the news release R together with the evaluation or rating results. This allows user U to consider whether to adopt the generated news release R as an actual news release, taking into account the evaluation or rating results.

[0038] The above-described system 100 can be implemented, for example, as a system 100 for generating news releases having the configuration described below.

[0039] (2. System configuration for generating news releases) FIG. 2 shows an example configuration of a system 100 for generating news releases.

[0040] The system 100 is connected to a database 200. The system 100 is also connected to at least one user terminal device 300 via a network N.

[0041] 2 shows three user terminal devices 300, the number of user terminal devices 300 is not limited to this. Any number of user terminal devices 300 may be connected to the system 100 via the network N.

[0042] The network N may be any type of network. For example, the network N may be the Internet or a LAN. The network N may be a wired network or a wireless network.

[0043] An example of the system 100 may be a computer (e.g., a server device) installed at a provider that provides a service for generating news releases. An example of the user terminal device 300 is a computer (e.g., a terminal device) used by a user who uses the service, but is not limited to this. The user may be, for example, a company's public relations officer. Here, the computer (server device or terminal device) may be any type of computer. For example, the terminal device may be any type of terminal device, such as a smartphone, tablet, personal computer, smart glasses, or smart watch.

[0044] Database 200 may store, for example, phrases generated by system 100 or generated news releases. The generated news releases may be stored along with a score or rating for likelihood of media coverage, for example.

[0045] FIG. 3 shows an example of a specific configuration of a system 100 for generating a news release.

[0046] The system 100 comprises an interface section 110, a processor section 120, and a memory section .

[0047] The interface unit 110 exchanges information with the outside of the system 100. The processor unit 120 of the system 100 can receive information from the outside of the system 100 and can send information to the outside of the system 100 via the interface unit 110. The interface unit 110 can exchange information in any format.

[0048] The interface unit 110 includes, for example, an input unit that allows information to be input to the system 100. It does not matter how the input unit allows information to be input to the system 100. For example, if the input unit is a receiver, the receiver may input information by receiving information from outside the system 100 via a network. Alternatively, if the input unit is a data reading device, the input unit may input information by reading information from a storage medium connected to the system 100.

[0049] The interface unit 110 includes, for example, an output unit that enables information to be output from the system 100. It does not matter in what manner the output unit enables information to be output from the system 100. For example, if the output unit is a transmitter, the transmitter may output information by transmitting it to an external device outside the system 100 via a network. Alternatively, if the output unit is a data writing device, the output unit may output information by writing it to a storage medium connected to the system 100.

[0050] The system 100 can, for example, transmit information to and / or receive information from the database 200 via the interface unit 110. The system 100 can, for example, transmit information to and / or receive information from the user terminal device 300 via the interface unit 110.

[0051] The system 100 can receive input information for generating a news release, for example, via the interface unit 110. As will be described later, the input information for generating a news release can be a topic for the news release. In addition to the topic, the input information can also include content or points that the user wants to explicitly emphasize in the news release (e.g., the implementation of a campaign, etc.). Furthermore, the input information can also include a framework for the public relations.

[0052] For example, the system 100 may output at least one phrase for the news release via the interface unit 110. Alternatively, the system 100 may also output an evaluation value of the at least one phrase for the news release. The system 100 may also output the generated news release via the interface unit 110.

[0053] The processor unit 120 executes the processing of the system 100 and controls the overall operation of the system 100. The processor unit 120 reads and executes a program stored in the memory unit 130. This allows the system 100 to function as a system that executes desired steps. The processor unit 120 may be implemented by a single processor or by multiple processors.

[0054] The memory unit 130 stores programs required to execute the processing of the system 100, data required to execute the programs, and the like. The memory unit 130 may also store a program for causing the processor unit 120 to execute processing to generate a news release (e.g., a program that implements the processing shown in FIG. 5, which will be described later). Here, how the program is stored in the memory unit 130 is not important. For example, the program may be pre-installed in the memory unit 130. Alternatively, the program may be stored in a non-transitory computer-readable storage medium and installed by reading the storage medium. Alternatively, the program may be installed in the memory unit 130 by being downloaded via a network. In this case, the type of network is not important. The memory unit 130 may be implemented by any storage means.

[0055] Database 200 may store, for example, phrases generated by system 100 or generated news releases. The generated news releases may be stored along with a score or rating for likelihood of media coverage, for example.

[0056] In the examples shown in FIGS. 2 and 3 , the database 200 is provided outside the system 100, but the present invention is not limited to this. At least a portion of the database 200 can also be provided inside the system 100. In this case, at least a portion of the database 200 may be implemented by the same storage means as the storage means that implements the memory unit 130, or by a storage means different from the storage means that implements the memory unit 130. In either case, at least a portion of the database 200 is configured as a storage unit for the system 100. The configuration of the database 200 is not limited to a specific hardware configuration. For example, the database 200 may be configured as a single hardware component or multiple hardware components. For example, the database 200 may be configured as an external hard disk drive attached to the system 100, as cloud storage connected via a network, or as a distributed network using blockchain technology or the like.

[0057] FIG. 4A shows an example of the configuration of the processor unit 120.

[0058] The processor unit 120 includes a receiving unit 121 , an acquiring unit 122 , and a generating unit 123 .

[0059] The receiving means 121 is configured to receive input information. The input information includes at least a topic for a news release. The topic for a news release may be an outline of a matter for which a news release is desired to be generated.

[0060] In addition to the topic, the input information may further include content or points that you want to explicitly emphasize in the news release (e.g., the implementation of a campaign, etc.). Furthermore, the input information may also include a framework for the public relations.

[0061] The input information received by the receiving means 121 is passed to the acquiring means 122 .

[0062] The acquiring means 122 is configured to acquire a plurality of expressions based on input information. The acquiring means 122 can acquire a plurality of expressions included in words generated by a plurality of language models. Here, the plurality of language models may be language models provided in the system 100 or may be language models provided externally to the system 100. The language models may, for example, constitute a so-called generative AI.

[0063] Multiple language models can interact to generate expressive phrases that sound as if they were created by a human copywriter and can be used in news releases.

[0064] For example, the plurality of language models can chat with each other and output expressions used in the chat. To this end, the acquiring unit 122 can instruct the plurality of language models to chat with each other.

[0065] Each of the multiple language models can impersonate a virtual person and engage in casual conversation. The language model impersonating a virtual person can converse using words that sound as if they were spoken by that person. The personality or setting of the virtual person to be impersonated may be determined in advance or may be set by the user. For example, when the user inputs the personality or setting of a virtual person into the system 100 via the interface unit 110, the personality or setting may be indicated to the language model via a prompt or the like.

[0066] The language models may be divided into a plurality of groups, and chat may be conducted within each group. Preferably, each group contains five or less language models, and more preferably, three language models.

[0067] Note that casual conversation is just one example, and the conversation is not limited to casual conversation as long as it is a form of speech in which expressions containing impressions or emotions or subjective expressions can appear.

[0068] The plurality of language models may further generate a diary or a journal and output expressions in the diary or journal. To this end, the obtaining means 122 may instruct at least one of the plurality of language models to generate a diary or a journal.

[0069] At least one of the plurality of language models can impersonate a virtual person and generate a diary or journal. The language model impersonating the virtual person can generate a diary or journal in words that appear as if the person had written them. The personality or setting of the virtual person to be impersonated may be predetermined or may be set by the user.

[0070] Note that a diary or journal is just one example, and the document is not limited to a diary or journal as long as it is written language in which subjective expressions of facts can appear.

[0071] The plurality of language models may further generate a poem or a verse and output the expression in the poem or a verse. To this end, the obtaining means 122 may instruct at least one of the plurality of language models to generate a poem or a verse.

[0072] At least one of the multiple language models can impersonate a virtual person and generate a poem or verse. The language model impersonating the virtual person can generate a poem or verse in words that sound as if the virtual person had written it. The personality or setting of the virtual person to be impersonated may be predetermined or may be set by the user.

[0073] Note that poetry is merely an example, and the text is not limited to poetry, but may be any written language in which literary expressions can appear, such as haiku, tanka, or song.

[0074] The acquiring means 122 may acquire words generated by a plurality of language models as a plurality of expressions as they are, or may acquire parts of words generated by a plurality of language models as a plurality of expressions. When acquiring parts of words, the acquiring means 122 may selectively acquire them, or may acquire them in response to an input by the user.

[0075] The representations obtained from the multiple language models are passed to a generating means 123 .

[0076] The generating means 123 is configured to generate at least one phrase for the news release using at least one expression from the retrieved plurality of expressions.

[0077] The generating means 123 extracts, for example, expressions that can be used as phrases in a news release from among a plurality of expressions. The expressions that can be used as phrases in a news release can be, for example, expressions that express emotions, and the emotions can be emotions related to excitement, amazement, sympathy, or doubt. The expressions that can be used as phrases in a news release can be, for example, expressions that conform to a public relations framework.

[0078] The generating unit 123 can extract expressions that can be used as phrases in news releases, for example, by using a language model (LM) or a large-scale language model (LLM). For example, when multiple expressions are input into the LM or LLM along with a prompt to extract expressions that can be used as phrases in news releases, expressions that can be used as phrases can be extracted. The number of expressions extracted can be varied depending on the accuracy.

[0079] The generation means 123 uses the extracted expressions to generate at least one phrase for the news release. For example, the generation means 123 may use the extracted expressions as they are as a phrase. Alternatively, the generation means 123 may combine some of the extracted expressions to generate a phrase. For example, the extracted expressions may be presented to the user via the interface unit 110, and one expression selected by the user may be used as a phrase as it is, or a phrase may be generated by combining multiple selected expressions.

[0080] The phrases generated in this way can sound as if they were written by a human copywriter, and the news releases generated using these phrases can also sound as if they were written by a human copywriter. System 100 makes it possible to easily generate news releases with rich expressions without relying on human copywriters.

[0081] For example, a user may create a news release using a phrase output from the system 100, or the news release may be generated by the system 100. In this case, the generating means 123 may also generate a news release including the generated phrase. The generating means 123 may generate a news release including the generated phrase using, for example, LM or LLM. The generated phrase may form, for example, a title, an abstract, a lead sentence, or a body sentence in the news release.

[0082] As an example, the results are shown below when three persona images (Misaki (female), Kenta (male), and Yui (female)) created by a language model are engaged in casual conversation. In this example, the topic entered is "The audiobook version of 'Twelve Kingdoms'."

[0083] Misaki: Listen up everyone! An audiobook version of "Twelve Kingdoms" is coming out! Kenta: Really? Speaking of "Twelve Kingdoms," I was really into it back in the day. Yui: Wow, that's amazing! I love this work too. I wonder what it will be like? Misaki: It will apparently start streaming in mid-June. The first volume will be "Shadow of the Moon, Shadow of the Sea," and it will be read by none other than Hisakawa Aya! Kenta: Oh, isn't that the girl who played Yoko in the anime version? It's so nostalgic. Yui: It's so luxurious! I love it when voice actors read aloud because it really draws you into the world of the work. Misaki: That's right! And apparently Shoya Ishige will be in charge of "The Devil's Child." Kenta: Ishige-san has been appearing in a lot of anime recently, right? He seems like a young up-and-coming talent. Yui: Yes, yes, his voice has a unique atmosphere that I like. It might be perfect for "Mashou no Ko." Misaki: Hey, it looks like there's a sample video too. Wanna take a listen? Kenta: Oh, I'd love to hear that! I wonder what it's like. Yui: I'm excited! I can't wait to hear it. (Listen to sample video) Misaki: Wow... Hisakawa's voice is nostalgic, but also feels fresh. It's really wonderful. Kenta: Yeah, it's really good. The intonation of his voice, the timing, it really makes you feel like a pro. Yui: It really feels like the story is unfolding right before my eyes. It makes my heart beat faster. Misaki: Hey, it seems there's also a giveaway for autographed autographed cards. Why don't you enter? Kenta: Wow, that sounds appealing. But how do I apply? Yui: I'm curious too! Tell me more, Misaki. Misaki: Well, first you create an account on audiobook.jp, then enter your email address in the Google form to apply. Kenta: Oh, I see. So? Misaki: If you purchase a "Twelve Kingdoms" album when it is released in June or listen to it on the unlimited listening plan, you'll be entered into the lottery. Yui: Wow, that's great! I'll definitely apply! Kenta: That's great, I might join too. But creating an account seems like a pain... Misaki: Don't worry, I don't think it's that difficult. Plus, if you create an account, you'll get news about the Twelve Kingdoms audiobook via email every month. Yui: That's convenient! That means we'll know the release date as soon as it's decided. Kenta: I see, then it sounds like it would be worth making. Misaki: Hey, how about we all listen to it together? Let's get together on the release day and listen over tea. Yui: That's a great idea! I totally agree! Kenta: Yeah, that's great. It'll be an excuse to get together after a long time. Misaki: Well, that's settled! I'm looking forward to it. Yui: By the way, Misaki-chan, you seem busy lately, is that okay? Misaki: Yeah... Actually, it's a little difficult to balance my school work and tutoring... Kenta: I see, it's a pretty hard schedule. Misaki: That's true. But audiobooks are great because I can read books while I'm on the move or doing housework. Yui: I know what you mean! I listen to it while I'm writing. It helps me stay focused and gives me inspiration. Kenta: Wow, so that's one way to use it. I listen to it on my commute, and it definitely feels like I'm making good use of my time. Misaki: That's right! It'll be even more fulfilling if I can listen to "Twelve Kingdoms" as well. Yui: Hey, don't you think that "Twelve Kingdoms" has some relevance to real-world issues? Kenta: What do you mean? Yui: For example, the difficulties Yoko faces in a different world, or what leadership is like... Misaki: Ah, I know! There are so many situations that could be used to teach children a lesson. Kenta: I see. I've just become a team leader at work, so this might be helpful. Yui: That's right! It's full of wisdom that we can apply to our daily lives. Misaki: I'm really looking forward to listening to it again. Kenta: By the way, are there any other audiobooks you'd like to listen to? Yui: Hmm, I think it would be Murakami Haruki's works. I think you can discover new charms in his writing when you listen to it out loud. Misaki: That's great! I'd probably choose an educational book. It would be more efficient to study while I'm on the move. Kenta: I'd like to listen to sci-fi or fantasy. It seems like it would be really immersive. Yui: The great thing about audiobooks is that everyone can enjoy them according to their own interests. Misaki: That's true, but I can't give up the benefits of paper books. Kenta: I understand. I also find it easier to read technical books on paper. Yui: I also find it easier to use a paper book when editing. Misaki: In the end, it might be best to take advantage of the best of both worlds. Kenta: That's true. It's wise to use them depending on the situation. Yui: Hey, I wonder if there will be a sequel to "Twelve Kingdoms"? Misaki: Oh, I'm looking forward to that too! It'd be great to hear a new story. Kenta: Seriously. If there's a sequel, we can all get excited about it again. Yui: Yes! Just thinking about it makes me excited. Misaki: That's right, how about a book club themed around the world of "Twelve Kingdoms"? Kenta: Oh, that's great. If we do it online, people far away can participate. Yui: Great idea! I'll help with the planning. Misaki: Yay! Then, let's start planning right away when the audiobook is released. Kenta: I'm looking forward to it. It's been a while since I've felt this excited. Yui: Me too! Thanks to "Twelve Kingdoms," I feel like I've found new things to enjoy. Misaki: I'm really happy to have come across such a wonderful piece of work. Kenta: Yeah, it's great to have friends to share these experiences with. Yui: That's right. Let's continue to have fun together! Misaki: Yes, I promise! I'm looking forward to what's to come.

[0084] Furthermore, an example of the results when a diary entry is generated using a language model is shown below. In this example, a diary entry is generated for a businessman persona with the topic "The audiobook version of 'Twelve Kingdoms'."

[0085] A businessman's daily diary Date: March 3, 2025 The alarm goes off in the morning. It's Monday, the start of the new week, and I get out of bed feeling a little gloomy. As I make coffee, I think about the flow of my work for the day. I've been working on the press release for the audiobook version of "Twelve Kingdoms," and excitement is building inside me. This is truly the arrival of a new era of listening. 8:00 AM - Commute During my commute, I listen to audiobooks of business books on the train as usual. According to recent research, the busier the business person, the more they rely on "listening to books." I am one of them. Learning by ear allows me to use my commute time meaningfully. 9:00 AM - Arrive at the office When I arrive at the office, my colleague is talking about the progress of the audiobook from last week. She is also looking forward to Aya Hisakawa's reading, and has high hopes for this industry-first attempt. We get excited as we discuss the new value that the fusion of voice acting and literature can bring. It stimulates our senses to think about the depth that an emotional reading can add to a story. 12:00 PM - Lunch During lunch, I received an email from a friend. "I've been so busy lately that I haven't had time to read, but audiobooks seem like something I could enjoy in my spare time." I couldn't agree more. It's clear to see that this method of efficiently gathering information while enjoying literature is becoming a popular new lifestyle. 3:00 PM - Presentation Preparation I'm currently preparing presentation materials for the afternoon meeting. I'm gathering data on our expansion strategy into the Asian market and the popularity of audiobooks. I plan to introduce last year's market growth rate and emphasize that the "depth of stories that are absorbed through the ears" has a major impact on business. 5:30 PM - Finish work After work, I leave the office and head home. I want to listen to a new audiobook tonight. The first volume will be released in mid-June, and I'm already looking forward to next month's release. The appeal of enjoying a story with your ears is that it's a different experience than usual. 6:30 PM - Arrive home After dinner, it's time to relax. I download the audiobook for "Shadow of the Moon, Shadow of the Sea." At first, the room is silent, but Aya Hisakawa's voice fills the air, drawing me into the world of the story. It's a new kind of emotion that can't be experienced visually, and it's an experience that makes you feel as if you're a part of the story. 9:00 PM - Getting ready for tomorrow It's late into the night, but the afterglow of the experience is still lingering. I organize the notes I took while listening and prepare for tomorrow's work. In the midst of busy days, this "listening and reading" nourishes the soul. I prepare for the next day, grateful to have been exposed to a new lifestyle. Today was another fulfilling day. I wonder what new experiences await me tomorrow.

[0086] Furthermore, below is an example of the results when the language model is used to generate poetry:

[0087] "The shadow of a flower blooming on a hazy window, The morning sky dyed with soft light, The distant hills where the bells of awakening ring, The heartbeat of the earth waking from its slumber. A sunshine arrow shines into my frozen heart. A rainbow of hope is reflected on the surface of the melting river. The branches have cast off their winter clothes and are clothed in new attire. Cherry blossoms dance on the wind, chasing the traces of fleeting dreams. The spirits of spring play the song of life. The chirps of birds break the silence. The scent of budding grass wafts along the path. The dance of young leaves swaying in the breeze. My heart, freed from the burden of winter, sheds tears of joy at the arrival of spring. Time goes by and the seasons begin to bud once again. The eternal cycle of life weaves a poem of life. Come, spring, messenger of healing, to the world awakening from its winter slumber.

[0088] Thus, even if the topic is given as "the audiobook version of 'Twelve Kingdoms'", the expressions contained in the generated words (i.e., chats, diaries, and poems) are not necessarily related to this topic. Rather, they contain many unrelated expressions. This allows the resulting phrases generated from these expressions to be expressive.

[0089] For example, the following phrases can be generated from the expressions contained in the chats, diaries, and poems mentioned above: In this example, these phrases were generated using LLM. (A) Drops of story that seep into your heart through your ears (i) Fly to another world on the wings of your voice (C) Fireworks of words that brighten up the silence (D) A soulful resonance that resonates across time (E) A garden in your ears that blooms in between busy schedules (K) A traveler walking the border between reality and fantasy (Ki) Invisible paintings woven by voices (H) A wind from another world creeping into everyday life (K) The infinite universe that spreads from the ear canal (K) A heart-warming painting woven with the magic of voice (S) Islands of words floating in a sea of silence (C) Awakening deep in the ears, sleeping sensitivity (S) A kaleidoscope in the air, painted by voices (C) Opening the door to hearing, liberating the imagination (So) The nectar of stories dripping from my ears (Ta) A symphony of the soul played by the vocal cords (C) A surprising treasure chest hidden in the labyrinth of the ear (Tsu) A ballerina of the heart, dancing on the wings of hearing (Te) A rainbow in the ears, woven by droplets of words (G) The warmth of your voice melts my frozen heart

[0090] These phrases are so expressive that they seem as if they were created by a human copywriter, expressions that would be impossible to generate using generative AI alone, which provides summary output.

[0091] The generating means 123 will generate a news release that includes at least one of these phrases.

[0092] Fig. 4B shows an example of the configuration of processor unit 120 in another embodiment. The embodiment shown in Fig. 4B has a similar configuration to the embodiment shown in Fig. 4A, except that processor unit 120 further includes evaluation means 124. Therefore, the configuration described with reference to Fig. 4A will not be described in detail here.

[0093] The processor unit 120 comprises a receiving means 121, an acquiring means 122, a generating means 123, and an evaluating means .

[0094] The receiving means 121 is configured to receive input information. The input information received by the receiving means 121 is passed to the obtaining means 122.

[0095] The obtaining means 122 is configured to obtain a plurality of representations based on the input information. The obtained representations are passed to the generating means 123.

[0096] The generating means 123 is configured to generate at least one phrase for the news release using at least one expression from the generated plurality of expressions, and to generate the news release including the generated at least one phrase. The generated news release is passed to the evaluating means 124.

[0097] The evaluation means 124 is configured to evaluate the generated news release.

[0098] The evaluation means 124 can evaluate and score expressions included in a news release based on preset evaluation indexes and output evaluation or scoring results. Each evaluation index has multiple items, and the evaluation means 124 calculates a score for each item by determining whether or not the criteria for each item are met. The evaluation means 124 can then output a total score calculated by adding up the scores for each item as the scoring result, or can output the result of determining whether or not the total score exceeds a threshold as the evaluation result.

[0099] The evaluation index may be an index for evaluating whether a news release is likely to be picked up by the mass media, and the evaluation means 124 can thereby evaluate and score the likelihood of the news release being picked up by the mass media. For example, the higher the total score as a scoring result, the more likely the news release is to be picked up by the mass media, and the more valuable or effective the news release is. For example, a news release with a total score exceeding a threshold value indicates the more likely the news release is to be picked up by the mass media, and the more valuable or effective the news release is. Mass media includes, but is not limited to, newspapers, magazines, television, radio, the Internet, etc.

[0100] The multiple items of the evaluation index may include, for example, the topicality or trendiness of the news release. Topicality or trendiness may be evaluated from multiple perspectives, including, but not limited to, the relevance of the expressions included in the news release to seasonal events, the agreement of the expressions included in the news release with current events or trends, the newsworthiness of the expressions included in the news release, the suitability of the expressions included in the news release for readers, and the reach of the expressions included in the news release.

[0101] The multiple items may include, for example, the attractiveness of the news release from a consumer perspective. The attractiveness from a consumer perspective may be evaluated from multiple perspectives, including, but not limited to, the clarity of the value to consumers of the expressions contained in the news release, the ability to solve readers' problems, the ability to call to action, the concreteness of the consumer experience, and the lack of a reader perspective.

[0102] The multiple items may include, for example, data and supporting performance of the news release. The data and supporting performance may be evaluated from multiple perspectives, including, but not limited to, the reliability of the data included in the news release, specificity, power of inciting action, specificity of consumer experience, and reader perspective.

[0103] The multiple items may include, for example, the appeal of the introduction of the news release. The appeal of the introduction may be evaluated from multiple perspectives, including, but not limited to, the clarity of the newsworthiness of the expressions contained in the news release, the use of data or concrete examples, the ability to attract the reader's attention, conciseness and clarity of focus, and timeliness.

[0104] The multiple items may include, for example, the market impact of the news release, which may be evaluated from multiple perspectives, including, but not limited to, the impact of the statements contained in the news release on market size, the industry bias, the impact on consumer behavior, the economic impact, and the contribution to sustainability.

[0105] The multiple items may include, for example, the originality of the news release. Originality may be evaluated from multiple perspectives, including, but not limited to, the uniqueness of the perspective expressed in the news release, the originality of the content, the freshness provided to readers, differentiation from competitors, and depth of in-depth information.

[0106] The evaluation index may include at least two of the six items described above, and may also include other items. Preferably, the evaluation index may include all six items described above. This may enable multifaceted evaluation and scoring.

[0107] The evaluation means 124 may evaluate and score the evaluation indexes on a rule basis, or may evaluate them using LM or LLM.

[0108] The evaluation means 124 can preferably calculate scores for each aspect of each item using a demerit system. This is because it is possible to evaluate the likelihood of being featured in the mass media with higher accuracy than with a point-addition system. In the demerit system, points can be deducted according to the degree to which the criteria for each of multiple aspects are not met.

[0109] For example, when evaluating topicality or trendiness among multiple items, the evaluation criterion is whether the content is directly related to seasonal events and attracts the interest of consumers and readers, and points are deducted depending on the extent to which it does not meet the criteria. For example, if the article is "partially related to seasonal events but lacks specificity and impact," one point is deducted; if the article is "vaguely related to seasonal events and lacks appeal," two points are deducted; and if the article is "completely unrelated to seasonal events," three points are deducted. Specifically, when evaluating an article about Christmas, one point is deducted for "limited-edition Christmas products and feature articles," two points are deducted for "contains elements related to Christmas but does not attract strong interest from readers," and three points are deducted for "dealing with products or themes unrelated to the Christmas season."

[0110] Using a concrete example, explain how to evaluate a passage about the chocolate festival "Salon du Chocolat 2025." For example, "Salon du Chocolat 2025 will be held from January 16th, where you can enjoy a luxurious chocolate experience created by cutting-edge chefs from France, Japan, and around the world. This year's theme is 'Moment / Infinity'. It is a special chocolate festival to coincide with the Valentine's season." If the phrase is generated, For "relevance to seasonal events," it is perfectly in line with the Valentine's season, so 0 points are deducted; for "alignment with current events and trends," it is in line with trends in the chocolate market, so 0 points are deducted; for "newsworthiness," the new theme and lineup of the event enhance its newsworthiness, so 0 points are deducted; for "suitability to readership," it is highly relevant to marketers and consumers, so 0 points are deducted; and for "wide reach," it is appealing to a wide range of people, from general consumers to industry insiders, so 0 points are deducted. Therefore, the evaluation value for this phrase is 15 points, with no deductions.

[0111] The scores or ratings from the rating means 124 may be output from the system 100 along with the generated news release.

[0112] The user can refer to the scoring or evaluation results to determine whether or not to adopt the news release as an actual news release.

[0113] Since system 100 generates phrases for news releases based on expressions contained in words that are not directly related to news releases, there is a risk that the quality of the generated news releases will decline as a result. However, by performing evaluation and scoring using evaluation means 124, news releases with poor evaluation and scoring results (i.e., low-quality news releases) are not adopted, thereby contributing to maintaining the quality of news releases.

[0114] In the above example, the evaluation means 124 evaluates and scores the news releases generated by the generation means 123, but the present invention is not limited to this. The evaluation means 124 can also evaluate and score news releases other than those generated by the generation means 123. For example, when a user inputs a news release that the user created or a news release created elsewhere into the system 100, the evaluation means 124 can evaluate and score the likelihood that the input news release will be picked up by the mass media.

[0115] 4A and 4B, the components of the processor unit 120 are provided in the same processor unit 120, but the present invention is not limited to this. A configuration in which the components of the processor unit 120 are distributed across multiple processor units is also within the scope of the present invention. In this case, the multiple processor units may be located in the same hardware component, or in separate hardware components located nearby or remotely.

[0116] Each component of the system 100 described above may be composed of a single hardware component or multiple hardware components. When composed of multiple hardware components, the manner in which the hardware components are connected does not matter. The hardware components may be connected wirelessly or by wire. The system 100 of the present invention is not limited to a specific hardware configuration. It is also within the scope of the present invention for the processor unit 120 to be configured using analog circuits rather than digital circuits. The configuration of the system 100 of the present invention is not limited to the one described above as long as it can realize its functions.

[0117] (3. Processing in the system for generating news releases) Figure 5 illustrates a data flow of processing in system 100 for generating a news release. Figure 5A illustrates data exchange between processor unit 120 of system 100 and multiple language models. Note that, although the multiple language models are shown as external to system 100 in this example, they may also be components of system 100.

[0118] In step S501, the receiving means 121 of the processor unit 120 receives input information. The input information includes at least a topic for a news release. The topic for a news release may be an outline of a matter for which a news release is to be generated. In addition to the topic, the input information may further include content or points that the user wants to explicitly emphasize in the news release (for example, the implementation of a campaign, etc.). Furthermore, the input information may include a framework for public relations.

[0119] In step S502, the acquiring means 122 of the processor unit 120 inputs the received input information into a plurality of language models to acquire a plurality of expressions. At this time, instructions for causing the plurality of language models to generate words may be input. The instructions for causing the plurality of language models to generate words may be, for example, instructions for causing the plurality of language models to chat, instructions for generating a diary, or instructions for generating poetry.

[0120] In step S503, the multiple language models generate words based on the input information. The multiple language models can generate words in accordance with instructions for generating words from the processor unit 120. The multiple language models can perform at least one of, for example, chatting, creating a diary or journal, and generating poetry or poems. For example, when instructed to chat, the multiple language models become virtual characters, and the virtual characters chat with each other based on the provided input information. For example, when instructed to create a diary, the multiple language models become virtual characters, and the virtual characters write the diary based on the provided input information. For example, when instructed to generate poetry, the multiple language models become virtual characters, and the virtual characters write poetry based on the provided input information.

[0121] In step S504, the multiple language models provide generated speech (eg, casual conversation, diary or journal, poem or verse) to the processor unit 120.

[0122] In step S505, the acquiring means 122 of the processor unit 120 acquires expressions included in the words provided in step S504. The acquiring means 122 may, for example, acquire words generated by a plurality of language models as expressions as they are, or may acquire parts of words generated by a plurality of language models as expressions. For example, the acquiring means 122 may acquire parts that express emotions during casual conversation as expressions.

[0123] In step S506, the generating means 123 of the processor unit 120 generates at least one phrase for the news release using at least one of the acquired expressions. For example, the generating means 123 may use the extracted expression as a phrase as is. Alternatively, the generating means 123 may combine some of the extracted expressions to generate a phrase.

[0124] The phrases generated in this way can sound as if they were written by a human copywriter, and news releases generated using these phrases can also sound as if they were written by a human copywriter. This process makes it possible to easily generate news releases with rich expressions without relying on human copywriters.

[0125] The processing in the system 100 may then continue to step S507, which may be the processing when the processor unit 120 has the configuration shown in FIG.

[0126] In step S507, the generating means 123 of the processor unit 120 generates a news release that includes at least one of the phrases generated in step S506. The phrases generated in step S506 can form, for example, a title, an abstract, a lead sentence, or a body sentence in the news release.

[0127] In step S508, the evaluation means 124 of the processor unit 120 evaluates and scores the news release generated in step S507. The evaluation means 124 evaluates expressions included in the news release based on preset evaluation indexes and can output a scoring result or an evaluation result. Each evaluation index has multiple items, and the evaluation means 124 calculates a score for each item by determining whether or not the criteria for each item are met. The evaluation means 124 may then output a total score calculated by adding up the scores for each item as the scoring result, or may output the result of determining whether or not the total score exceeds a threshold as the evaluation result.

[0128] The evaluation index may be an index for evaluating and scoring whether or not a news release is likely to be picked up by the mass media, thereby enabling the evaluation means 124 to evaluate and score the likelihood of the news release being picked up by the mass media.

[0129] The evaluation value or evaluation result is output from the system 100 along with the generated news release, and the user can refer to the scoring result or evaluation result to decide whether or not to adopt it as an actual news release.

[0130] In the example described above with reference to FIG. 5, the processes are described as being performed in a specific order, but the order of each process is not limited to that described and may be performed in any order that is logically possible.

[0131] In the example described above with reference to Figure 5, the processing of each step shown in Figure 5 can be realized by the processor unit 120 and a program stored in the memory unit 130, but at least one of the processing of each step shown in Figure 5 may also be realized by a hardware configuration such as a control circuit.

[0132] In the above example, the system 100 is a computer (e.g., a server device) installed at a provider that provides a service for generating news releases, but the present invention is not limited to this. The system 100 may be any information processing device that includes a processor.

[0133] The present invention is not limited to the above-described embodiments. It is understood that the scope of the present invention should be interpreted only by the claims. It is understood that a person skilled in the art can implement an equivalent scope based on the description of the present invention and common technical knowledge from the description of specific preferred embodiments of the present invention. [Example]

[0134] (Example) A news release was generated using the system 100 of the present invention. The topic given was information about "Rakuyaku AI," a generative AI SaaS solution for pharmaceutical companies. Specifically, the following input sentence was given: <input statement> Launch of "Rakuyaku AI," a generative AI SaaS solution for pharmaceutical companies https: / / www.rozetta.jp / rakuyakuai / Message for pharmaceutical companies (including CRO companies) The most important thing to say: Rakuyaku AI, which was previously a contract development solution service, will now be offered as a SaaS product. Service Features: (1) Consistency check ⇒Sentence to sentence is the only thing you can do ┗Until now, visual inspection was done by manpower, and competitors only provided word search services ⇒ Answers in under 1 second ★ Calculate the percentage of correct answers based on your own research (imitating the T4OO method) ★ Interview with pharmaceutical manufacturers about how long it has taken so far (2) Search function ⇒PMDA full text search ┗Until now, four layers of data were examined and compared one by one. ⇒You can search for drugs of the same type and with the same effects side by side (display them with just one button!) (3) Writing support tool (CSR editor) ⇒AI assists medical writers with the time-consuming task of writing documents (4) AI question answering tool (RAG specialized for the pharmaceutical industry) ⇒Utilize AI to achieve smoother searches and more efficient operations! Background of the service launch: Until now, Rakuyaku AI has been a contract development solution service, but in the course of sales activities, the company has narrowed down the major issues facing the pharmaceutical industry to four, and has turned them into a SaaS product using generative AI.

[0135] For this input sentence, the system 100 generated a news release based on phrases obtained from the results of having multiple language models conduct casual conversations, generate diaries, and generate poetry.

[0136] Below are the title, suggested titles, and lead sentences of the generated news release. ·title The moment when AI surpasses human accuracy! The one and only "Rakuyaku AI" with a 98% accuracy rate and response time of less than 1 second has become the new standard in the pharmaceutical industry. Title contenders AI surpasses human limits in just one second! Launch of "Rakuyaku AI," a new standard in the pharmaceutical industry "Dad, do I no longer have to work late?" Innovative AI solution that amazes children and improves work efficiency and accuracy Lead sentence We are entering an era where AI surpasses human accuracy. On January 21, 2025, Rakuyaku AI will be launched as a new standard in the pharmaceutical industry. This innovative SaaS solution performs data integrity checks in less than one second, boasts a 98% accuracy rate, and features industry-first functionality that reduces information search time by 50%. This reduces the workload of pharmaceutical company personnel and allows them to focus on strategic work. This solution, perfect for starting the new year, will promote efficiency across the industry and pave the way for the future. For more information, please visit the official website (www.rakuyakuai.com).

[0137] (Comparative Example) A news release was generated using ChatGPT o1 Pro mode provided by OpenAI. The topic was the same as in the previous example, information about "Rakuyaku AI," a generative AI SaaS solution for pharmaceutical companies. The following sentence was given as a prompt: The input sentence was the same as in the previous example. prompt You are a PR specialist. Create a press release based on the following information. A press release consists of a title, subtitle, lead, and body. Analyze your readership and create a press release that will interest them. #Story information: [Enter text here]

[0138] A news release was generated in response to this prompt.

[0139] Below are the title, subtitle, and lead of the generated news release: ·title The long-awaited SaaS version of "Rakuyaku AI" has been released -- an AI solution that improves operational efficiency for pharmaceutical companies has been released ·subtitle Japan's only comprehensive "sentence-to-sentence" consistency check function, PMDA full-text search, and AI-powered writing support and Q&A tools. Lead sentence Rosetta Inc. (Head office: Shinjuku-ku, Tokyo; CEO: XXXX) will begin offering "Rakuyaku AI," an AI solution for pharmaceutical companies and CRO companies, as a SaaS product. Rakuyaku AI, which has previously been offered as a contract development model, has been revamped into a format that can be quickly adopted and used by a wide range of pharmaceutical professionals, simultaneously improving operational efficiency and reducing risk.

[0140] (Consideration) We compare news releases generated by the system 100 of the present invention with those generated by chatGPT. The titles and title candidates of the news releases generated by the present invention contain phrases with rich expressions that seem human-created, such as personal expressions like "AI surpasses human accuracy," superlative expressions like "one of a kind," and conversational expressions like "Dad, do you want to work late?" In contrast, the titles and subtitles of the news releases generated by chatGPT consist only of summary phrases that list facts. [Industrial Applicability]

[0141] The present invention is useful in that it provides a system or the like that enables news releases to be generated using sentences that have rich expressions as if they were written by a human. [Explanation of symbols]

[0142] U User D. Expressions in Diaries C. Expressions in casual conversation P Expression in poetry R News Release 100 systems 200 Database Department 300 User terminal device

Claims

1. 1. A system for generating a news release, comprising: receiving means for receiving input information including topics for a news release; an acquiring means for acquiring a plurality of expressions based on the input information, the plurality of expressions being included in words generated by a plurality of language models; generating means for generating at least one phrase for a news release using at least one expression from the plurality of expressions; A system comprising:

2. the generating means further generates a news release including the generated at least one phrase; The system comprises: The system of claim 1 further comprising an evaluation means for evaluating the generated news release.

3. 3. The system of claim 2, wherein the evaluation means evaluates the generated news release by determining the likelihood of the generated news release being covered by mass media.

4. The evaluation means evaluates expressions included in the generated news release by: determining whether the expression satisfies criteria for a plurality of items, thereby calculating a score for each of the plurality of items; determining whether the total score obtained by adding up the scores of each item exceeds a threshold; The system of claim 2 , comprising:

5. The system of claim 4, wherein the plurality of items include at least one of the topicality of the news release, the appeal of the news release from a consumer perspective, the data and performance evidence contained in the news release, the appeal of the introductory part of the news release, the market impact of the news release, or the uniqueness of the news release.

6. The system according to claim 1 , wherein the acquiring means instructs the plurality of language models to conduct casual conversation, and the plurality of expressions include expressions used in casual conversation between the plurality of language models.

7. The system of claim 1 , wherein the acquisition means instructs at least one of the plurality of language models to generate a diary or journal, and the plurality of expressions includes expressions in the diary or journal.

8. The system of claim 7 , wherein the plurality of expressions includes onomatopoeic or mimetic words or expressions in hiragana.

9. The system of claim 1 , wherein the obtaining means instructs at least one of the plurality of language models to generate a poem, and the expression comprises an expression in a poem.

10. The system of claim 1 , wherein the generating means generates a title for the news release using at least one representation of the plurality of representations.

11. The system of claim 1 , wherein the generator generates a summary for the news release using at least one expression from the plurality of expressions.

12. The system according to claim 1 , wherein the generating means extracts, from the plurality of expressions, an expression that expresses emotion, wonder, sympathy, or doubt as the at least one expression.

13. An information processing method for generating a news release, the information processing method being performed on a computer having a processor, the information processing method comprising: the processor receiving input information including a topic for a news release; The processor obtains a plurality of expressions based on the input information, the plurality of expressions being included in vocabulary generated by a plurality of language models; and generating at least one phrase for a news release using at least one expression from the plurality of expressions; An information processing method including:

14. 1. A program for generating a news release, the program running on a computer having a processor, the program comprising: receiving input information including topics for a news release; Obtaining a plurality of expressions based on the input information, the plurality of expressions being included in words generated by a plurality of language models; generating at least one phrase for a news release using at least one expression of the plurality of expressions; A program causing the processor to perform processing including the steps of:

Images