Information processing apparatus

Abstract

An information processing apparatus configured to provide information to a user via a natural language dialogue service includes a control unit configured to: extract behavioral information related to a planned behavior expected to be performed by the user with a probability exceeding a predetermined value in the user's behavior; execute at least a portion of a process for generating a response corresponding to a planned task related to the planned behavior based on the behavioral information before the expected timing of the planned behavior execution, and store the execution result obtained by executing at least a portion of the process in a storage unit; and when the user requests a given task related to the planned task, generate a response corresponding to the given task using the execution result stored in the storage unit, and output the generated response to the user.

Description

Technical Field

[0001] This disclosure relates to information processing devices. Background Technology

[0002] When an agent provides information to a user, it performs processes such as information retrieval, information analysis, and response generation. Using high-performance engines for information analysis and response generation enables high-speed processing, but it often involves relatively high costs. For example, Japanese Unexamined Patent Application Publication No. 2016-9091 (JP 2016-9091 A) discloses a technique for generating response text based on a trade-off between response time and the reliability of response content. In this technique, response text from a high-speed, low-reliability engine is first output, followed by response text from a low-speed, high-reliability engine. Summary of the Invention

[0003] The methods used in the prior art are not designed to reduce costs. Furthermore, since the response text generated in response to a request is output twice, the generation of the first response text may ultimately be wasted.

[0004] This disclosure aims to balance response speed with processing cost when generating responses to provide information to users, and to minimize the cost associated with generating responses.

[0005] One aspect of this disclosure is an information processing apparatus configured to provide information to a user via a natural language dialogue service. The information processing apparatus includes a control unit configured to: extract behavioral information related to a planned behavior expected to be performed by the user with a probability exceeding a predetermined value in the user's behavior; execute at least a portion of a process for generating a response corresponding to a planned task related to the planned behavior based on the extracted behavioral information before the expected timing of the planned behavior, and store the execution result obtained by executing at least a portion of the process in a storage unit; and when the user requests a given task related to the planned task, generate a response corresponding to the given task using the execution result stored in the storage unit, and output the response corresponding to the given task to the user.

[0006] Other solutions provide a method executed by the aforementioned information processing apparatus, a program for causing a computer to execute the method, and a computer-readable storage medium for storing the program in a non-transitory manner.

[0007] This disclosure can balance response speed and processing cost when generating a response to provide information to a user, and minimize the cost associated with generating the response. Attached Figure Description

[0008] The features, advantages, and technical and industrial significance of exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, wherein like reference numerals denote like elements, and wherein: Figure 1 An outline of the information provision system according to this disclosure is shown; Figure 2 The hardware configuration of the vehicle-mounted device is shown; Figure 3 The software configuration of the vehicle-mounted device is shown; Figure 4 This is the first flowchart related to the response processing of tasks performed by the onboard device; Figure 5 This is a second flowchart related to the response processing of tasks performed by the onboard device; and Figure 6 This demonstrates the coordination between software modules of the onboard unit during task response processing. Detailed Implementation

[0009] In recent years, advances in machine learning have increased the number of products equipped with language models. For example, natural language dialogue capabilities can be added to target products by incorporating large language models (LLMs) into them, whose accuracy has been improved through machine learning using large-scale datasets. One area where natural language dialogue is useful is the automotive industry. For example, by equipping in-vehicle devices with LLMs, information can be obtained without operating touch panels or similar devices. This functionality is particularly useful when the user's hands are occupied, such as while driving. For example, natural language dialogue capabilities can be installed in embedded devices in vehicles, such as in-vehicle devices. However, given the data size and cost, a configuration that accesses the LLM via a network is preferred. Accessing the LLM via a network enables the use of large-scale language models that are difficult to install in local storage.

[0010] With improvements in this natural language dialogue capability, various types of information can be provided to users. To achieve appropriate information delivery to users, it is desirable to respond fluently to their requests for task execution, that is, to respond to requests as quickly as possible. On the other hand, achieving high-speed responses often increases the load on the device performing information processing to provide information, which may lead to higher costs.

[0011] To provide users with appropriate information, it is desirable to respond as quickly as possible and minimize associated costs. However, this problem has not been adequately addressed in conventional technologies. The information processing apparatus according to this disclosure solves this problem.

[0012] An information processing apparatus according to one aspect of this disclosure is configured to provide information to a user via a natural language dialogue service. The information processing apparatus includes a control unit configured to: extract behavioral information related to a planned behavior expected to be performed by the user with a probability exceeding a predetermined value in the user's behavior; execute at least a portion of a process for generating a response corresponding to a planned task related to the planned behavior based on the behavioral information before the expected timing of the planned behavior execution, and store the execution result obtained by executing at least a portion of the process in a storage unit; and when the user requests a given task related to the planned task, generate a response corresponding to the given task using the execution result stored in the storage unit, and output the response corresponding to the given task to the user.

[0013] Information can be provided to users in various forms. For example, an external agent can be used, positioned outside the information processing device. This external agent can be, for example, a system capable of performing dialogues in natural language and providing services using language models, each dedicated to a specific domain. The language model can be a large language model (LLM). A large language model is, for example, a language model trained to perform dialogue tasks in natural language. Examples of domains handled by such external agents include directions, local information guidance, vehicle operation guidance, tourist guidance, and everyday conversations. These external agents are typically configured to be remotely accessible via a network.

[0014] Alternatively, information can be provided to the user by using a local agent included in the information processing device. The language model included in the local agent is a language model capable of performing dialogue tasks in natural language, but unlike external agents, it can be a language model that has not been trained to be specific to any particular domain and can be generated at a relatively low cost.

[0015] The information processing apparatus of this disclosure can use either an external agent or a local agent. In either case, there is a trade-off between the speed of response to a task request and the cost of generating the response. Therefore, for actions expected to be performed by the user with a relatively high probability (the planned actions described above), the control unit of the information processing apparatus performs predetermined preprocessing of the response to the task associated with such action, taking into account the high probability. Then, when the task request is actually received from the user, the result of the preprocessing is used to reduce the load associated with generating the response to the task (i.e., the response used to provide information to the user). As a result, a balance is achieved between the speed of response to the task and the associated cost, thereby achieving appropriate information delivery.

[0016] In other words, for preprocessing purposes, the control unit extracts behavioral information related to the user's planned behavior. Planned behavior refers to the behavior the user is expected to perform from a probabilistic perspective. In this case, the user is more likely to need various types of information about the planned behavior, and it can be said that the probability of the user issuing an information provision request to the information processing device (i.e., a request for a planned task related to the planned behavior or a task that has some correlation with the planned task) is relatively high. Therefore, based on the extracted behavioral information, the control unit performs predetermined preprocessing before the user actually makes a task request. Predetermined preprocessing is at least a part of, and can be, a process for generating a response corresponding to the planned task. Alternatively, at least a part of the process can be a process for retrieving and collecting information for generating a response corresponding to the planned task. The result of this preprocessing is stored in a storage unit and used for subsequent information provisioning processing, i.e., generating a response when a task request is actually received from the user.

[0017] Here, the task actually requested by the user (the given task) may not be consistent with the planned task used in preprocessing. However, if the requested task is related to the planned task used in preprocessing, it is assumed that the result obtained through preprocessing (the execution result described above) can be used. Therefore, when a given task related to a planned task is requested, the control unit generates a response corresponding to the given task using the execution result stored in the storage unit and outputs the generated response to the user. Even when the actually requested given task is consistent with the planned task used in preprocessing, the control unit can similarly use the execution result to generate a response. In this way, by performing the processing for response generation in advance, the processing load when making the actual task request can be reduced, thereby achieving a balance between response speed and processing cost, and minimizing the cost associated with generating the response.

[0018] In the following description, specific embodiments of the present disclosure will be illustrated with reference to the accompanying drawings. Unless otherwise stated, the hardware configurations, module configurations, and functional configurations described in the embodiments are not intended to limit the scope of the present disclosure to these configurations.

[0019] First Embodiment System Overview Reference Figure 1 The overall configuration of the provision system according to the first embodiment is described. The provision system of this embodiment includes an on-board device 10 installed in a vehicle 1. The on-board device 10 is capable of accessing an agent (external agent 30) that provides natural language information services via a network 20 (e.g., a mobile communication network).

[0020] The vehicle-mounted device 10 is installed in a networked vehicle capable of communicating with various devices wirelessly. The vehicle-mounted device 10 may include a data communication module (DCM) for connecting components of the vehicle 1 (e.g., electronic control units (ECUs) and vehicle terminals) to the network 20. In this embodiment, the vehicle-mounted device 10 is able to access the Internet via the network 20 and connect to an external agent 30 providing services.

[0021] Vehicle 1 (or components installed in vehicle 1) is able to provide various services by communicating with external devices via onboard unit 10. Examples of such services include navigation services, remote control services (e.g., remote air conditioning), onboard Wi-Fi (registered trademark) services, and emergency call services.

[0022] The in-vehicle device 10 also features voice input / output functionality and is capable of performing natural language dialogues with vehicle occupants. The in-vehicle device 10 includes a control unit 11 that performs voice input and output and voice recognition, as well as a local agent. The local agent includes a language model trained to enable natural language dialogue. For example, the local agent can provide pre-defined functions through dialogue, such as route search or vehicle operation (e.g., air conditioning). The local agent is also configured to provide natural language information services in the same manner as the external agent 30. The language model used by the local agent is stored in a storage device included in the in-vehicle device 10.

[0023] The size of the language model used to provide information services using natural language dialogue varies greatly depending on its function. For example, there are a wide variety of dialogue categories, including local information, travel guides, vehicle-related information (operator manuals), and everyday conversations, and to cover all these categories, it is necessary to prepare a language model trained on a large-scale dataset. The size of such a language model can range from several gigabytes to hundreds of gigabytes, making it difficult to install in an in-vehicle device 10 with limited storage capacity. Therefore, the information service functions using a large-scale language model are assigned to an external agent 30. The in-vehicle device 10 is equipped with a relatively small-scale language model, and the local agent uses this relatively small-scale language model to provide natural language information services in response to requests from a limited number of users.

[0024] In the description of this embodiment, expressions such as "accessing an external agent" and "connecting to an external agent" refer to accessing or connecting to an external device that provides services via an external agent. In this embodiment, each of the multiple services provided using a large-scale language model operates as an "external agent" on a corresponding one of the multiple external devices, and the in-vehicle device 10 is able to perform a dialogue by connecting to any of these external agents.

[0025] Hardware configuration Next, the hardware configuration of the apparatus that constitutes the providing system will be described. Figure 2 An example of the hardware configuration of the vehicle-mounted device 10 according to this embodiment is illustrated schematically.

[0026] The vehicle-mounted device 10 can be configured as a computer including a processor (central processing unit (CPU), graphics processing unit (GPU), etc.), main storage (random access memory (RAM), read-only memory (ROM), etc.) and auxiliary storage (erasable programmable read-only memory (EPROM), hard disk drive, removable media, etc.). The auxiliary storage stores the operating system (OS), various programs, various tables, etc. As described below, various functions (software modules) for a predetermined purpose can be implemented by executing programs stored in the auxiliary storage. However, some or all of the functions can alternatively be implemented as hardware modules using hardware circuitry such as application-specific integrated circuits (ASICs) or field-programmable gate arrays (FPGAs).

[0027] The vehicle-mounted device 10 includes a control unit 11, a storage unit 12, a wireless communication module 13, and an input / output unit 14 as hardware components.

[0028] The control unit 11 is a computing unit that executes predetermined programs to implement various functions of the vehicle-mounted device 10. The control unit 11 can be implemented by, for example, a hardware processor such as a CPU. The control unit 11 may include RAM, ROM, and cache memory.

[0029] Storage unit 12 serves as a device for storing information and is composed of a storage medium such as RAM, a hard disk, or flash memory. Storage unit 12 stores programs executed by control unit 11 and data used by the programs.

[0030] The wireless communication module 13 is a communication device for wirelessly communicating with a predetermined network. In this embodiment, the wireless communication module 13 is configured to communicate with the network 20, which is a mobile communication network. The wireless communication module 13 may include an embedded universal integrated circuit card (eUICC) (e.g., a Subscriber Identity Module (SIM) card). The SIM card is configured as a microcomputer including a CPU and storage devices. The SIM card stores information for connecting to the network 20 and receiving authentication.

[0031] The input / output unit 14 is a unit that receives input from the user of the device and presents information to the user. The input / output unit 14 typically includes devices for inputting and outputting audio, such as microphones and speakers. The input / output unit 14 may also include devices for providing visual information (displays, etc.).

[0032] Software Configuration Next, the software configuration of the devices that constitute the system will be described. Figure 3 The software configuration of the vehicle-mounted device 10 according to this embodiment is schematically illustrated. In this embodiment, the control unit 11 of the vehicle-mounted device 10 includes four software modules: a dialogue receiving unit 111, an agent unit 112, a route guidance unit 113, and a processing unit 114. Each software module can be implemented by the control unit 11 (e.g., a CPU) that executes a program stored in the storage unit 12. Information processing performed by the software modules is equivalent to information processing performed by the control unit 11 (e.g., the CPU).

[0033] The dialogue receiving unit 111 acquires speech delivered by a vehicle occupant (hereinafter also referred to as "user") via the input / output unit 14. The dialogue receiving unit 111 performs predetermined processing on the acquired audio data and performs speech recognition. The content of the speech is thus converted into text. The dialogue receiving unit 111 then sends the text obtained as a result of speech recognition to the agent unit 112. The dialogue receiving unit 111 also outputs the response sent from the agent unit 112 (hereinafter referred to as "response text") as audio or on a display.

[0034] Agent unit 112 provides information services. Specifically, agent unit 112 uses a language model stored in its own device (local language model 121, described later) to provide information services. Agent unit 112 can act as a virtual agent (local agent). The local agent can provide services that can be completed within its own device (e.g., everyday conversations). When the conversation involves specialized knowledge, agent unit 112 sends the utterance text to one or more external agents 30 available via network 20, obtains the results, and provides them to the user, thereby providing information services. For example, when it is determined that the user is looking for information about vehicle 1, agent unit 112 selects an external agent 30 that can provide operating instructions for vehicle 1. When it is determined that the user is looking for information about facilities or shops, agent unit 112 selects an external agent 30 that can provide such information. The selection of external agents 30 can be based on the results of analyzing the content of the utterance. For example, agent unit 112 can calculate a relevance score for each external agent 30 based on the words or context contained in the utterance and send the utterance text to the external agent 30 with a relevance score greater than or equal to a predetermined value.

[0035] Furthermore, agent unit 112 uses the results of preprocessing performed by processing unit 114, as described later, to generate a response corresponding to a given task requested by the user. This response generation will be described in detail later.

[0036] The route guidance unit 113 receives a starting point and destination specified by the user, or receives a command (e.g., a command to set the location information of the destination recommended to the user) from the agent unit 112, and generates a route connecting the starting point and destination based on road map data, etc. The route guidance unit 113 also outputs information about the generated route via the input / output unit 14.

[0037] Processing unit 114 performs at least a portion of the processing for generating a response corresponding to a planned task associated with a planned behavior, which is the behavior that the user is expected to perform with a probability exceeding a predetermined value among the behaviors the user can take. The predetermined probability serves as a threshold for identifying planned behaviors to be preprocessed by processing unit 114 and can be set taking into account factors such as the processing load associated with processing unit 114. The preprocessing performed by processing unit 114 refers to the processing performed before agent unit 112 generates a response corresponding to the actual given task from the user, and is performed to generate data that can be used for generating that response.

[0038] Examples of planned behaviors include behaviors performed by the user at a relatively high frequency, and behaviors that the user is likely to perform as part of a future schedule, i.e., behaviors that the user is expected to perform with a probability exceeding a predetermined value. More specifically, examples of behaviors performed by the user at a relatively high frequency include behaviors that the user has performed more frequently than a predetermined frequency in the user's past behavior history. Processing unit 114 is capable of extracting behavioral information related to the planned behavior based on information about the user's behavior history included in attribute data 123, which will be described later. Examples of behaviors that the user is likely to perform as part of a future schedule include behaviors included in the user's schedule information. Processing unit 114 is capable of extracting information related to the user's behaviors included in the user's schedule information in attribute data 123 as behavioral information related to the planned behavior.

[0039] Processing unit 114 then executes at least a portion of the processing for generating a response corresponding to a planned task related to the planned action. The planned task is a task requested by the user when the user wants to perform a planned action. The planned task can be derived using a local agent included in the vehicle device 10, or it can be derived through rule-based processing. "At least a portion of the processing" executed by processing unit 114 refers to part or all of the processing for generating a response corresponding to the planned task via agent unit 112. For example, at least a portion of the processing may include processing for searching and collecting information for generating the response, or it may be the response generation processing itself based on the collected information.

[0040] At least a portion of the execution results of the processing performed by processing unit 114 are stored in storage unit 12. The stored execution results constitute preprocessed data 124, which will be described later. The processing performed by processing unit 114 will be described in detail later.

[0041] The onboard unit 10 includes a storage unit 12 that stores a local language model 121, agent information 122, attribute data 123, and preprocessed data 124. The local language model 121 is a language model trained to perform dialogue tasks in natural language. The local language model 121 is a language model generated at a relatively low cost and can be, for example, a language model used for general conversations. Compared to large-scale language models used by external agents accessible via a network, the local language model 121 is a lightweight language model. Therefore, external and local agents can be used selectively. For example, when the conversation involves specialized knowledge, the utterance text is sent to an external agent, while utterances in a conversation where a fast response is desired can be processed by a local agent.

[0042] Agent information 122 is a collection of data related to multiple external agents available to the vehicle-mounted device 10. Agent information 122 may include, for example, identifiers and names of external agents, information about the scope specific to the external agents, and information about their access destinations (such as the network address of the external device providing the external agent). By referring to agent information 122, agent unit 112 is able to determine the external agent to which the spoken text will be sent.

[0043] Attribute data 123 is a collection of data related to the attributes of a user interacting with the vehicle-mounted device 10. Attribute data 123 may include, for example, personal information such as the user's gender, age, height, and weight. Attribute data 123 may also include information about the user's family structure. Attribute data 123 may also include information about the user's past behavioral history and schedule information. Past behavioral history may be accumulated in the attribute data based on task requests made by the user and corresponding responses provided via the dialogue receiving unit 111 in the vehicle-mounted device 10. Alternatively, the vehicle-mounted device 10 may access an external server or mobile terminal (such as a smartphone owned by the user) that stores the user's behavioral history, obtain behavioral history information from the external server or mobile terminal, and store it in the attribute data. Regarding schedule information, the vehicle-mounted device 10 may access an external server or mobile terminal (such as a smartphone owned by the user) that stores the user's schedule information, obtain schedule information from the external server or mobile terminal, and store it in the attribute data. When multiple users exist in the vehicle, attribute data 123 may also include data for each user.

[0044] The preprocessed data 124 is data related to the execution result of at least a portion of the processing performed by the processing unit 114 as described above. Therefore, the preprocessed data 124 is appropriately updated based on the preprocessing performed by the processing unit 114.

[0045] The process of providing information to users Figure 4 This is a flowchart illustrating the first stage of the process for providing information to a user. The processing in the first stage mainly corresponds to the preprocessing performed by the processing unit 114. Figure 5 This is a flowchart illustrating the second phase of the process for providing information to the user. The processing in the second phase mainly corresponds to the response generation process performed by the agent unit 112. Figure 6 The coordination between software modules of the control unit 11 of the on-board unit 10 during these processing flows is illustrated. The processing flow performed by the control unit 11 will now be described with reference to these figures.

[0046] First, the description Figure 4 The processing flow is shown. This processing can be performed each time the attribute data 123, which is user information, is updated, or it can be repeated at predetermined intervals. In step S101, processing unit 114 extracts behavioral information related to the user's planned behavior. A planned behavior can be an action that the user performs more frequently than a predetermined frequency in the user's past behavioral history. For example, if the user's behavioral history indicates that the user checks the news more frequently at 7 a.m. than a predetermined frequency, then "checking the news at 7 a.m." can be a planned behavior. Processing unit 114 is able to extract behavioral information related to such planned behaviors based on the user's behavioral history information contained in attribute data 123. Alternatively, planned behaviors can be behaviors contained in the user's schedule information. In particular, behaviors such as family-related activities (e.g., family birthday parties or family trips) or business-related matters (e.g., business trips or dining with business partners) can be considered more important planned behaviors for the user. For example, if a "family birthday party" is planned according to the user's schedule information, then "family birthday party" can be a planned behavior. Processing unit 114 can extract information about user behavior included in the schedule information contained in attribute data 123 as behavioral information related to planned behavior. After step S101 is completed, the process proceeds to step S102.

[0047] In step S102, it is determined whether the time has arrived for preprocessing to be performed by processing unit 114. As described above, a planned behavior is an action that is expected to be performed by the user with a probability exceeding a predetermined value, and the timing when the behavior can be performed can be anticipated. For example, for the planned behavior of "checking the news at 7 a.m.", the execution timing will be 7 a.m., and for the planned behavior of "family birthday party", the execution timing will be scheduling the date and time of the birthday party. If the preprocessing performed by processing unit 114 is performed too early, information may be retrieved and collected at a time too far from the execution timing of the planned behavior, and from the perspective of generating data to be used to generate a response to the user's task request, the processing may not be considered appropriate preprocessing. Therefore, the determination in step S102 is performed to determine whether the appropriate time for preprocessing has arrived. The timing for performing preprocessing can be determined based on the content of the planned behavior. For example, when the planned behavior is an action that can be performed daily, the timing for performing preprocessing may be, for example, one hour before the planned behavior. When the planned behavior is performed once a year, the timing for performing preprocessing may be, for example, one week before the planned behavior. If a positive decision is made in step S102, the process proceeds to step S103. If a negative decision is made in step S102, step S102 is repeated.

[0048] In step S103, since the time for performing preprocessing has arrived, processing unit 114 performs preprocessing via agent unit 112 to generate data that can be used to generate a response to a future given task from the user; that is, information retrieval and information collection performed while preparing for the user's future given task. Specifically, under the premise that the planned behavior is somewhat related to the future given task from the user, in S103, processing related to generating a response corresponding to the planned task associated with the user's planned behavior is performed based on the extracted behavioral information. For example, when the user's planned behavior is "checking the news at 7 a.m.," the planned task could be "collecting information about Japanese politics and economics," as determined based on the user's attributes (such as occupation). Alternatively, when the user's planned behavior is "family birthday party," the planned task could be "searching for candidate gifts for children," as determined based on the user's attributes (such as family structure).

[0049] During the processing in step S103, the user's future given task cannot be accurately predicted. Therefore, in step S103, processing for preparing the future given task is performed using behavioral information related to the planned behavior. Specifically, processing unit 114 passes the planned task to agent unit 112 and obtains the response to the planned task as the preprocessing execution result. Upon receiving the planned task, agent unit 112 identifies its content, obtains agent information 122 about the external agent from storage unit 12, and determines the external agent 30 to which the planned task will be sent using the agent information 122. The content of the planned task can be determined using the local language model 121 or another machine learning model. Alternatively, the content of the planned task can be determined through rule-based processing. Agent unit 112 may also determine that it is not necessary to send the planned task to the external agent. In this case, the planned task can be processed by agent unit 112 (and local language model 121). Subsequently, in step S104, processing unit 114 stores the obtained execution result as preprocessing data 124 in storage unit 12.

[0050] The above describes Figure 4 The processing flow is shown below.

[0051] Next, we will describe Figure 5 The processing flow is shown below. For example, this processing is performed when a user requests a given task. In step S201, a request to perform a given task related to information provision is received from the user riding in vehicle 1. This processing is performed by the dialogue receiving unit 111. Specifically, the dialogue receiving unit 111 obtains the utterance related to the request from the user through the input / output unit 14. For example, the input / output unit 14 converts the utterance obtained through a microphone, etc., into audio data, and the dialogue receiving unit 111 obtains the audio data. The dialogue receiving unit 111 performs predetermined speech recognition processing on the obtained audio data and converts the audio data into text. While retaining the response to the utterance, the dialogue receiving unit 111 sends the text obtained through speech recognition to the agent unit 112. This text is referred to as "utterance text".

[0052] In step S202, it is determined whether the given task received in step S201 is a task related to the planned task corresponding to the preprocessed data 124. In other words, it is determined whether the preprocessed data 124 can be used to generate a response to the given task. Specifically, the agent unit 112 can calculate the correlation score between the given task and the planned task, and when the correlation score is greater than or equal to a predetermined value, it can make a positive determination in step S202. When a positive determination is made in step S202, the process proceeds to step S203. When a negative determination is made in step S202, the process proceeds to step S205.

[0053] In response to the affirmative determination in step S202, agent unit 112 acquires preprocessed data 124 in step S203. Subsequently, in step S204, agent unit 112 generates a response to a given task from the user using the preprocessed data 124. Specifically, agent unit 112 performs information provision processing based on the utterance text related to the given task sent by dialogue receiving unit 111 and the preprocessed data. More specifically, upon receiving the utterance text and preprocessed data, agent unit 112 can generate a response to the given task using the same agent as the external agent performing the processing related to the planned task (i.e., the processing in step S103). The generated response is sent to agent unit 112 as response text, and agent unit 112 outputs the response as a response to the given task to the user (i.e., provides information) through dialogue receiving unit 111. Specifically, dialogue receiving unit 111 generates audio data based on the response text and outputs the audio data through input / output unit 14. Dialogue receiving unit 111 can generate audio data for reading out the response text, or it can generate information accompanying the audio data. Examples of information accompanying the audio data include user interface screens. User interface screens may include response text in text form.

[0054] In response generation in step S204, preprocessed data (information collected for the planned task) is used. Therefore, in step S204, it is not necessary to collect the information contained in the response. This significantly reduces the processing load associated with generating the response. In other words, the processing for generating a response to a user's given task can be distributed more efficiently. Figure 4 The processing in the first stage shown and Figure 5 The second stage of processing is shown. This reduces the load associated with generating a response to a given task and makes it possible to provide a relatively fast response to the user while keeping the required performance of the large language model (LLM) installed in the external agent 30 and the local language model 121 at a low level. In other words, this allows for a balance between response speed and processing cost and minimizes the cost associated with response generation.

[0055] In response to the negative decision in step S202, agent unit 112 generates a response to the given task from the user in step S205. In this case, preprocessed data 124 is not used. Agent unit 112 performs information retrieval and information collection to generate a response to the given task. Thereafter, agent unit 112 uses the collected information to generate a response and outputs the response to the user. The output to the user is as described above. When the processing in step S205 is performed, no processing is allocated for generating the response, thus not reducing the processing load.

[0056] As mentioned above, Figure 4 and Figure 5 The processing shown can balance response speed and processing cost when generating a response to provide information to the user, and minimize the cost associated with generating the response.

[0057] exist Figure 5 In step S204, response generation is performed using preprocessed data 124. However, response generation can alternatively be performed by performing additional information retrieval and information collection to generate a response corresponding to a given task, and using both preprocessed data 124 and additional information to generate the response. Such additional information collection makes the content of the response more suitable for output to the user. However, given the purpose of generating preprocessed data 124, it is undesirable to increase the processing load associated with such additional information collection. Therefore, when performing additional information collection, the scope of information to be collected can be limited to reduce the processing load. For example, processing such as information collection can be performed on new information that appears during a period after the timing of preprocessing (i.e., after the processing of steps S103 and S104).

[0058] Variation Example The above embodiments are merely examples, and modifications can be made to this disclosure as appropriate without departing from its spirit and scope. For example, the processes and means described in this disclosure can be combined as needed, provided there are no technical inconsistencies. In a variation of this disclosure, the information processing apparatus can be configured as a server device installed in a fixed manner, rather than as an in-vehicle device installed in a vehicle.

[0059] The above embodiments illustrate an example of deploying an external agent on an external device connected to the Internet. However, the external agent can alternatively be deployed on other types of devices. For example, the external agent can be deployed on an edge server accessible from the vehicle. The external agent can also be distributed across multiple edge servers or cloud systems. Although the embodiments show the in-vehicle device 10 as an information processing apparatus according to this disclosure, the information processing apparatus can alternatively be implemented as a device or equipment not installed in a vehicle.

[0060] When an external agent is deployed across multiple edge servers, the vehicle-mounted device 10 can obtain its own vehicle location information and identify the edge server on which the desired external agent is deployed based on that location information. For example, there may be a situation where an external agent providing local information can be deployed on multiple edge servers located in different regions. In this case, the vehicle-mounted device 10 can identify the external agent deployed on the geographically nearest edge server based on the vehicle's location information. The local information processed by the external agent may vary depending on the edge server. For example, an external agent providing local information for region A may be deployed on an edge server located in region A, and an external agent providing local information for region B may be deployed on an edge server located in region B. The vehicle-mounted device 10 can obtain the local information of the region where its vehicle is located by identifying the edge server corresponding to that region and accessing the external agent provided by that edge server.

[0061] Processes described as being performed by a single device can be divided among multiple devices. Processes described as being performed by different devices can be performed by a single device. In computer systems, the hardware configuration (server configuration) used to implement each function can be flexibly modified.

[0062] This disclosure also enables implementation by providing a computer with a computer program that performs the functions described in the above embodiments, and enabling one or more processors of the computer to read and execute the program. Such a computer program can be provided to the computer via a non-transitory computer-readable storage medium that can be connected to the computer's system bus, or it can be provided to the computer via a network. Examples of non-transitory computer-readable storage media can include any type of disk, such as a magnetic disk (floppy disk (registered trademark), hard disk drive (HDD), etc.), an optical disk (CD-ROM, DVD, Blu-ray disc, etc.), a read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), a magnetic card, flash memory, an optical card, and any other type of medium suitable for storing electronic instructions.

Claims

1. An information processing apparatus configured to provide information to a user via a natural language dialogue service, the information processing apparatus comprising a control unit configured to: Extract behavioral information related to planned behaviors that are expected to be performed by the user with a probability exceeding a predetermined value, based on the user's behavior. Before the expected timing for executing the planned action, at least a portion of a process for generating a response corresponding to a planned task related to the planned action based on the extracted action information is executed, and the execution result obtained by executing the at least a portion of the process is stored in a storage unit. When the user requests a given task related to the scheduled task, a response corresponding to the given task is generated by using the execution result stored in the storage unit, and the response corresponding to the given task is output to the user.

2. The information processing apparatus according to claim 1, wherein the control unit is configured to: extract information related to behaviors that the user has performed more frequently than a predetermined frequency in the user's past behavior history as the behavior information.

3. The information processing apparatus according to claim 1, wherein the control unit is configured to: extract information related to the user's behavior included in the user's schedule information as the behavior information.

4. The information processing apparatus according to any one of claims 1 to 3, wherein the control unit is configured to: as at least part of the processing, at least execute information based on the extracted behavioral information for generating the response corresponding to the planned task, and store the collected information as the execution result in the storage unit.

5. The information processing apparatus of claim 4, wherein the control unit is configured to: when the user requests the given task, collect additional information for generating the response corresponding to the given task, use the execution result stored in the storage unit and the additional information to generate the response corresponding to the given task, and output the response corresponding to the given task to the user.

Images