Information provision device, information provision method, and information provision program

Abstract

According to one embodiment, an information provision device includes a collection unit and a generation unit. The collection unit collects the provided data source from a provider who has satisfied a condition related to the reliability. The generation unit generates a language model that is generated by learning the collected data source and that generates an answer to the cue input by the user.

Description

TECHNICAL FIELD

[0001] The disclosed embodiment relates to an information providing apparatus, an information providing method, and an information providing program. BACKGROUND

[0002] Conventionally, a system is known that generates an answer text for a question text input by a user using a generation model (for example, refer to Patent Literature 1).

[0003] PRIOR ART DOCUMENT PATENT LITERATURE Patent Literature 1: Japanese Patent Application Laid-Open No. 2022-503838 SUMMARY

[0004] In the conventional technology, there is room for improvement in improving the credibility of the provided information.

[0005] For example, the system described in Patent Literature 1 generates an answer text based on a news report. However, in a case where a news report is collected from an arbitrary provider, there are cases where the credibility of the news report is low depending on the origin of the news report.

[0006] The present application has been made in view of the above circumstances, and aims to improve the credibility of the provided information.

[0007] An information providing apparatus of one mode of the technical solution has a collection section that collects a provided information source from a provider that has satisfied a condition related to credibility, and a generation section that generates a language model that learns the collected data source to generate an answer for a prompt word input by a user.

[0008] According to one mode of the technical solution, it is possible to improve the credibility of the provided information. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 A diagram for explaining an outline of the information providing apparatus of the embodiment.

[0010] Figure 2 A functional block diagram for showing a configuration example of the information providing apparatus of the embodiment.

[0011] Figure 3 A diagram for explaining a change in the amount of data.

[0012] Figure 4 A flowchart for explaining a flow of the credibility determination processing.

[0013] Figure 5 A flowchart for explaining a flow of the pre-training processing.

[0014] Figure 6 A flowchart for explaining a flow of a process of the information providing apparatus.

[0015] Figure 7 A diagram for schematically showing an example of a hardware configuration of a computer functioning as the information providing apparatus. DETAILED DESCRIPTION

[0016] Hereinafter, the present application will be described by way of embodiments, but the following embodiments do not limit the application of the claims. In addition, the combination of features described in the embodiments is not necessarily all required for the technical solution of the application.

[0017] USING Figure 1 A flow of a process of the information providing apparatus of the embodiment is explained. Figure 1 A diagram for explaining an outline of the information providing apparatus of the embodiment. In Figure 1 , a configuration of the information providing system 1 including the information processing system 10 of the embodiment is shown.

[0018] As shown in Figure 1 , the information providing section 1 includes the information providing apparatus 10, the data source providers 20, and the answer providers 30. The entities of the data source providers 20 are servers 20-1 for publishing information or user terminals 20-2 held by users, vehicles 20-3 having an automatic driving function, and the like. The servers 20-1 are servers managed by, for example, a newspaper company, a news agency, a broadcasting organization, a publishing company, a network news providing organization, and the like. The servers 20-1 publish information on the Internet through a website or a social networking service (SNS). In addition, the respective servers 20-1 or the user terminals 20-2, the vehicles 20-3 of the data source providers 20, for example, one million in total, can all be connected to the information providing apparatus 10 in a plug-in manner. That is, the information providing apparatus 10 has a leadership function as a so-called command tower, collects information transmitted from all the chips mounted on one million of the data source providers 20, and performs learning.

[0019] The answer providers 30 are providers of answers to the user's questions using a language model possessed by the information providing apparatus 10. The entities of the answer providers 30 are user terminals 30-1 or vehicles 30-2 having an automatic driving function, and the like.

[0020] As shown in Figure 1 , the information providing apparatus 10 acquires and provides information related to the providers from the data source providers 20 (step S1). The information related to the providers is information of attribute information (business content or business scale, number of employees, number of member registrations) or a business development area of the providers, and the like.

[0021] Next, the information providing apparatus 10 determines whether the provider satisfies the condition related to credibility based on the information related to the provider (step S2). For example, for a provider whose business content is news distribution (newspaper company, news agency, broadcasting station, publishing company, web news providing company), the information providing apparatus 10 determines that the condition related to credibility is satisfied. That is, since the credibility of the data source of the news report provided by the provider whose business content is news distribution is high, the information providing apparatus 10 determines that the condition related to credibility is satisfied. Further, the information providing apparatus 10 determines that the provider whose business scale is a certain amount or more satisfies the condition related to credibility. The business scale is, for example, the number of distribution (for a newspaper company, the number of issued copies, for a news agency, the number of users who distribute, for a broadcasting station, the rating, for a publishing company, the number of issued copies, for a web news providing company, the number of distribution or the number of registered users, etc.) of news reports. That is, for a provider that distributes news reports to a certain number or more of users, the information providing apparatus 10 determines that the condition related to credibility is satisfied.

[0022] Further, in a case where the data source provider 20 is a machine such as the user terminal 20-2 or the vehicle 20-3, the information providing apparatus 10 determines whether the manufacturer of the machine satisfies the condition related to credibility. For example, the information providing apparatus 10 determines that the manufacturer whose sales amount is a certain amount or more satisfies the condition related to credibility.

[0023] Next, the information providing apparatus 10 permits communication connection with respect to the data source provider 20 that has satisfied the condition related to credibility, and collects data sources from the data source provider 20 (step S3). In addition, the information providing apparatus 10 and the data source provider 20 are connected through a virtual private network (VPN).

[0024] The data source is data in the form of text, sound, image, etc. that the server 20-1 of the data source provider 20 distributes. Further, the data source is the user's posted content (text, sound, image, etc.) that the user terminal 20-2 provides. The data source is the past movement path, the congestion situation of the road or sidewalk in movement, the movement time required to reach the destination, the operation (accelerator, brake, steering wheel, etc.) of the vehicle 20-3 when the vehicle 20-3 moves to the destination, etc. that the vehicle 20-3 provides.

[0025] Further, the data source includes comment data posted by a user (distribution target user) with respect to distributed data. The comment data is, for example, a user's comment with respect to a news report (distributed data).

[0026] The information providing apparatus 10 determines the trustworthiness of the collected data source. For example, the information providing apparatus 10 determines the trustworthiness based on the information of the provider. Specifically, the information providing apparatus 10 determines the trustworthiness based on the attribute information of the provider and the category of the data source. For example, in a case where the provider that mainly publishes news on economy provides a data source on economy, the information providing apparatus 10 determines that the trustworthiness is high, and in a case where the provider provides a data source other than on economy, the information providing apparatus 10 determines that the trustworthiness is low. In addition, the trustworthiness is not limited to binary values of high / low, and can be expressed in the form of a 10-level score of 1-10. Furthermore, for a data source provided by a provider having a business scale of a certain level or more, the information providing apparatus 10 can determine that the trustworthiness is high. The business scale is, for example, the number of publications for news publication (the number of issues for a newspaper, the number of users for a news agency, the rating for a broadcasting station, the number of issues for a publishing company, the number of publications or the number of registered users for a web news providing company, and the like). Furthermore, the information providing apparatus 10 can determine the trustworthiness of the data source provided this time based on the trustworthiness of the data source provided by the provider in the past. Specifically, in a case where the trustworthiness of the data source provided by the provider in the past is all high, the information providing apparatus 10 determines that the trustworthiness of the data source provided this time is high. Furthermore, in a case where the trustworthiness of the data source on a specific field among the data source provided by the provider in the past is all high, in a case where the data source provided this time is on the specific field, the information providing apparatus 10 determines that the trustworthiness is high.

[0027] The information providing apparatus 10 performs learning of the model based on the acquired data source. The model is a language model. As the language model, for example, ChatGPT of OpenAI is known (reference: https: / / openai.com / blog / chatgpt). The language model can also be a generative model using a neural network such as a generative adversarial network (GAN) or a variational autoencoder (VAE).

[0028] In the present embodiment, the language model generates a text of an answer to a text of a question (hereinafter, a prompt word) input by the user. Furthermore, the model is learned in advance using the data source. The learning of the model can be performed by a known machine learning method.

[0029] Thus, the model can generate an answer based on the data source. For example, in a case where the prompt word includes a keyword of "electric vehicle", the model can generate an answer based on a news report associated with the keyword. In other words, by learning, the content of the news report can have an influence on the model to generate an answer.

[0030] In step S1, the information providing device 10 continuously acquires data sources. Therefore, the amount of data sources acquired by the information providing device 10 increases continuously over time.

[0031] The information providing device 10 has two models that use different data sources for learning. The two models are a standard model and an advanced model.

[0032] Information providing device 10 learns two language models based on data sources (step S4). Specifically, information providing device 10 learns a standard model based on past data sources. In addition, information providing device 10 learns an advanced model based on current data sources.

[0033] Here, "past data source" refers to the data source acquired before a certain time period prior to the time when the model learning begins. For example, assuming the certain time period is two years, if learning begins at "2023 / 6 / 22 16:05", then the past data source would be the data source acquired from the start time of data source acquisition to "2021 / 6 / 22 16:05".

[0034] On the one hand, the current data source is the data source acquired up to the time point when model learning begins. For example, if learning begins at "2023 / 6 / 22 16:05", the current data source is the data source acquired from the start time of data source acquisition to "2023 / 6 / 22 16:05".

[0035] In other words, the information providing device 10 generates a standard model and an advanced model. The standard model's learning cutoff point is the past data source acquired before a predetermined time period. The advanced model's learning cutoff point includes the past data source acquired up to the present of the predetermined time period.

[0036] The model can be learned once per second or once every shorter interval (e.g., once per nanosecond).

[0037] Furthermore, when learning advanced models, the information providing device 10 uses data sources whose reliability meets the predetermined conditions, obtained within a predetermined time period, for example. Specifically, the information providing device 10 uses data sources with higher reliability to learn advanced models. Thus, it is possible to learn advanced models with higher reliability while reducing the learning workload.

[0038] In addition, the information providing device 10 receives from the user of the answer provider a designation of providers prohibited from being used for language model learning, and generates a language model based on the data source of each user learning from providers other than the designated provider.

[0039] For example, when the user receives a directive prohibiting the use of Company A's data source for learning, the information providing device 10 learns data sources from providers other than Company A. Then, the generated language model is a user-specific language model that has received the directive. That is, when the user inputs a prompt, the user-specific language model is input to generate an answer. As a result, the answer generated from the language model can exclude data sources from providers prohibited by the user. Thus, for example, it is possible to provide an answer that excludes data sources from providers that do not meet the user's preferences or that the user does not trust. In other words, the information providing device 10 according to this embodiment can provide appropriate information in accordance with the user (the user).

[0040] Furthermore, as a condition for generating such a language model based on each user, the information providing device 10 can use whether the user has joined a paid membership program (described later) as a condition. That is, for users who have joined a paid membership program, the information providing device 10 can generate the aforementioned dedicated language model, while for users who have joined a freemium program, no dedicated language model is generated (a common language model for all users is used).

[0041] Furthermore, when learning the aforementioned common language model for users, the information providing device 10 can adjust the learning amount of each provider in the language model based on the specified reception results received from the user that are prohibited from being used for language model learning. The language model is used to provide answers to users who have joined the freemium program.

[0042] For example, for providers with a large number (number of users) designated as prohibited from use in language model learning, the information providing device 10 reduces the learning amount when using the data source provided by that provider for common language model learning. On the other hand, for providers with a small number (number of users) designated as prohibited from use in language model learning, the information providing device 10 increases the learning amount when using the data source provided by that provider for common language model learning. Furthermore, the learning amount can be determined, for example, based on the designated number (number of users). Additionally, for providers with a designated number (number of users) exceeding a predetermined number, the information providing device 10 can prohibit their use in language model learning. Furthermore, for providers with a designated number (number of users) exceeding a predetermined number, the information providing device 10 can exclude them from the data source provider list (terminate the contract as a provider).

[0043] In other words, the more providers are prohibited from being used for language model learning, the more the information providing device 10 judges them as providers with lower user credibility and reduces the learning load; conversely, the fewer the number of prohibited providers, the more the information providing device 10 judges them as providers with higher user credibility and increases the learning load. Thus, by using the received results from the prohibited providers to learn the user's common language model, the credibility of the answers provided by the language model can be improved. That is, the information providing device 10 according to the embodiment can provide users with more credible information.

[0044] Furthermore, the information providing device 10 adjusts the learning amount of each provider in the language model based on user utilization of the services provided by the providers. Specifically, the information providing device 10 adjusts the learning amount of each provider in the language model based on the number of users who have registered for the service, the number of times the service is used, and the frequency of use (time or number of times). For example, the information providing device 10 increases the learning amount of data sources provided by providers with a larger number of registered users, a higher number of times the service is used, and a higher frequency of use than the data sources provided by other providers. Additionally, the learning amount can be determined based on the number of users who have registered for the service, the number of times the service is used, and the frequency of use. In other words, the better the service usage of a provider, the higher the credibility of the data source provided by that provider is judged by the information providing device 10, and the higher the learning amount is. This improves the credibility of the answers provided by the language model. That is, the information providing device 10 according to this embodiment can provide users with more credible information.

[0045] Furthermore, the information providing device 10 uses the published data and comment data included in the data source for model learning. This comment data can include supplementary information to the published data, more detailed information, information from other perspectives, etc. Therefore, by using such comment data for learning, the accuracy of the responses output from the model can be further improved.

[0046] Furthermore, the information providing device 10 learns its model based on comments containing positive content (positive comments) and comments containing negative content (negative comments) from the comment data. Positive content comments include those that praise, agree, affirm, or supplement the published data. Negative content comments include those that criticize, oppose, or negate the published data, those containing extreme language, and those that differentiate the content.

[0047] For example, the information providing device 10 prohibits the use of negative comments in the comment data for model learning. This prevents the model from outputting responses based on negative comments, thus reducing the unpleasant experience for the user receiving the response.

[0048] Furthermore, when the number of comments in the comment data exceeds a threshold, the information providing device 10 uses that comment data for model learning. In other words, the information providing device 10 determines whether the number of comments in each piece of published data exceeds a threshold. If it does, both the published data and the comment data are used for model learning; if the number of comments is less than the threshold, only the published data is used for learning. This avoids situations where the number of comments is too small to ensure comment diversity.

[0049] Furthermore, the information providing device 10 adjusts the amount of data used for model learning based on the number of positive and negative comments within the total number of comments. For example, if the proportion of negative comments in all comments of the comment data exceeds a threshold, the information providing device 10 prohibits the use of all comments or negative comments for model learning.

[0050] Furthermore, the information providing device 10 ensures that the learning amount for published data where the proportion of positive comments is higher than the proportion of negative comments is higher than that for published data where the proportion of positive comments is lower than the proportion of negative comments. For example, in the case of published data where the proportion of positive comments is lower than the proportion of negative comments, the information providing device 10 does not use the content related to negative comments in that published data for model learning.

[0051] Furthermore, the information providing device 10 excludes providers whose negative comments exceed a predetermined number from the data source collection objects. This allows for the exclusion of providers who publish a large number of negative comments—in other words, providers whose data publication credibility is low—from the data source collection objects, thereby improving the credibility of the collected data source.

[0052] In this step, the information providing device 10 receives prompt words input from the user to the language model (step S5). Then, the information providing device 10 uses the model to generate a response to the prompt words based on the user's suggestion (step S6).

[0053] Additionally, prompts from the user are sent to the information providing device 10 via the answer provider 30. Furthermore, the information providing device 10 sends the answer to the answer provider 30.

[0054] Information providing device 10 acts as a server to provide answers. The server contains premium paid membership plans and secondary freemium plans. Paid membership plans are subscription services that charge a fixed fee for a specific period (e.g., annually or monthly), or pay-as-you-go plans where the fee is determined based on usage. Furthermore, freemium plans are free. Freemium plans can also be replaced with plans that cost less than paid membership plans (e.g., basic plans).

[0055] The information providing device 10 differentiates the service quality for each plan based on the differences in the learning data used for the models employed. Specifically, when a user subscribes to a paid membership plan, the information providing device 10 uses an advanced model to generate the answer. Conversely, when a user subscribes to a freemium plan, the information providing device 10 uses a standard model to generate the answer.

[0056] The information providing device 10 provides the generated answer to the user (step S7). The information providing device 10 can receive prompts and provide answers through a chat-like user interface. Furthermore, if the answer provider 30 is the vehicle 30-2, the information providing device 10 may display the answer on a display screen mounted on the vehicle 30-2 (e.g., a navigation device). Alternatively, if the prompt is an autonomous driving request to the destination, the information providing device 10 controls the vehicle 30-2's control system to perform autonomous driving.

[0057] Furthermore, when a user joins a paid membership program, the information providing device 10 can update the advanced model based on the latest data source to generate further answers. This updated model is called the real-time model. The information providing device 10 can provide supplementary answers generated using the real-time model.

[0058] The information providing device 10 continuously acquires data sources to update its language model during the period between generating and providing the generated answer. Using a real-time model, the information providing device 10 is able to generate answers that reflect the differences that occur during that period.

[0059] Next, when providing users with answers related to the data source, the information providing device 10 pays a reward to the data source provider 20 (step S8). For example, the information providing device 10 determines the content of the reward based on the number of times answers related to the data source are provided (or the number of users). Furthermore, the information providing device 10 can differentiate the rewards for providing a standard model and providing an advanced model. For example, the information providing device 10 may offer higher rewards for providing answers related to the data source using an advanced model than for providing them using a standard model.

[0060] Furthermore, the information providing device 10 pays a reward based on the credibility determined in step S4. For example, the higher the credibility, the more reward the information providing device 10 pays. Additionally, the information providing device 10 can determine the reward based on the credibility of multiple data sources previously provided to the user. Specifically, the more data sources with high credibility provided within a predetermined time period, the more reward the information providing device 10 pays. Thus, according to this embodiment, the information providing device 10 provides rewards to the provider based on credibility, thereby enabling the payment of appropriate rewards to the provider.

[0061] In this way, the information providing device 10 collects data sources only from the data source provider 20 that has met the conditions related to credibility, thereby avoiding the mixing of low-credibility data sources in language model learning and thus improving the credibility of the provided information.

[0062] Furthermore, in the information providing system 1, the information providing device 10 and the data source provider 20 (server 20-1, user terminal 20-2, and vehicle 20-3) are embedded with semiconductor chips of the same manufacturer. These semiconductor chips themselves can perform machine learning or deep learning independently.

[0063] Furthermore, the semiconductor chips used by the information providing device 10 or the data source provider 20 are chips adapted to their respective housing sizes. For example, assuming the chip size of the information providing device 10 is XL, the server 20-1 or vehicle 20-3 uses L size, and the user terminal 20-2 uses S or M size. Additionally, for the data source provider 20, which has a smaller housing size than the user terminal 20-2, an SS size is used, and a system-on-a-chip (SoC) is incorporated to achieve single-chip integration. The number of semiconductor chips for XL size is 200 watts, for L size it is 50 watts, for M size it is 20 watts, for S size it is 10 watts, and for SS size it is 2 watts, but the number of tiles for each size is just an example and is not limited to the above. If auxiliary devices such as cameras and microphones need to be equipped in the user terminal 20-2, dedicated chips for these auxiliary devices, independent of the semiconductor chips, can be set in the free space near the semiconductor chips in the system-on-a-chip (SoC).

[0064] By using semiconductor chips from the same manufacturer in both the information providing device 10 and the data source provider 20, a closed system can be established between them, thus maintaining a secure state between them. In other words, this not only effectively prevents hacking, virus infection, and deepfakes between the information providing device 10 and the data source provider 20, but also ensures their privacy.

[0065] use Figure 2 The configuration of the information providing device 10 is explained. Figure 2 A functional block diagram illustrating an example configuration of an information providing device for an implementation method.

[0066] like Figure 2 As shown, the information providing device 10 includes a communication unit 11, a storage unit 12, and a control unit 13.

[0067] The communications unit 11 sends and receives information with the data source provider 20 or the response provider 30 via the network.

[0068] The storage unit 12 is implemented using semiconductor storage elements such as random access memory (RAM) and flash memory, or storage devices such as hard disk drives (HDDs), solid-state drives (SSDs), and optical discs. The storage unit 12 stores various programs and various data. The storage unit 12 includes user information 121, standard model information 122, advanced model information 123, and real-time model information 124.

[0069] User information 121 contains information about the plans each user has joined. For example, user information 121 is information that associates a user ID with a paid membership plan or a freemium plan.

[0070] Standard model information 122, advanced model information 123, and real-time model information 124 respectively contain information about the parameters used to construct the standard model, advanced model, and real-time model. Parameters used to construct the model include, for example, the weights and biases of the neural network. These parameters are updated during model learning.

[0071] The control unit 13 is a controller, and may include, for example, a microcomputer with a central processing unit (CPU), read-only memory (ROM), random access memory (RAM), input / output ports, and various circuits. Alternatively, the control unit 13 may be constructed from hardware such as application-specific integrated circuits (ASICs) or field-programmable gate arrays (FPGAs). The control unit 13 includes an acquisition unit 31, a determination unit 32, a collection unit 33, a generation unit 34, a provisioning unit 35, and a payment unit 36.

[0072] The Acquisition Department 31 obtains information related to the data source provider 20. This information includes the provider's attribute information (business content, business scale, number of employees, number of registered members) or information such as the regions where the business is conducted.

[0073] The determination unit 32 determines whether a provider meets the credibility-related conditions based on information related to the provider. For example, the determination unit 32 determines that a provider whose business content is news publishing (newspapers, news agencies, broadcasters, publishers, online news providers) meets the credibility-related conditions. That is, because the data source of news reports provided by providers whose business content is news publishing is relatively credible, the determination unit 32 determines that they meet the credibility-related conditions. In addition, the determination unit 32 determines that providers with a certain or greater business scale meet the credibility-related conditions. Business scale is, for example, the volume of news publications (circulation for newspapers, number of users for news agencies, viewership for broadcasters, circulation for publishers, and number of publications or registered users for online news providers, etc.). In other words, for providers that publish news reports to a certain or greater number of users, the determination unit 32 determines that they meet the credibility-related conditions.

[0074] Furthermore, when the data source provider 20 is a machine such as user terminal 20-2 or vehicle 20-3, the determination unit 32 determines whether the machine manufacturer meets the conditions related to credibility. For example, the determination unit 32 determines that a provider with sales exceeding a certain amount meets the conditions related to credibility.

[0075] Furthermore, the determination unit 32 determines the credibility of the collected data sources. For example, the determination unit 32 determines credibility based on the information provided by the provider. Specifically, the determination unit 32 determines credibility based on the provider's attribute information and the category of the data source. For example, if a provider that mainly publishes economic-related news provides an economic-related data source, the determination unit 32 determines it to be highly credible; if it provides a data source other than economic-related news, it determines it to be less credible. In addition, credibility is not limited to a binary value of high / low, but can also be expressed as a 10-level rating from 1 to 10. Furthermore, for data sources provided by providers with a certain business scale, the determination unit 32 can determine them to be highly credible. Business scale is, for example, the volume of news publications (for newspapers, circulation; for news agencies, the number of users; for broadcasters, viewership; for publishers, circulation; for online news providers, the number of publications or registered users, etc.). Furthermore, the determination unit 32 can determine the credibility of the data source provided this time based on the credibility of the data sources previously provided by the provider. Specifically, if the credibility of all data sources previously provided by the provider is high, the determination unit 32 determines the credibility of the data source provided this time to be high. Furthermore, if the credibility of all data sources in a specific field provided by the provider in the past is high, then the determination unit 32 determines that the credibility of the data source provided this time is in a specific field.

[0076] The collection unit 33 collects data sources from data source providers 20 that are determined by the determination unit 32 to meet the conditions related to credibility. The collection unit 33 stores the collected data sources in the storage unit 12. Furthermore, the collection unit 33 continuously collects data sources asynchronously with the actions of other processing units. That is, the collection unit 33 continuously collects data sources asynchronously with processes such as model training, response generation, and provision.

[0077] like Figure 3 As shown, the amount of data stored from the collection unit 33 to the storage unit 12 changes over time. Figure 3 A graph illustrating the changes in data volume.

[0078] Figure 3 The horizontal axis represents time (moment). Furthermore, Figure 3 The vertical axis represents the amount of data. The amount of data at the current time t2 is v2. Furthermore, the amount of data at time t1, which is a time interval T earlier than time t2, is v1. Also, v1 is smaller than v2.

[0079] Furthermore, the advanced model is trained at time t2, and it is assumed that the time to generate an answer using the trained advanced model is t2 + Δt. Since the collection unit 33 continuously collects data, the amount of data increases by Δv during the time period Δt.

[0080] The generation unit 34 generates language models. Specifically, the generation unit 34 generates a standard model and an advanced model. Specifically, the generation unit 34 learns two language models based on data sources. Specifically, the generation unit 34 learns the standard model based on past data sources. In addition, the generation unit 34 learns the advanced model based on current data sources.

[0081] Furthermore, the generation unit 34 updates the model's parameters through learning. The generation unit 34 can regenerate the model at each learning opportunity, or it can reflect data sources that differ from the previous learning in the model.

[0082] Furthermore, the generation unit 34 determines the credibility of the data source collected from the provider and decides whether to use it for language model learning based on the credibility. For example, the generation unit 34 parses the content of the data source (text parsing, sound parsing, image parsing), and if there is a possibility that the content of the data source is false, it prohibits the use of the data source for language model learning.

[0083] Furthermore, when learning the advanced model, the generation unit 34 uses data sources whose credibility, as determined in step S4, meets predetermined conditions for learning from data sources acquired within a predetermined time period. Specifically, the generation unit 34 uses data sources with higher credibility to learn the advanced model.

[0084] In addition, the generation unit 34 receives from the user of the answer provider a designation of providers prohibited from being used for language model learning, and generates a language model based on each user learning data sources from providers other than the designated providers.

[0085] For example, when a user receives a directive prohibiting the use of Company A's data source for learning, the generation unit 34 learns data sources from providers other than Company A. Then, the generated language model is a user-specific language model that has received the directive. That is, when the user inputs a prompt, the user-specific language model is input to generate an answer. As a result, the answer generated from the language model can exclude data sources from providers prohibited by the user. Therefore, for example, it is possible to provide an answer that excludes data sources from providers that do not meet the user's preferences or that the user does not trust. In other words, the generation unit 34 can provide appropriate information tailored to the user (the user).

[0086] Furthermore, as a condition for generating this user-based language model, the generation unit 34 can use whether each user has joined the paid membership program (described later) as a condition. That is, for users who have joined the paid membership program, the generation unit 34 can generate the aforementioned dedicated language model, while for users who have joined the freemium program, no dedicated language model is generated (a common language model for all users is used).

[0087] Furthermore, when learning the aforementioned common language model for users, the generation unit 34 can adjust the learning amounts of the providers in the language model based on the specified reception results received from the user that are prohibited from being used for language model learning. The language model is used to provide answers to users who have joined the freemium program.

[0088] For example, for providers with a large number (number of users) designated as prohibited from use in language model learning, the generation unit 34 reduces the learning amount when using the data source provided by that provider for common language model learning. Conversely, for providers with a small number (number of users) designated as prohibited from use in language model learning, the generation unit 34 increases the learning amount when using the data source provided by that provider for common language model learning. Furthermore, the learning amount can be determined based on the designated number (number of users). Additionally, for providers with a designated number (number of users) exceeding a predetermined number, the generation unit 34 may prohibit their use in language model learning. Furthermore, for providers with a designated number (number of users) exceeding a predetermined number, the generation unit 34 may exclude them from the data source provider list (terminate from the provider's contract).

[0089] In other words, the more providers are prohibited from being used for language model learning, the more the generation unit 34 judges them as providers with low user credibility and reduces the learning amount; conversely, the fewer the specified providers, the more the information providing device 10 judges them as providers with high user credibility and increases the learning amount. Thus, by using the received results from the prohibited providers to learn the user's common language model, the credibility of the answers provided by the language model can be improved. That is, according to the generation unit 34, more credible information can be provided to the user.

[0090] Furthermore, the generation unit 34 adjusts the learning amount of each provider in the language model based on user utilization of the services provided by the providers. Specifically, the generation unit 34 adjusts the learning amount of each provider in the language model based on the number of registered users, the number of times the service is used, and the frequency of use (time or number of uses). For example, the generation unit 34 increases the learning amount of data sources provided by providers with a larger number of registered users, a higher number of times the service is used, and a higher frequency of use than the data sources provided by other providers. Additionally, the learning amount can be determined based on the number of registered users, the number of times the service is used, and the frequency of use. In other words, the better the service usage of a provider, the higher the credibility of the data source provided by that provider is judged by the generation unit 34, and the higher the learning amount is increased. This improves the credibility of the responses provided by the language model. That is, according to the generation unit 34, more credible information can be provided to users.

[0091] Furthermore, the generation unit 34 uses the published data and comment data included in the data source for model learning. This comment data can include supplementary information, more detailed information, information from other perspectives, etc., thus, by using such comment data for learning, the accuracy of the responses output from the model can be further improved.

[0092] Furthermore, the generation unit 34 learns its model based on comments with positive content (positive comments) and comments with negative content (negative comments) in the comment data. Positive content comments include those that praise, agree, affirm, or supplement the published data. Negative content comments include those that criticize, oppose, or negate the published data, those containing extreme language, and those that differentiate the content.

[0093] For example, the generation unit 34 prohibits the use of negative comments in the comment data for model learning. This prevents the model from outputting answers based on negative comments, thus reducing the unpleasant experience for users receiving the responses.

[0094] Furthermore, when the number of comments in the comment data exceeds a threshold, the generation unit 34 uses that comment data for model learning. In other words, the generation unit 34 determines whether the number of comments in the comment data of each published data set exceeds a threshold. If it does, both the published data and the comment data are used for model learning; if the number of comments is less than the threshold, only the published data is used for learning. This avoids situations where the number of comments is too small to ensure comment diversity.

[0095] Furthermore, the generation unit 34 adjusts the amount of learning material used for the model based on the number of positive and negative comments among the total number of comments. For example, if the proportion of negative comments in all comments of the comment data exceeds a threshold, the generation unit 34 prohibits the use of all comments or negative comments for model learning.

[0096] Furthermore, the generation unit 34 ensures that the learning amount for published data where the proportion of positive comments is higher than the proportion of negative comments is higher than that for published data where the proportion of positive comments is lower than the proportion of negative comments. For example, in the case of published data where the proportion of positive comments is lower than the proportion of negative comments, the generation unit 34 does not use the content related to negative comments in that published data for model learning.

[0097] Furthermore, the generation unit 34 excludes providers whose posts contain negative comments exceeding a certain threshold from the data source collection objects. This allows for the exclusion of providers who post a large number of negative comments—in other words, providers whose posting data has high credibility—from the data source collection objects, thereby improving the credibility of the collected data source.

[0098] The provision unit 35 uses the generated language model to generate and provide responses to the prompts received from the user.

[0099] Department 35 refers to user information 121 to determine the user's plan. Department 35 determines whether the user has joined a paid membership plan or a free value-added plan.

[0100] The providing unit 35 inputs the prompt words into a language model and generates an answer. The providing unit 35 uses any one of multiple language models to generate an answer based on the prompt words input by the user, wherein the multiple language models are learned from different data sources. For example, the providing unit 35 uses either a standard model or an advanced model to generate an answer.

[0101] The providing unit 35 provides the user with the answer generated by the language model and information representing the data source provider, whereby the data source is used for learning the language model, and the language model is used to generate the answer. For example, the providing unit 35 causes the user terminal to display an information providing screen. For example, the providing unit 35 can receive and respond to prompts through a chat-style user interface. Furthermore, if the answer provider 30 is the vehicle 30-2, the providing unit 35 may display the answer on a display screen mounted on the vehicle 30-2 (e.g., a navigation device). Alternatively, when the prompt is an autonomous driving request to the destination, the providing unit 35 controls the control device of the vehicle 30-2 to perform autonomous driving.

[0102] Payment unit 36 ​​pays a fee to the data source provider 20 when providing users with answers related to the data source. For example, payment unit 36 ​​determines the content of the fee based on the number of times (or the number of users) an answer is provided related to the data source. Furthermore, payment unit 36 ​​can differentiate the fees for providing a standard model and an advanced model. For example, payment unit 36 ​​may offer higher fees for providing answers related to the data source through the advanced model than for providing them through the standard model.

[0103] Furthermore, the payment unit 36 ​​pays a reward based on the credibility determined in step S4. For example, the higher the credibility, the more reward the payment unit 36 ​​pays. Additionally, the payment unit 36 ​​can determine the reward based on the credibility of multiple data sources provided to the user in the past. Specifically, the more data sources with high credibility provided within a predetermined time period, the more reward the payment unit 36 ​​pays.

[0104] use Figure 4 Explain the process for determining credibility. Figure 4 A flowchart illustrating the process of determining credibility.

[0105] like Figure 4 As shown, the information providing device 10 obtains information related to the data source provider 20 (step S101).

[0106] Next, the information providing device 10 determines whether the conditions related to credibility are met based on the information related to the provider (step S102).

[0107] If the conditions related to credibility are met (step S102: Yes), the information providing device 10 adopts the provider as the data source (step S103) and ends the process. On the other hand, if the conditions related to credibility are not met (step S102: No), the information providing device 10 does not adopt the provider and ends the process.

[0108] use Figure 5Explain the process of pre-training. Figure 5 A flowchart illustrating the pre-training process.

[0109] like Figure 5 As shown, the information providing device 10 continuously acquires data sources (step S201). Next, the information providing device 10 determines whether it is time to generate a model (step S202).

[0110] If it is not a model generation opportunity (step S202: No), the information providing device 10 returns to step S201 and continues to acquire the data source. On the other hand, if it is a model generation opportunity (step S202: Yes), the information providing device 10 proceeds to step S203.

[0111] The information providing device 10 uses past data sources from a certain time period to train the standard model (step S203). Furthermore, the information providing device 10 uses current data sources to train the advanced model (step S204). For example, the information providing device 10 uses current data sources to train the advanced model and uses data sources from two years ago to train the standard model.

[0112] use Figure 6 Explain the process of providing and processing information. Figure 6 A flowchart illustrating the information provision process. The information provision device 10 is capable of performing pre-training processing and information provision processing in parallel.

[0113] like Figure 6 As shown, the information providing device 10 continuously acquires data sources (step S301). The information providing device 10 determines whether there is an inquiry from a user (step S302).

[0114] If there is no inquiry from the user (step S302: No), the information providing device 10 returns to step S301 to continue acquiring the data source. On the other hand, if there is an inquiry from the user (step S302: Yes), the information providing device 10 determines the user's plan (step S303).

[0115] If the user's plan is a freemium plan (step S303: freemium), the information providing device 10 proceeds to step S304. On the other hand, if the user's plan is a paid membership plan (step S303: paid membership), the information providing device 10 proceeds to step S306.

[0116] In step S304, the information providing device 10 inputs the query content into the standard model and generates an answer. Then, the information providing device 10 provides the generated answer to the user (step S305). Next, the information providing device 10 pays a fee to the provider of the data source related to the answer (step S311), ending the process.

[0117] In step S306, the information providing device 10 inputs the query content into the advanced model and generates an answer. Then, the information providing device 10 provides the generated answer to the user (step S307).

[0118] Further, the information providing device 10 updates the advanced model using the latest data source (step S308). The information providing device 10 inputs the query content into the updated advanced model (real-time model) and generates an answer (step S309). The information providing device 10 provides the generated answer to the user (step S310).

[0119] According to this embodiment, by establishing a cooperative relationship with companies that are generally recognized as excellent enterprises, and using such enterprises as data source providers, the reliability of the service that provides answers using language models can be improved.

[0120] Furthermore, according to this implementation method, it is possible to provide a general answer to the free value-added service to individual free members, and a high-precision answer to the paid membership service to individual paid members, enterprises, governments and other public institutions.

[0121] Figure 7 This diagram schematically illustrates an example of the hardware configuration of a computer functioning as an information providing device. A program installed on computer 1200 enables computer 1200 to function as one or more "parts" of the apparatus of this embodiment, or to perform operations associated with the apparatus of this embodiment or one or more "parts," and / or to perform the process of this embodiment or a stage of that process. Such a program, in order to enable computer 1200 to perform specific operations associated with several or all of the blocks in the flowcharts and block diagrams described in this specification, can be executed by CPU 1212.

[0122] The computer 1200 of this embodiment includes a CPU 1212, RAM 1214, and a graphics controller 1216, which are interconnected via a main controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive, and an IC card driver, which are connected to the main controller 1210 via an input / output controller 1220. The DVD drive can be a DVD-ROM drive or a DVD-RAM drive, etc. The storage device 1224 can be a hard disk drive or a solid-state drive, etc. The computer 1200 also includes input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0123] CPU 1212 operates according to the program stored in ROM 1230 and RAM 1214, thereby controlling each unit. Graphics controller 1216 retrieves image data generated by CPU 1212 from frame buffers or other sources provided in RAM 1214 or from itself, and the image data is displayed on display device 1218.

[0124] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 within the computer 1200. The DVD drive reads programs or data from a DVD-ROM or similar source and provides them to the storage device 1224. The IC card driver reads programs and data from an IC card and / or writes programs and data to the IC card.

[0125] ROM 1230 stores boot programs and / or programs that depend on the hardware of computer 1200, which are executed by computer 1200 upon activation. Input / output chip 1240 can also connect various input / output components to input / output controller 1220 via USB port, parallel port, serial port, keyboard port, mouse port, etc.

[0126] The program is provided by a computer-readable storage medium such as a DVD-ROM or an IC card. The program is read from the computer-readable storage medium, or installed in a storage device 1224, RAM 1214, or ROM 1230, which are also examples of computer-readable storage media, and executed by the CPU 1212. The information processing described within these programs is read by the computer 1200, enabling cooperation between the program and the aforementioned various types of hardware resources. An apparatus or method can be constructed by implementing the manipulation or processing of information according to the use of the computer 1200.

[0127] For example, when communication is performed between computer 1200 and an external device, CPU 1212 can execute a communication program loaded in RAM 1214 and perform command communication processing on communication interface 1222 based on the processing described in the communication program. Under the control of CPU 1212, communication interface 1222 reads transmission data stored in a transmission buffer provided in a recording medium such as RAM 1214, storage device 1224, DVD-ROM, or IC card, and sends the read transmission data to the network, or writes received data received from the network into a receive buffer provided on the recording medium.

[0128] In addition, CPU 1212 can read all or a portion of files or databases stored in external recording media such as storage device 1224, DVD drive (DVD-ROM), IC card, etc., from RAM 1214, and perform various types of processing on the data in RAM 1214. CPU 1212 can then write the processed data back to the external recording medium.

[0129] Various types of information, such as programs, data, tables, and databases, are stored in the recording medium, which can handle information processing. The CPU 1212 can perform various types of processing on data read from the RAM 1214 and write the results back to the RAM 1214. These various types of processing include operations specified by a sequence of program instructions, as described anywhere in this disclosure, such as information processing, conditional judgment, conditional branching, unconditional branching, and information retrieval / replacement. Furthermore, the CPU 1212 can retrieve information from files, databases, etc., within the recording medium. For example, if the recording medium contains multiple entries, each with an attribute value of a first attribute associated with a second attribute value, the CPU 1212 can retrieve from these multiple entries an entry that matches the condition specifying the first attribute value, and read the attribute value of the second attribute stored in that entry, thereby obtaining the attribute value of the second attribute associated with the first attribute that satisfies a preset condition.

[0130] The aforementioned program or software module can be stored on or near the computer 1200 on a computer-readable storage medium. Alternatively, a recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the program to the computer 1200 via the network.

[0131] In this embodiment, the boxes in the flowcharts and block diagrams may represent stages of a process for performing an operation or "parts" of a device having the function of performing an operation. Specific stages and "parts" may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable storage medium, and / or a processor provided with computer-readable instructions stored on a computer-readable storage medium. Dedicated circuitry may include both digital and / or analog hardware circuitry, as well as integrated circuits (ICs) and / or discrete circuitry. Programmable circuitry may include, for example, reconfigurable hardware circuitry such as field-programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), including logical AND, logical OR, logical XOR, logical NAND, logical NOR and other logical operations, flip-flops, registers, and storage elements.

[0132] Computer-readable storage media can include any tangible device capable of storing instructions executable by a suitable device. Consequently, a computer-readable storage medium having instructions stored therein will possess an executable product comprising instructions for creating a scheme to perform operations specified in a flowchart or block diagram. Examples of computer-readable storage media include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable storage media include floppy disks (registered trademark), magnetic disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact optical disc read-only memory (CD-ROM), digital versatile optical disc (DVD), Blu-ray disc (registered trademark), memory sticks, integrated circuit cards, etc.

[0133] Computer-readable instructions may include any of the source code or object code described in any combination of one or more programming languages, including assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or object-oriented programming languages ​​such as Smalltalk (registered trademark), JAVA (registered trademark), C++, and existing procedural programming languages ​​such as the "C" programming language or similar programming languages.

[0134] Regarding computer-readable instructions, these instructions are executed to generate a scheme for performing the operations specified in a flowchart or block diagram by a processor or programmable circuit of a general-purpose computer, special-purpose computer, or other programmable data processing device. These computer-readable instructions can be provided locally or via a wide area network (WAN) such as a local area network (LAN) or the Internet. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0135] Although the present invention has been described above using various embodiments, its scope is not limited to those described in the above embodiments. It will be apparent to those skilled in the art that various modifications or alterations can be made to the above embodiments. As will be clear from the claims, such modifications or alterations can also be included within the scope of the present invention.

[0136] It should be noted that the execution order of actions, processes, steps, and stages in the apparatus, systems, programs, and methods shown in the claims, description, and drawings can be implemented in any order, unless specifically stated as "before," "prior to," etc., and the output of the previous process is not used for the subsequent process. Even if the flow of actions in the claims, description, and drawings is described using terms such as "firstly," "next," etc., for convenience, it does not mean that they must be implemented in that order.

[0137] Explanation of reference numerals in the attached figures 10 Information providing device 11 Ministry of Communications 12 Storage Department 13 Control Department 31 Acquisition Department 32 Judgment Department 33 Collection Department 34 Generation Department 35. Supply Department 36. Payment Department 121 User Information 122 Standard Model Information 123 Advanced Model Information 124 Real-time model information.

Claims

1. An information providing device, characterized in that, It has a collection section and a generation section. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The generation unit generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user.

2. The information providing device according to claim 1, characterized in that, It also has a judgment department. The determination unit determines whether the credibility-related conditions are met based on the information provided by the provider.

3. The information providing device according to claim 2, characterized in that, The determination unit determines whether the trustworthiness-related conditions are met based on the provider's attribute information.

4. The information providing device according to claim 3, characterized in that, The determination unit determines whether the credibility-related conditions are met based on the provider's business content.

5. The information providing device according to claim 3, characterized in that, The determination unit determines whether the credibility-related conditions are met based on the provider's business scale.

6. The information providing device according to claim 1, characterized in that, The provider is the manufacturer of the machine that provides the data source.

7. The information providing device according to claim 6, characterized in that, The machine in question is a user terminal.

8. The information providing device according to claim 6, characterized in that, The machine in question is a vehicle.

9. The information providing device according to claim 1, characterized in that, It also has a payment department. The payment department pays a fee to the provider of the data source when it provides the user with the answer related to the data source.

10. The information providing device according to claim 1, characterized in that, The generation unit determines the credibility of the data source collected from the provider, and determines whether it is used for learning the language model based on the credibility.

11. An information providing method performed by an information providing device, characterized in that, This includes the collection process and the generation process. The collection process gathers data from providers that have met and whose credibility-related conditions have been met. The generation process generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user.

12. An information provider, characterized in that, The computer performs the collection and generation steps. The collection step gathers the provided data sources from providers who have met and whose credibility-related conditions have been met. The generation step generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user.

13. An information providing device, characterized in that, It has a collection department, a judgment department, a generation department, and a payment department. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The determination unit determines the credibility of the collected data sources. The generation unit generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user. When the payment department provides a user with the answer related to the data source, it pays the provider of the data source a reward based on the credibility of the answer.

14. An information providing device, characterized in that, It has a collection section, a decision section, and a generation section. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The determination unit determines the credibility of the collected data sources. The generation unit generates a language model, which learns from the data source whose credibility meets predetermined conditions to generate an answer based on the prompt words input by the user.

15. An information providing device, characterized in that, It has a collection section and a generation section. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The generation unit generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user. The generation unit receives from the user a designation of the provider that is prohibited from being used for language model learning, and generates the language model based on the data source of each user learning from the provider other than the designated provider.

16. An information providing device, characterized in that, It has a collection section and a generation section. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The generation unit generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user. The generation unit receives from the user a designation that the provider is prohibited from being used in the language model learning process, and adjusts the learning amount of each provider in the language model based on the received designation. The language model is used to provide the answer to users other than the user.

17. An information providing device, characterized in that, It has a collection section and a generation section. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The generation unit generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user. The generation unit adjusts the learning amount of each provider in the language model based on the user's utilization of the services provided by the provider.

18. An information providing device, characterized in that, It has a collection section and a generation section. The collection unit collects the provided data sources from providers who have met the conditions related to credibility. The generation unit generates a language model, which learns from the collected data sources to generate responses based on prompts input by the user. The data source includes the publishing data published by the target users and the comment data submitted by the target users in response to the publishing data.

Images