Natural language generation device, natural language generation method, and program

Abstract

To provide technology that enables more efficient modification of dialogue systems that generate responses based on information provided by users. [Solution] An information processing device that updates constraint information affecting the answers generated in a dialogue with a user, which is a dialogue interface that generates answers based on information provided by a user, comprising: an extraction unit that extracts areas that need improvement based on a record of the dialogue; a generation unit that receives input of knowledge information useful for improving the areas that need improvement and generates update instruction information for updating the constraint information based on the knowledge information; and an update unit that updates the constraint information based on the update instruction information.

Description

Technical Field

[0001] The present invention relates to the technology of natural language generation devices, natural language generation methods, and programs.

Background Art

[0002] Conventionally, research has been conducted on dialogue systems using large language models (LLMs). For example, Patent Document 1 describes a system that generates a dialogue model by performing machine learning using chat dialogue information as teacher data and answers questions from users using the dialogue model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the conventional technology, when the dialogue system cannot give an appropriate answer to a specific pattern, a lot of labor has been spent on modifying the dialogue system. For example, it takes time to collect failure cases for analysis. Also, for example, when modifying the dialogue model, it takes time because expertise is required for collecting learning data and re-learning the model. Also, expertise is required when modifying the external data sources used by the dialogue system, and it takes time to formulate a policy on how to modify. Thus, in the conventional technology, it may not be possible to efficiently modify a natural language-based dialogue system.

[0005] In view of the above circumstances, an object of the present invention is to provide a technology that enables more efficient modification of a dialogue system that generates answers based on information provided by a user.

Means for Solving the Problems

[0006] One aspect of the present invention is an information processing device for updating constraint information that affects the answers generated in a dialogue with a user, which is a dialogue interface that generates answers based on information provided by a user, and comprises: an extraction unit that extracts areas that need improvement based on a record of the dialogue; a generation unit that receives input of knowledge information useful for improving the areas that need improvement and generates update instruction information for updating the constraint information based on the knowledge information; and an update unit that updates the constraint information based on the update instruction information.

[0007] One aspect of the present invention is the above-described information processing device, wherein the extraction unit presents the extracted areas requiring improvement to the administrator of the dialogue device, and the generation unit receives input of the knowledge information from the administrator.

[0008] One aspect of the present invention is the above-described information processing apparatus, further comprising an input unit for inputting the knowledge information to the generation unit, wherein the extraction unit supplies the areas requiring improvement to the input unit, and the input unit extracts knowledge information useful for improvement from an external data source for the areas requiring improvement supplied from the extraction unit and inputs it to the generation unit.

[0009] One aspect of the present invention is the above-described information processing device, wherein the input unit is configured using a dialogue model that has been trained to respond to user input.

[0010] One aspect of the present invention is the above-described information processing device, further comprising a dialogue unit that provides an interface for dialogue with the administrator of the dialogue device, wherein the dialogue unit identifies the information necessary as knowledge information based on the areas to be improved and presents it to the administrator, receives input of the knowledge information from the administrator, and inputs the knowledge information entered by the administrator into the generation unit.

[0011] One aspect of the present invention is the above-described information processing device, wherein the dialogue unit is configured using a dialogue model that has been trained to respond to user input.

[0012] One aspect of the present invention is the above-described information processing device, wherein the generation unit modifies the generated update instruction information based on the feedback information when feedback information regarding modifications to the generated update instruction information is input by the administrator.

[0013] One aspect of the present invention is the above-described information processing apparatus, wherein the generation unit stores both first constraint information, which is constraint information before update, and second constraint information, which is constraint information after update based on the update instruction information modified based on the feedback information, and the dialogue unit presents to the administrator, in a comparable manner, the output when the same dialogue is performed with the first constraint information and the second constraint information, respectively.

[0014] One aspect of the present invention is the above-described information processing device, wherein the extraction unit and the generation unit are configured using a dialogue model that has been trained to respond to user input.

[0015] One aspect of the present invention is the above-described information processing device, wherein the generation unit performs preprocessing on the input knowledge information, such as formatting the input knowledge information or extracting key information for updating the constraint information, and inputs the preprocessed knowledge information into the dialogue model.

[0016] One aspect of the present invention is the above-described information processing device, wherein the extraction unit classifies the causes of the dialogue extracted as areas requiring improvement into categories corresponding to the policy for correcting the constraint information and presents them to the administrator.

[0017] One aspect of the present invention is the above-described information processing device, wherein the dialogue device interacts with the user through language-based dialogue information exchanged with the user and visual information supplementing the dialogue information, and the record of the dialogue includes historical information relating to the dialogue information and the visual information.

[0018] One aspect of the present invention is the above-described information processing device, wherein the dialogue interface is provided as a robot capable of physical movement, and the historical information relating to the visual information includes historical information relating to the physical movement performed by the robot.

[0019] One aspect of the present invention is the above-described information processing device, wherein the dialogue device includes a camera for reading the user's facial expressions, and the recording of the dialogue includes the dialogue information and historical information relating to the user's facial expressions captured by the camera or recognized from the image.

[0020] One aspect of the present invention is the above-described information processing device, wherein the dialogue device includes a microphone for sensing the sound environment of a space in which the user is interacting, and the recording of the dialogue includes the dialogue information and historical information relating to the sound environment to be captured by the microphone or recognized from the sound.

[0021] One aspect of the present invention is the above-described information processing device, wherein the recording of the dialogue includes the dialogue information and historical information relating to the user's voice or the user's evaluation of the dialogue device as recognized from the voice.

[0022] One aspect of the present invention is the above-described information processing device, wherein the record of the dialogue includes historical information relating to communication errors of the dialogue device.

[0023] One aspect of the present invention is an information processing method for updating constraint information that affects an answer generated by a dialogue device that provides a dialogue interface for a computer to generate an answer based on information provided by a user in a dialogue with the user, wherein the computer extracts points to be improved to be improved based on a record of the dialogue, receives an input of knowledge information useful for improving the points to be improved, generates update instruction information for updating the constraint information based on the knowledge information, and updates the constraint information based on the update instruction information.

[0024] One aspect of the present invention is a program for causing a computer to execute an update process for updating constraint information that affects an answer generated by a dialogue device that provides a dialogue interface for generating an answer based on information provided by a user in a dialogue with the user, wherein the update process includes a process of extracting points to be improved to be improved based on a record of the dialogue, a process of receiving an input of knowledge information useful for improving the points to be improved, generating update instruction information for updating the constraint information based on the knowledge information, and a process of updating the constraint information based on the update instruction information.

Advantages of the Invention

[0025] According to the present invention, it becomes possible to more efficiently perform the correction of a dialogue system that generates an answer based on information provided by a user.

Brief Description of the Drawings

[0026] [Figure 1] It is a diagram showing an example of the system configuration of the dialogue system 1 of the embodiment. [Figure 2] It is a diagram showing a configuration example of the dialogue device 100 of the embodiment. [Figure 3] It is a diagram (part 1) showing an example of the flow of a process for a dialogue device 100 of the embodiment to update constraint information of a dialogue model through a dialogue with a user. [Figure 4]This is a diagram (part 2) showing an example of the processing flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model through dialogue with the user. [Figure 5] Figure (3) shows an example of the processing flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model through dialogue with the user. [Figure 6] Figure (4) shows an example of the processing flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model through dialogue with the user. [Modes for carrying out the invention]

[0027] Embodiments of the present invention will be described in detail below with reference to the drawings.

[0028] Figure 1 shows an example of the system configuration of the dialogue system 1 of the embodiment. The dialogue system 1 is a system that receives natural language input from a user and outputs a response to that natural language input. The dialogue system 1 includes, for example, a dialogue device 100, a user terminal device 200, and an administrator terminal device 300. The dialogue device 100, the user terminal device 200, and the administrator terminal device 300 can communicate with each other via a network NW. The network NW may be a network using wireless communication or a network using wired communication. The network NW may be configured using, for example, the Internet or a local area network (LAN). The network NW may be configured by combining multiple networks.

[0029] The user terminal device 200 is a terminal device used by user U of the dialogue system 1. For example, the user terminal device 200 may be a smartphone, tablet, personal computer, or other terminal device. A user interface operates on the user terminal device 200. By operating the user interface, user U can input arbitrary natural language sentences to the dialogue device 100 and obtain responses to those natural language sentences from the dialogue device 100. The user interface may be a dedicated application program or a web application provided via a web browser.

[0030] The administrator terminal device 300 is a terminal device used by the administrator M of the dialogue system 1. For example, the administrator terminal device 300 may be a terminal device such as a smartphone, tablet, or personal computer. The administrator interface operates on the user terminal device 300. Administrator M can send and receive information with the dialogue system 100 by operating the administrator interface. The administrator interface may be a dedicated application program or a web application provided via a web browser.

[0031] Figure 2 shows an example configuration of the dialogue device 100 according to the embodiment. The dialogue device 100 comprises, for example, a communication unit 110, a storage unit 120, and a control unit 130. The control unit 130 is configured using, for example, a processor such as a CPU (Central Processing Unit) and memory. The control unit 130 functions as a dialogue unit 131, an improvement area extraction unit 132, an update instruction generation unit 133, and an update execution unit 134 when the processor executes a program. Note that all or part of the functions of the control unit 130 may be implemented using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The above program may be recorded on a computer-readable recording medium. Computer-readable recording media include, for example, portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, semiconductor storage devices (e.g., SSDs: Solid State Drives), and storage devices such as hard disks and semiconductor storage devices built into computer systems. The above program may be transmitted via a telecommunications line.

[0032] The communication unit 110 is a communication device that connects the dialogue device 100 to the network NW. The communication unit 110 is, for example, a network interface. The network interface may be a device that performs wireless communication or a device that performs wired communication.

[0033] The storage unit 120 is configured using a storage device such as a magnetic hard disk drive or a semiconductor storage device. The storage unit 120 stores, for example, an interaction log 121, an update patch 122, and a replacement file 123. Details of the interaction log 121, the update patch 122, and the replacement file 123 will be described later.

[0034] The control unit 130 acquires information in natural language entered by the user into the user terminal device 200 from the user terminal device 200, generates a response to the natural language, and sends it back to the user terminal device 200. More specifically, the control unit 130 comprises a dialogue unit 131, an improvement area extraction unit 132, an update instruction generation unit 133, and an update execution unit 134.

[0035] The dialogue unit 131 performs dialogue processing that generates a response to natural language text supplied from the user terminal device 200 and transmits it to the user terminal device 200. For example, the dialogue unit 131 may be composed of a large language model (LLM) that has been trained to respond to input natural language text. For example, GhatGPT®, Claude®, Gemini®, etc., may be used as the LLM. In this case, the dialogue unit 131 can obtain a response to the natural language text as its output by inputting the supplied natural language text into the LLM.

[0036] More specifically, the LLM of the dialogue unit 131 is subjected to prompt engineering (for example, a technique called "Policy Prompt") which controls the output of the LLM based on constraint information such as specific policies, rules, and knowledge. In this embodiment, it is assumed that the constraint information is stored in an external data source. One technique that allows the LLM to refer to an external data source is called "RAG (Retrieval-Augmented Generation)". In the dialogue device 100 of this embodiment, the constraint information is stored in an external data source and made accessible to the LLM. As a result, the dialogue device 100 of this embodiment can be tuned without retraining the LLM by changing the constraint information stored in the external data source.

[0037] In addition to the above, the dialogue unit 131 stores information related to the dialogue process as a dialogue log 121 in the storage unit 120 so that it can be referenced later. For example, the dialogue log 121 includes natural language input from the user terminal device 200 and the content of the dialogue unit 131's response to that natural language input. Furthermore, the dialogue log 121 may also include a history of data processing performed in the process of generating a response to the input natural language input, and a history of referencing the external data source 40.

[0038] The Improvement Area Extraction Unit 132 extracts areas that need improvement (areas requiring improvement) from the content of the dialogue recorded in the dialogue log 121. Areas requiring improvement can be defined as cases where the dialogue failed (failed cases) among the group of dialogue cases that took place with the user. For example, the Improvement Area Extraction Unit 132 can be configured using an LLM. In this case, for example, the Improvement Area Extraction Unit 132 can obtain the areas requiring improvement for the dialogue as output by inputting the dialogue log 121 and natural language instructing the LLM to extract areas requiring improvement from the content of the dialogue log 121.

[0039] For example, the improvement area extraction unit 132 inputs the dialogue log 121 into the LLM and inputs instructions to the LLM such as, "From the input dialogue log, extract the cases where location guidance failed." For example, if the dialogue log records an exchange where the question "Where is the restroom?" is answered with "It's at the south end of the second floor of the main building," the LLM is expected to determine that an appropriate dialogue took place because a location was answered to the question "Where?". On the other hand, if the dialogue log records an exchange where the same question is answered with "You can see it if you turn around," the LLM may determine that an appropriate dialogue did not take place because a location was not answered to the question "Where?". In this way, it is expected that the LLM may recognize that an appropriate dialogue needs improvement even if it is context-dependent.

[0040] In anticipation of such cases, the improvement area extraction unit 132 may be configured to determine the areas requiring improvement using a rule-based method as described below. (1) Cases where the dialogue time is too long (exceeding the threshold time) are identified as failure cases. (2) Cases in which user U repeatedly asks questions of the same nature are identified as failure cases. (3) Cases in which the dialogue was terminated midway (for example, cases in which there were no words indicating the end of the dialogue, such as "thank you" or "see you later") are identified as failure cases. The improvement area extraction unit 132 may also extract areas requiring improvement by instructing the LLM to extract failure cases using the rule-based method described above.

[0041] Furthermore, the improvement area extraction unit 132 may be configured to extract the dialogue that was taking place at the time as a failure case when it detects a failure in internal processing related to the dialogue. For example, if the dialogue system 1 has a function that allows the LLM to call and execute an external function, such as the "Function Calling" function of ChatGPT (registered trademark), the dialogue unit 131 may be configured to save a communication log with the external function in addition to the dialogue log 121. In this case, the improvement area extraction unit 132 may be configured to determine the success or failure of the call to the external function based on the communication log and to extract the dialogue when the call to the external function failed (a request error occurred) as a failure case.

[0042] In addition to these methods, the improvement area extraction unit 132 may be configured to determine whether a dialogue is a failure case by inputting the dialogue log of the dialogue to be judged into a trained model that has been trained by machine learning to take the dialogue log of each dialogue as input and output whether the dialogue is a success or failure.

[0043] The update instruction generation unit 133 generates instructions (update instructions) to update constraint information stored in an external data source. More specifically, the update instruction generation unit 133 receives input of knowledge information (knowledge information) that serves as material for LLM to determine how to improve the parts of the dialogue that need improvement extracted by the part extraction unit 132, and generates update instructions based on that knowledge information.

[0044] For example, the update instruction generation unit 133 generates an alternative file 123 containing updated constraint information and an update patch 122, which is a program for updating the constraint information stored in the external data source to the content described in the alternative file, as update instructions. For example, if a dialogue scene that failed to properly guide the user to the location of the toilets on campus is extracted as an area requiring improvement, text information such as "The toilets are located at the south end of the second floor of the main building. Please generate an update patch and an alternative file to add this knowledge to the external data source" is input to the update instruction generation unit 133. In this case, for example, the update instruction generation unit 133 recognizes the content of the knowledge to be added based on the text information and generates an alternative file 123 containing that knowledge and an update patch 122 that adds the knowledge from the alternative file 123 to the external data source. The update instruction generation unit 133 then stores the generated alternative file 123 and update patch 122 in the storage unit 120.

[0045] The update execution unit 134 updates the constraint information stored in the external data source based on the update instructions generated by the update instruction generation unit 133. For example, the update execution unit 134 updates the constraint information to the contents described in the alternative file 123 by executing the update patch 122.

[0046] [First Embodiment] Figure 3 is a diagram (part 1) showing an example of the process flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model (LLM) through dialogue with the user. In Figure 3, S101 to S105 represent the phase in which the user U and the dialogue device 100 interact (hereinafter referred to as the "dialogue phase"), and S201 to S206 represent the phase in which the constraint information stored in the external data source 400 is modified based on the content of the dialogue that took place in the dialogue phase to improve the dialogue function (hereinafter referred to as the "improvement phase").

[0047] [Dialogue Phase] In the dialogue phase, first, user U inputs natural language into the dialogue device 100 via the user terminal device 200 (S101). The dialogue unit 131 of the dialogue device 100 generates a response to the natural language input in S101 based on constraint information (policy 41 and knowledge 42) stored in the external data source 400 (S102). The dialogue unit 131 presents the response generated in S102 to user U (S104). Meanwhile, the dialogue unit 131 records information related to the dialogue processing in S102 in the dialogue log 121 (S105). Steps S101 to S105 may be executed repeatedly.

[0048] [Improvement Phase] Next, in the improvement phase, the improvement area extraction unit 132 first refers to the dialogue log 121 and extracts areas that appear to require improvement (areas requiring improvement) from the dialogues that took place during the dialogue phase (S201). The improvement area extraction unit 132 presents a list of the extracted areas requiring improvement (dialogue list) to the administrator M (S202). For example, the improvement area extraction unit 132 sends the dialogue list to the administrator terminal device 300 for display. Administrator M checks the contents of the dialogue list presented in S202 and inputs the knowledge information necessary to improve the listed dialogues into the dialogue device 100 (S203). The improvement area extraction unit 132 may also determine the knowledge information that is lacking to provide an appropriate answer to the dialogues in the areas requiring improvement (lack of knowledge information) and include that lack of knowledge information in the dialogue list.

[0049] Next, in the dialogue device 100, the update instruction generation unit 133 generates update instruction information INS (including update patch 122 and alternative file 123) to reflect the knowledge information entered in S203 into the constraint information (S204A). The update instruction information INS may reflect the knowledge information entered by administrator M into policy 41, into knowledge 42, or into both policy 41 and knowledge 42. The update instruction generation unit 133 determines, as appropriate, what information to update the constraint information to, according to the knowledge information entered by administrator M.

[0050] Furthermore, the update instruction generation unit 133 may be configured to refer to the current constraint information as needed when generating the update instruction information INS (S204B). For example, the update instruction generation unit 133 may consider the current constraint information and the new knowledge information entered by the administrator M to generate an alternative file 123 containing the information to be added to the current constraint information, or it may generate an alternative file 123 containing the current constraint information with the new knowledge information added.

[0051] Next, the update execution unit 134 updates the constraint information stored in the external data source 40 based on the update instruction information INS generated in S204A. For example, in this embodiment, the update execution unit 134 reflects the contents of the alternative file 123 in the constraint information by executing the update patch 122 (S205) (S206).

[0052] Once the constraint information update is complete, the dialogue unit 131 can interact with user U based on the updated constraint information. In the dialogue after the constraint information update, the knowledge 42 and policies 41 for improving the areas that need improvement identified in the previous improvement phase are stored as constraint information. Therefore, in the current dialogue phase, the dialogue unit 131 is expected to engage in a more favorable dialogue regarding the above-mentioned areas that need improvement than in the dialogue phase before the improvement. By repeating these dialogue phases and improvement phases, the dialogue system 1 is updated to enable smoother conversations with user U.

[0053] [Second Example] Figure 4 is a diagram (part 2) showing an example of the processing flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model (LLM) through dialogue with the user. The flow in Figure 4 differs from the case in Figure 3 in that, instead of the administrator M inputting knowledge information about areas requiring improvement to the update instruction generation unit 133, the dialogue device 100 inputs the knowledge information. That is, in this case, the control unit 130 of the dialogue device 100 is equipped with a knowledge information input unit 135 in addition to the configuration in Figure 2. The knowledge information input unit 135 may be composed of an LLM. In Figure 4, the same reference numerals are used for the same processing flow as in Figure 3, and the explanation of the same processing flow as in Figure 3 is omitted.

[0054] The knowledge information input unit 135 inputs a list of areas requiring improvement (dialogue list) extracted by the area requiring improvement extraction unit 132 in S202, and obtains knowledge information useful for improving the areas requiring improvement from the second external data source 50 based on this dialogue list (S203A). The knowledge information input unit 135 inputs the knowledge information obtained in S203A to the update instruction generation unit 133 (S203B). Here, the second external data source 50 can be any data source that is accessible from the knowledge information input unit 135 and contains information useful for improving the areas requiring improvement. For example, the second external data source 50 is the internet. In this case, the knowledge information input unit 135 may obtain knowledge information useful for improving the areas requiring improvement by, for example, searching on an internet search site.

[0055] For example, the knowledge information input unit 135 can input the dialogue log 121 of a failure case into the LLM and input instructions to the LLM such as, "Please extract the lack of knowledge (or rules) that caused the dialogue to fail," thereby recognizing the knowledge (or rules) necessary to improve the dialogue in that failure case. Alternatively, the knowledge information input unit 135 may be configured to extract frequently occurring words from the dialogue log 121 of a failure case and recognize the extracted frequently occurring words as the knowledge (or rules) necessary for improvement. In this case, the extraction of frequently occurring words may be achieved by rule-based processing or by instructing the LLM to extract frequently occurring words.

[0056] As shown in Figure 4, since administrators do not need to input knowledge information, the costs associated with system operation can be reduced.

[0057] [Third Embodiment] Figure 5 is a diagram (part 3) showing an example of the processing flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model (LLM) through dialogue with the user. The flow in Figure 5 differs from the case in Figure 3 in that, instead of the administrator M directly inputting knowledge information to the update instruction generation unit 221, the administrator M inputs knowledge information to the update instruction generation unit 133 via the second dialogue unit 136. That is, in this case, the control unit 130 of the dialogue device 100 includes the second dialogue unit 136 in addition to the configuration in Figure 2. For example, the second dialogue unit 136 may be configured by an LLM.

[0058] In this case, for example, the second dialogue unit 136 inputs a dialogue list and, based on this dialogue list, obtains knowledge information useful for improving the areas that need improvement through dialogue with the administrator M. For example, the second dialogue unit 136 presents the areas that need improvement listed in the dialogue list to the administrator M (S203C) and requests the administrator M to input knowledge information useful for improving the presented areas that need improvement. In response to this request, the administrator M answers (inputs) knowledge information useful for improving the presented areas that need improvement to the second dialogue unit 136. The knowledge information that the administrator M inputs to the second dialogue unit 136 may be knowledge that the administrator M already knew, or it may be knowledge that the administrator M learned as a result of an investigation conducted in response to the request in S203C.

[0059] Furthermore, for example, the second dialogue unit 136 may be configured to determine the knowledge information (missing knowledge information) that is lacking in order to provide an appropriate answer to the dialogue about areas requiring improvement, and to request that administrator M input this missing knowledge information. In this case, the second dialogue unit 136 may be configured to determine the missing knowledge information based on the constraint information and the contents of the dialogue list stored in the external data source 40. Also, the interface between the second dialogue unit 136 and administrator M may be in the form of a text-based chat, or it may be configured as a voice-based interface using a microphone or intercom.

[0060] With the second dialogue unit 136 configured in this way, for example, administrator M can input knowledge information useful for improving areas that need improvement to the second dialogue unit 136 by engaging in the following exchanges (dialogues) as described in (1) to (3). (1) Second Dialogue Section 136: "I'm not sure if there's a diaper changing station in the second-floor restroom." (2) Manager M: "There is no changing table in the toilet on the second floor. But there is one in the toilet on the first floor." (3) Second Dialogue Section 136: "Thank you. I'll update the knowledge base with that information."

[0061] [Fourth embodiment] Figure 6 is a diagram (part 4) showing an example of the process flow in which the dialogue device 100 of the embodiment updates the constraint information of the dialogue model (LLM) through dialogue with the user. In Figure 3, when the update instruction generation unit 133 generates update instruction information INS, the update execution unit 134 updates the constraint information based on the generated update instruction information INS. In contrast, the flow in Figure 6 differs from the case in Figure 3 in that the constraint information is updated only when the administrator M confirms the generated update instruction information INS and decides to execute the update.

[0062] In this case, even if the update instruction information INS is generated in S204A, the update execution unit 134 does not immediately update the constraint information, but waits for an update execution instruction from the update instruction generation unit 133. Meanwhile, the administrator M obtains the update instruction information INS generated in S204A from the dialogue device 100 and checks the validity of its contents (S204C). If the administrator M determines that there are points that need to be corrected in the update instruction information INS, they input the correction instruction to the update instruction generation unit 133 (S204D). For example, the correction instruction includes feedback information to allow the update instruction generation unit 133 to recognize the points that need to be corrected and from what perspective the correction should be made. When the correction instruction is input in S204D, the update instruction generation unit 133 recognizes the content that needs to be corrected based on the feedback information and reflects it in the update instruction information INS (S204E). Steps S204C to S204E are repeated until the administrator M can confirm the validity of the update instruction information INS. In this way, by repeatedly inputting feedback information and modifying the update instruction information INS, the likelihood of obtaining the desired output can be increased.

[0063] Furthermore, the modification of the update instruction information INS may be achieved by editing the already generated update instruction information INS. Alternatively, for example, the update instruction generation unit 133 may modify the update instruction information INS by changing the parameters of the process that outputs the update instruction information INS based on the feedback information and re-outputting the update instruction information INS. The update instruction generation unit 133 may also determine whether or not to re-output the update instruction information INS based on the content of the update instruction information INS that was subject to validation and the feedback information. In addition, the update instruction generation unit 133 may perform additional learning based on such output (update instruction information INS) and feedback information. This can improve the accuracy of the inference of the update instruction generation unit 133.

[0064] Generally, the method of further learning a trained model through human feedback is called reinforcement learning from human feedback (RLHF). In RLHF, a human gives OK (positive feedback) or NG (negative feedback) to the output of the trained model, and the trained model is updated based on this feedback information. This allows the output of the trained model to be adjusted to better align with human intentions and expectations.

[0065] In contrast, the fourth embodiment stores feedback information as the same session information (e.g., prompt history, session memory, metadata, etc.) and sequentially optimizes the output based on the stored feedback information. In other words, in the fourth embodiment, there is no need to retrain the LLM (e.g., update weight parameters) based on the feedback information. Therefore, according to the fourth embodiment, more accurate update instruction information INS can be generated efficiently.

[0066] On the other hand, in S204C, if administrator M confirms the validity of the update instruction information INS, administrator M instructs the update instruction generation unit 133 to perform an update of the constraint information (S204F). Upon receiving the update execution instruction in S204F, the update instruction generation unit 133 instructs the update execution unit 134 to perform an update of the constraint information (S204G). Then, upon receiving the update execution instruction in S204G, the update execution unit 134 updates the constraint information stored in the external data source 40 based on the generated update instruction information INS (S205).

[0067] If there are no points to be corrected in the update instruction information INS generated in S204A, administrator M may enter the update execution instruction S204F without entering the correction instruction S204D. In this case, S204D and S204E are omitted, and the update execution unit 134 updates the constraint information based on the update instruction information INS generated in S204A.

[0068] According to the dialogue system 1 of the embodiment described above, it becomes possible to modify the dialogue system using natural language more efficiently.

[0069] <Variation> In the above embodiment, a case was described in which the dialogue unit 131, the unit for extracting areas requiring improvement 132, the unit for generating update instructions 133, the unit for executing updates 134, the unit for inputting knowledge information 135, and the second dialogue unit 136 are provided in the dialogue device 100. However, some or all of these functional units, other than the dialogue unit 131, may be configured separately from the dialogue device 100. For example, the dialogue device 100 may be equipped with the dialogue unit 131 and be communicatively connected to a separate information processing device equipped with the unit for extracting areas requiring improvement 132, the unit for generating update instructions 133, the unit for executing updates 134, the unit for inputting knowledge information 135, and the second dialogue unit 136.

[0070] In the above embodiment, we described a case in which the dialogue unit 131 of the dialogue device 100 mainly provides a dialogue interface using natural language. However, the dialogue unit 131 may be configured to provide a so-called multimodal dialogue interface that can also recognize information other than natural language. For example, the dialogue unit 131 may be configured to take input such as text, images, and audio directly or via a file, and to output audio or images that indicate a response to those inputs.

[0071] In the above embodiment, a case in which the output of the dialogue unit 131 is controlled by constraint information has been described, but such control of output by constraint information may also be applied to functional units other than the dialogue unit 131. For example, the output of the improvement area extraction unit 132, the update instruction generation unit 133, the knowledge information input unit 135, or the second dialogue unit 136, which can be configured using LLM similar to the dialogue unit 131, may be controlled by applying dedicated constraint information to each of them. By controlling the output by applying dedicated constraint information in this way (for example, defining the output format, NG rules, file extensions, etc.), it is possible to suppress the input and output of unnecessary information and make it easier to obtain the desired output, thereby improving the performance of the improvement phase.

[0072] [Update instruction generation unit] (1) Preprocessing of knowledge information The update instruction generation unit 133 may be configured to instruct the LLM to perform preprocessing in order to improve the accuracy of the update instruction information INS. For example, the update instruction generation unit 133 may instruct the LLM to format the data input as knowledge information. Alternatively, for example, if knowledge information in text format is input, the update instruction generation unit 133 may instruct the LLM to extract key information from the text data. For example, the update instruction generation unit 133 may input the instruction to the LLM, "Please organize and extract the key and value for the knowledge to be added," from the input text data. This allows, for example, if knowledge information such as "The toilet is located at the south end of the second floor of the main building" is input, the knowledge information can be converted into the form "Info_Type=Location, Name=“Toilet”, value=“South end of the second floor of the main building”". By converting the knowledge information in advance into a form that is easy for the LLM to handle through such preprocessing, the accuracy of inference by the LLM can be improved.

[0073] (2) Input support function for feedback information In the fourth embodiment, a configuration was described in which administrator M inputs feedback information to the update instruction information INS generated by the update instruction generation unit 133. However, it is conceivable that administrator M may need a certain level of expertise to judge the validity of the update instruction information INS. Therefore, in order to make it easier for administrator M to judge the validity of the update instruction information INS, the dialogue device 100 may be configured to show administrator M an example of how the content of the dialogue will change as a result of feedback information being input by administrator M.

[0074] In this case, for example, the update instruction generation unit 133 may be configured to store both the constraint information before the update and the constraint information after it has been updated by the update instruction information INS which has been modified based on the input feedback information, and to present the output of the same dialogue performed in both cases to the administrator M in a comparable manner. For example, the same dialogue may be a failure pattern extracted from the dialogue log 121. With the update instruction generation unit 133 configured in this way, the content of the dialogue based on the constraint information before and after the update is presented to the administrator M, so that even if the administrator M does not have expertise in the update instruction information INS, it is possible to confirm the validity of the feedback information from the change in the content of the dialogue before and after the update.

[0075] [Identification of areas requiring improvement] (1) Classification of the causes of the failure of the dialogue In addition to the above functions, the improvement area extraction unit 132 may be configured to have a function for classifying the cause of failure cases. For example, the improvement area extraction unit 132 may be configured to input the dialogue log 121 of a failure case to the LLM and instruct the LLM to classify the cause of the failure of the dialogue in the failure case. Alternatively, for example, the improvement area extraction unit 132 may be configured to input the dialogue log 121 of a failure case to a trained model that has been trained by machine learning to take the dialogue log 121 of a failure case as input and output a classification result of the cause of the failure of the dialogue in the failure case, thereby obtaining a classification result (cause type) of the cause of the failure of the dialogue in the failure case.

[0076] With the improved area extraction unit 132 configured in this way, it is possible to recognize the type of constraint information that needs to be corrected when improving the dialogue of a failure case, making it possible to correct the constraint information more accurately and improving the accuracy of the improvement area. For example, if the cause type is an error in the response policy, the improvement accuracy can be expected to be improved by notifying the update instruction generation unit 133 that the target of correction is policy 41. Also, for example, if the cause type is a lack of knowledge or an error in reasoning, the improvement accuracy can be expected to be improved by notifying the update instruction generation unit 133 that the target of correction is knowledge 42. Furthermore, for example, if the cause type is a communication error or an error in speech recognition, the improvement accuracy can be expected to be improved by notifying the update instruction generation unit 133 that correction of constraint information is unnecessary, thereby preventing a decrease in the accuracy of the dialogue of the dialogue unit 131 due to unnecessary corrections.

[0077] (2) Output of evaluation report In addition to the above functions, the Improvement Area Extraction Unit 132 may be configured to generate an evaluation report on the performance of the Dialogue Unit 131 based on the information of the extracted failure cases. For example, the evaluation report may list the number of extracted failure cases, the percentage of extracted failure cases that the extracted failure cases represent of the total number of dialogue cases, specific examples of failure cases, and the types of causes. The Improvement Area Extraction Unit 132 may be configured to format the evaluation report using a rule-based system or LLM, or it may be configured to summarize the information to be included in the evaluation report using LLM. With the Improvement Area Extraction Unit 132 configured in this way, it becomes possible to measure the performance of the dialogue system 1 and the dialogue model (LLM), and to evaluate the validity of customer satisfaction based on the measurement results.

[0078] [Dialogue Section] The dialogue unit 131 may be configured to output visual information that supplements the dialogue information, in addition to inputting and outputting language-based dialogue information (such as voice and chat). In this case, the dialogue device 100 may include a display device for outputting the visual information. For example, the visual information may be displayed in the form of a map. For example, if user U inputs the question "Where is the toilet?" into the dialogue unit 131, the dialogue unit 131 may display a map on its display unit showing the route from the current location to the nearest toilet, along with a language-based response such as "Please look here." By providing visual information along with the dialogue information in this way, the dialogue device 100 can improve the accuracy of information transmission to user U.

[0079] In the above embodiment, a configuration was described in which user U inputs voice to user terminal device 200 and dialogue device 100 acquires the voice data from user terminal device 200. However, the dialogue unit 131 may have a microphone for voice input, in which case user terminal device 200 may be omitted. For example, the dialogue unit 131 may be configured as a robot with physical characteristics such as a humanoid or animal form. In this case, the robot may be equipped with a communication interface for communicating with the main body of dialogue device 100, as well as input / output devices such as a microphone, speaker, and touch panel. In this case, the robot may also be configured to transmit information through body movements in addition to such input / output devices. For example, when asked about a destination location during a dialogue with user U, the robot may be configured to move its arm and point in the direction of the destination location. Such movements can also be considered a type of visual information in the sense that they are perceived by user U's vision.

[0080] Furthermore, it is conceivable that the nature of dialogue information may change due to the supplementation of visual information, making it difficult to determine the success or failure of the dialogue based solely on the dialogue log. For example, if the response given in the above example is "Please look here," it is difficult to determine whether the dialogue was successful or not because the text-based information does not reveal what the visual information at that time was. Therefore, when supplementing dialogue information with visual information, the dialogue unit 131 may be configured to record the dialogue log 121 in association with the visual information displayed at the time the dialogue took place, along with the content of the dialogue that was performed. In this case, the dialogue unit 131 may record the image of the visual information itself in the dialogue log 121, record text information summarizing the content of the visual information in the dialogue log 121, or convert the visual information into features and record it in the dialogue log 121. This allows the subsequent improvement area extraction unit 132 to extract areas that need improvement based on both the dialogue information and the visual information.

[0081] [Dialogue logging enhancement] (1) Expansion using image data Even if the content of the dialogue log 121 confirms in text that user U said "thank you" to the dialogue device 100, it is possible that user U is not actually satisfied with the dialogue with the dialogue device 100. In such cases, the dialogue system 1 may be equipped with a camera for capturing user U's facial expression during the dialogue, and configured to input the image data captured by the camera, along with the dialogue log 121, to the improvement area extraction unit 132. In this case, the improvement area extraction unit 132 may be configured to read user U's facial expression from the input image and extract dialogue failure cases based on the read facial expression and the content of the dialogue log 121. The method for reading a person's facial expression from an image may be implemented using any existing technology. In this case, the improvement area extraction unit 132 may be configured to perform a process to detect user U's facial expression from the image data and input the detection result to the LLM instead of the image data itself. Also, if the LLM supports multimodal processing, the improvement area extraction unit 132 may input the image data itself to the LLM.

[0082] (2) Augmentation using audio data Furthermore, the dialogue system 1 may be equipped with a microphone for inputting the voice spoken by user U, and may be configured to input the voice data recorded by the microphone, along with the dialogue log 121, to the improvement area extraction unit 132. In this case, the improvement area extraction unit 132 may be configured to read user U's satisfaction level with the dialogue device 100 from the input voice, and to extract dialogue failure cases based on the read satisfaction level and the contents of the dialogue log 121. For example, the improvement area extraction unit 132 can extract feature quantities from the voice data that relate to features that may be influenced by user U's thoughts and emotions, such as tone of voice, manner of speaking, intonation, and speed, and estimate user U's evaluation of the dialogue with the dialogue device 100 (satisfied or dissatisfied, etc.) by comparing or classifying the extracted feature quantities or index values ​​obtained based on them with thresholds. In this case, the improvement area extraction unit 132 may be configured to input the estimated evaluation result to the LLM instead of the voice data itself. Also, if the LLM supports multimodal operation, the improvement area extraction unit 132 may input the voice data itself to the LLM.

[0083] Furthermore, if the user terminal device 200 is owned by individual users, the camera and microphone built into the user terminal device 200 may be used. Also, if the user terminal device 200 is shared by an unspecified number of users (for example, an information terminal), a camera that captures images of the area around the user terminal device 200 and a microphone that inputs sounds from the area around the user terminal device 200 may be separately installed. With such a configuration, the text-based dialogue log 121 can be expanded with non-verbal information such as the voice and images of user U during the dialogue, so that the improvement area extraction unit 132 can more appropriately determine whether or not it is a failure case when extracting areas that need improvement.

[0084] (3) Expansion using sound environment information In situations where the sound environment of the dialogue space is poor, such as when there is a lot of noise, users may have difficulty hearing the voice, and the accuracy of the dialogue system 1 in recognizing the user's voice will also decrease, which may cause dialogue failures. Therefore, the dialogue device 100 may be configured to input time-series information (sound environment information) indicating the state of the sound environment during the dialogue in the dialogue space, along with the dialogue log 121, to the improvement area extraction unit 132. In this case, the dialogue system 1 may be equipped with a microphone for sensing the sound environment of the dialogue space (hereinafter referred to as the "environment microphone") in addition to the microphone used for dialogue with the user. In this case, the dialogue system 1 generates and stores sound environment information based on the voice data input from the environment microphone. Furthermore, in this case, the improvement area extraction unit 132 may be configured to recognize the state of the sound environment during the dialogue based on the sound environment information, and to extract dialogue failure cases based on the recognition result and the contents of the dialogue log 121. With this configuration, the text-based dialogue log 121 can be expanded with sound environment information of the dialogue space, so the improvement area extraction unit 132 can more appropriately determine whether or not it is a failure case when extracting areas that need improvement.

[0085] (4) Expansion using motion control information As described above, the dialogue unit 131 may be configured as a robot or other entity capable of performing physical actions. In this case, visual information presented by the robot through physical actions can also be used as information to extend the dialogue log 121, along with the visual information that can be presented by the image display described above. For example, if the robot points to a location inquired about by the user U using the direction of its arm, the dialogue device 100 stores the control information of such action along with the dialogue log 121. In this case, the improvement area extraction unit 132 may be configured to recognize the robot's actions during the dialogue based on the control information and to extract dialogue failure cases based on the recognition result and the contents of the dialogue log 121. With such a configuration, the text-based dialogue log 121 can be extended with information about the actions (outputs) performed by the dialogue unit 131, so the improvement area extraction unit 132 can more appropriately determine whether or not a case is a failure when extracting areas that need improvement.

[0086] Although embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention. [Industrial applicability]

[0087] This invention is applicable to systems that provide a human-to-human interaction interface based on constraint information. [Explanation of Symbols]

[0088] 1. Dialogue System 40 External data sources 41 Policies 42 Knowledge 50 Second external data source 100 Dialogue device 110 Communications Department 120 Storage section 121 Dialogue Log 122 Update Patch 123 Alternative files 130 Control Unit 131 Dialogue Section 132 Section for identifying areas requiring improvement 133 Update instruction generation unit 134 Update Execution Unit 135 Knowledge Information Input Section 136 Second Dialogue Section 200 User terminal devices 221 Update instruction generation unit 300 Administrator terminal device 300 User terminal devices 400 External data sources

Claims

1. An information processing device that updates constraint information that affects the response generated in a dialogue with a user, which is a dialogue device that provides a dialogue interface that generates a response based on information provided by the user, An extraction unit that extracts areas requiring improvement based on the record of the aforementioned dialogue, A generation unit that receives input of knowledge information useful for improving the aforementioned areas requiring improvement, and generates update instruction information for updating the constraint information based on the knowledge information, An update unit that updates the constraint information based on the update instruction information, An information processing device equipped with the following features.

2. The extraction unit presents the extracted areas requiring improvement to the administrator of the dialogue device. The generation unit receives input of the knowledge information from the administrator. The information processing apparatus according to claim 1.

3. The system further comprises an input unit for inputting the knowledge information to the generation unit, The extraction unit supplies the areas requiring improvement to the input unit. The input unit extracts knowledge information useful for improvement from an external data source regarding the areas requiring improvement supplied from the extraction unit, and inputs it into the generation unit. The information processing apparatus according to claim 1.

4. The input unit is configured using a dialogue model that has been trained to respond to the user's input. The information processing apparatus according to claim 3.

5. The device further comprises a dialogue unit that provides an interface for dialogue with the administrator of the aforementioned dialogue device, The aforementioned dialogue unit, Based on the aforementioned areas requiring improvement, identify the necessary information as knowledge information and present it to the administrator. The system receives input of the knowledge information from the administrator and inputs the knowledge information entered by the administrator into the generation unit. The information processing apparatus according to claim 1.

6. The dialogue unit is configured using a dialogue model that has been trained to respond to user input. The information processing apparatus according to claim 5.

7. When the generation unit receives feedback information from the administrator regarding corrections to the generated update instruction information, it corrects the update instruction information based on the feedback information. The information processing apparatus according to claim 5.

8. The generation unit stores both the first constraint information, which is the constraint information before the update, and the second constraint information, which is the constraint information after it has been updated by the update instruction information modified based on the feedback information. The dialogue unit then presents to the administrator, in a comparable manner, the output of the same dialogue performed with the first constraint information and the second constraint information, respectively. The information processing apparatus according to claim 7.

9. The extraction unit and the generation unit are configured using a dialogue model that has been trained to respond to the user's input. The information processing apparatus according to claim 1.

10. The generation unit performs preprocessing on the input knowledge information, such as formatting it or extracting key information for updating the constraint information, and inputs the preprocessed knowledge information into the dialogue model. The information processing apparatus according to claim 9.

11. The extraction unit, with respect to the dialogue extracted as areas requiring improvement, classifies the causes of the extracted areas requiring improvement into categories corresponding to the policy for correcting the constraint information and presents them to the administrator. The information processing apparatus according to claim 2.

12. The dialogue device interacts with the user through language-based dialogue information exchanged with the user and visual information that supplements the dialogue information. The record of the dialogue includes historical information relating to the dialogue information and the visual information. The information processing apparatus according to claim 1.

13. The aforementioned dialogue interface is provided as a robot capable of physical movement. The historical information relating to the visual information includes historical information relating to the physical actions performed by the robot. The information processing apparatus according to claim 12.

14. The dialogue device is equipped with a camera for reading the user's facial expressions. The record of the dialogue includes language-based dialogue information exchanged with the user, and historical information relating to the user's facial expressions captured by the camera or recognized from the images. The information processing apparatus according to claim 1.

15. The dialogue device is equipped with a microphone for sensing the sound environment of the space in which the user interacts. The recording of the dialogue includes language-based dialogue information exchanged with the user, and historical information regarding the sound environment captured by the microphone or the sound environment recognized from the sound. The information processing apparatus according to claim 1.

16. The record of the dialogue includes language-based dialogue information exchanged with the user, and historical information relating to the user's voice or the user's evaluation of the dialogue device as recognized from the voice. The information processing apparatus according to claim 1.

17. The record of the aforementioned dialogue includes historical information regarding communication errors in the dialogue device. The information processing apparatus according to claim 1.

18. Computers An information processing method for updating constraint information that affects the answers generated in a dialogue with a user, provided by a dialogue device that provides a dialogue interface that generates answers based on information provided by the user, The aforementioned computer, Based on the record of the aforementioned dialogue, we will identify areas that need improvement, The system accepts input of knowledge information useful for improving the aforementioned areas requiring improvement, and generates update instruction information for updating the constraint information based on the aforementioned knowledge information. The constraint information is updated based on the update instruction information. Information processing methods.

19. On the computer, A program for causing a dialogue device that provides a dialogue interface that generates answers based on information provided by a user to perform an update process to update constraint information that affects the answers generated in the dialogue with the user, The aforementioned update process is: A process to extract areas that need improvement based on the record of the aforementioned dialogue, A process that accepts input of knowledge information useful for improving the aforementioned areas requiring improvement, and generates update instruction information for updating the constraint information based on the aforementioned knowledge information, A process to update the constraint information based on the update instruction information, including, program.

Images