Device and method

Abstract

The present invention generates an advertisement for a user as a customer at a time that suits the user's perspective.　In an interactive server 100 of the present disclosure, a voice recognition unit 101 acquires a user's utterance from a user terminal. Then, a trust degree measurement unit 103 measures the user's degree of trust in interaction on the basis of the user's utterance. A response generation unit 109 generates a response on the basis of provided information (for example, user interests and an advertisement to be provided to the user) acquired in response to the utterance, in accordance with the degree of trust. In the present disclosure, the interactive server 100 and a user terminal 200 interact with each other using voice, but the interaction is not necessarily limited to voice, and may be performed using text or the like.

Description

Device and Method

[0001] The present invention relates to a device and a method for interacting with a user.

[0002] Patent Document 1 describes providing an information presentation device that can present an advertisement with more appropriate content at a more appropriate timing to a user. Specifically, in a dialogue system that performs more detailed analysis processing while referring to cases assuming user speech, an advertisement is presented and a response for presenting the advertisement is made. By doing so, an advertisement that matches the user's preference is selected from among the advertisements and presented at an appropriate timing, or a simple response to the advertisement inquiry can be returned.

[0003] Japanese Patent Application Laid-Open No. 2002-2976926

[0004] However, in the invention described in Patent Document 1, although the timing of presenting an advertisement for a product that has emerged in the dialogue with the user is considered, and an advertisement for a product or the like that the user is interested in is presented, there are cases where it is not the timing for the user to accept the advertisement. That is, advertisements are often presented from the perspective of the company, and there is a high possibility of being unpleasant for the customer.

[0005] Therefore, an object of the present invention is to provide a device and a method for generating an advertisement at a timing according to the user's perspective for a user who is a customer.

[0006] The device of the present disclosure includes a dialogue information acquisition unit that acquires dialogue information from a user, a measurement unit that measures the reliability of the user's response to the dialogue based on the dialogue information, and a response generation unit that generates a response based on the provided information acquired according to the dialogue information according to the reliability.

[0007] According to the present invention, an advertisement can be generated at a timing according to the user's perspective for a user who is a customer.

[0008] Figure 1 is a system configuration diagram including the dialogue server 100 and user terminal 200 of this disclosure. Figure 2 is a diagram showing the functional configuration of the dialogue server 100 of this disclosure. Figure 3 is a schematic diagram showing the confidence measurement process. Figure 4 is a diagram showing the process of measuring RT confidence based on specific examples. Figure 5 is a diagram showing the His confidence measurement process based on specific examples. Figure 6 is a diagram showing an overview of the internal estimation process. Figure 7 is a diagram showing the detailed processing of the learning process of the estimation model for estimating the user's internal state. Figure 8 is a schematic diagram showing the action (output) in response to the state (input) in the judgment unit 105. Figure 9 is a schematic diagram showing an overview of the interest acquisition process by the interest acquisition unit 106. Figure 10 is a flowchart showing the operation of the dialogue server 100. Figure 11 is a diagram showing the input information and the output thereto of the emotion recognition unit 102. Figure 12 is a diagram explaining the processing of multiple phrases. Figure 13 is a diagram showing the response to the user's utterance via the NN. Figure 14 shows an example of the hardware configuration of a dialogue server 100 and a user terminal 200 according to one embodiment of the present disclosure.

[0009] Embodiments of this disclosure will be described with reference to the attached drawings. Where possible, the same parts will be denoted by the same reference numerals, and redundant descriptions will be omitted.

[0010] Figure 1 is a system configuration diagram including the dialogue server 100 and user terminal 200 of this disclosure. The dialogue server 100 receives access from the user terminal 200 and engages in dialogue. That is, the dialogue server 100 receives a message from the user terminal 200, generates a response message in response to that message, and sends it to the user terminal 200. The dialogue server 100 is equipped with a dialogue AI (for example, LLM (Large Language Models)) and can engage in dialogue by analyzing the user's utterance (voice or text) sent from the user terminal 200 and generating a response utterance (voice or text). Note that the functionality of the dialogue server 100 may also be contained within the user terminal 200.

[0011] If the interaction is voice-based, the other party (generated AI) may be displayed on the screen, creating the illusion of talking to a real person. If it is text-based, it can be conducted in a chat format.

[0012] Figure 2 is a diagram showing the functional configuration of the dialogue server 100 of this disclosure. As shown in the figure, the dialogue server 100 is composed of a speech recognition unit 101, an emotion recognition unit 102, a confidence level measurement unit 103, an inner thoughts estimation unit 104, a judgment unit 105, an interest acquisition unit 106, an advertisement storage unit 107, a speech history storage unit 108, an action history storage unit 108a, a response generation unit 109, and a speech synthesis unit 110.

[0013] The speech recognition unit 101 is the part that receives speech transmitted from the user terminal 200, recognizes the speech, and generates text information.

[0014] The emotion recognition unit 102 recognizes the user's emotions based on the textual information and acoustic information of the speech recognized by the speech recognition unit 101. Techniques for obtaining emotions from textual and acoustic information are existing techniques. For example, one possible method is to pass the textual information through an ANN (Attention Neural Network) to create feature vectors (real number vectors), and obtain real values ​​such as F0 (fundamental frequency) or F1 (formant) as acoustic information, and then use machine learning to classify emotions into multiple classes from the combined feature vector. Furthermore, in addition to textual or acoustic information, video information (image information) obtained by image recognition of the user's facial expressions may also be added to obtain emotions. In this disclosure, the emotion recognition result is multi-valued (joy, anger, etc.), and a score is assigned to the recognition result.

[0015] Let's explain in more detail. Figure 11 shows the input information and output of the emotion recognition unit 102. As shown in the figure, the emotion recognition unit 102 takes text information, sound information, and images (facial expressions) as input and outputs scores (probability values) for several emotions. As shown in the figure, the emotion recognition unit 102 takes text information, sound information, and images (facial expressions) as input and outputs scores for joy, anger, and sadness. In the figure, the output is joy: 0.9, anger: 0.05, and sadness: 0.05. In this case, the emotion recognition unit 102 recognizes the user's utterance as joy, which has a score of 0.9 (the maximum value).

[0016] When recognizing facial expressions, the dialogue server 100 is equipped with an image acquisition unit (not shown), which acquires the user's facial expressions from the user terminal 200.

[0017] The confidence level measurement unit 103 is the part that measures the user's level of trust in the dialogue. This confidence level is a value that indicates how much the user trusts the dialogue server 100 (generating AI) through the dialogue, based on the emotions that have been recognized.

[0018] First, the level of trust in a dialogue is broken down into "(A) credibility," "(B) empathy," and "(C) logic." Furthermore, the trust level measurement unit 103 classifies and associates the content of the dialogue AI's utterances with these "(A) credibility," "(B) empathy," and "(C) logic" as follows: (A) Utterances about facts (news, etc.): "Example: It looks like it will be sunny tomorrow." (B) Empathetic utterances: "Example: That must have been tough." (C) Utterances about suggestions: "Example: Since it is said that XX, how about XX?" (D) Others: Utterances not used in the calculation of trust level. If the user's feelings towards the above (A), (B), and (C) are positive, it is assigned a + (plus), and if they are negative, it is assigned a - (minus), and the trust level is calculated. The estimated emotions are assumed to be "anger," "joy," "anxiety," "excitement," "positive," "negative," "affirmative," "negative," "sadness," "surprise," "neutral," "disgust," and "fear," and a value is set to be added to or subtracted from the confidence level for each. For example, anger is set to -1, and joy is set to +1.

[0019] Figure 3 schematically illustrates the reliability measurement process. Figure 3(a) shows how to obtain the RT reliability score through real-time processing. As shown in the figure, emotion recognition is performed from user utterances, and emotions are recognized. In the figure, joy and anger are recognized, and the reliability measurement unit 103 scores them to measure the RT reliability.

[0020] Figure 3(b) shows the calculation of a score derived from past utterances (His). As shown in the figure, the utterance history storage unit 108 stores the time, utterance content, and emotion in association with each other. This utterance history is information that associates past utterances and their emotions (recognized by the emotion recognition unit 102) with the time. The confidence level measurement unit 103 scores the His confidence level based on the positive and negative tendencies of the utterance content and emotions of the dialogue server 100 (dialogue AI).

[0021] Figure 4 shows the process for measuring RT confidence based on a specific example. In this figure, the user's utterance is "xxx, xxx". In response, the dialogue server 100 (generating AI) makes an utterance classified as one of the following: (A) an utterance about facts, (B) an utterance of empathy, (C) an utterance about a suggestion, or (D) other. If the user's utterance is "Oh really!", it is recognized as an emotion of surprise, and the confidence level is increased by 1. In other words, the user moves towards a state of trusting the dialogue server 100. Conversely, if the utterance is "What? Is that true?", it is recognized as a negative emotion, and the confidence level is decreased by 1. In other words, the user moves towards a state of distrusting the dialogue server 100.

[0022] In this disclosure, utterances and confidence levels are classified into (A), (B), and (C), and the confidence level is calculated by summing them up. However, confidence levels may also be calculated according to the classified items, such as confidence levels obtained from (A) only, or confidence levels obtained from (A) and (B) only. Since the type of confidence level may differ, calculating the confidence level accordingly may allow for more appropriate judgments.

[0023] In this disclosure, utterances and confidence levels are classified into (A), (B), and (C), and the confidence level is calculated by summing them up. However, it is also possible to calculate the confidence level according to the classified items, such as using the confidence level obtained from (A) only, or the confidence level obtained from (A) and (B) only. The choice of which confidence level from (A), (B), or (C) to use is closely related to the dialogue system to be built, and by selecting it appropriately, the judgment ability of the dialogue system can be improved.

[0024] Similarly, based on the user's utterance in response to what the dialogue server 100 has said, the emotion recognition unit 102 recognizes the user's emotions, and a level of trust in the dialogue (dialogue server 100) is determined according to those emotions.

[0025] In this disclosure, RT confidence is based on multiple phrases actually spoken by the user, and these multiple phrases may contain a mixture of multiple emotions. In this disclosure, these mixed emotions are quantified and expressed as RT confidence. In this disclosure, these multiple phrases may be a series of utterances (or text information) made by the user, or utterances (or text information) contained within a predetermined time interval (e.g., a few seconds).

[0026] In this disclosure, RT confidence is based on multiple phrases actually spoken by the user, and these multiple phrases may contain a mixture of multiple emotions. In this disclosure, these mixed emotions are quantified and expressed as RT confidence. Furthermore, in this disclosure, these multiple phrases may be a series of utterances (or text information) spoken by the user, utterances (or text information) contained within a predetermined time interval (e.g., a few seconds), or utterances separated by a predetermined silent interval (pause).

[0027] Figure 12 illustrates the processing of multiple phrases. Figure 12(a) shows multiple phrases. For example, consider the phrase, "I am a cat (pause)...". Here, "pause" indicates a silent interval. In this way, when there is a silent interval, the phrase is divided into parts before and after it. In Figure 12(b), the part "I am a cat" is calm: 0.9, and the part "I only remember crying meow in a dark, damp place" is sad: 0.9. Each phrase is divided by a silent interval, and the corresponding emotion is output.

[0028] Figure 5 shows the process for measuring His confidence level based on a specific example. In the figure, the total value over a certain period in the past is used as the past confidence level. As mentioned above, the unit for aggregating the confidence level in the figure may be a series of utterances or a predetermined time period. The RT confidence level of the utterances actually spoken as described above is stored as confidence level history information in the utterance history storage unit 108, and this is used to aggregate the His confidence level. Note that the confidence level may decrease as you go further back in time. The utterances to be aggregated may be limited to only the responses to (A), (B), and (C) above.

[0029] Figure 5(a) shows the total confidence level over a certain period in the past. Confidence levels outside the thick border are outside that period and are excluded from the confidence level calculation. Figure 5(b) shows that confidence levels decrease as you go further back in time. As shown in the figure, the weighting coefficients are reduced as you go further back in time to minimize the impact of the reduction in the total confidence level. Similar to Figure 5(a), confidence levels outside the thick border are excluded from the summation process. The borders in Figure 5 are defined by predetermined periods or utterance counts.

[0030] The internal character estimation unit 104 is the part that estimates the user's internal character (personality, awareness, and cognitive biases) from their speech history. The internal character estimation unit 104 estimates the user's internal character (scored) using, for example, a pre-trained estimation model. In addition to speech history, behavioral history may also be considered. This internal character estimation process by the internal character estimation unit 104 is a known technique. Its outline is described below.

[0031] Figure 6 shows an overview of the processing. In Figure 6(a), the internal estimation unit 104 features the speech history (time, content of speech, emotion) and estimates personality, consciousness, and cognitive bias using a pre-trained estimation model. Furthermore, as shown in Figure 6(b), the internal estimation unit 104 can also feature the behavioral history (location information, purchase information) to estimate personality, consciousness, and cognitive bias, and utilize this information. The behavioral history, like the speech history, is information that is compiled in advance by the dialogue server 100. Location information indicating when and where the user is, and purchase information indicating when and where and what was purchased, are stored as the behavioral history, but of course, other behaviors may also be included.

[0032] Figure 7 shows the detailed processing steps of the learning process for an estimation model used to estimate the user's inner thoughts. Figure 7(a) shows the training data prepared in advance for training. This training data consists of features and ground truth labels for each user. The ground truth labels are information obtained in advance through questionnaires and represent information that indicates the user's inner thoughts (personality, awareness, or cognitive biases, etc.). The features are based on the user's speech history. These features are obtained by applying the speech history (or location information and terminal logs (operation logs, app launch logs, etc.)) to a Neural Network (NN). In Figure 7(a), these features are shown in two dimensions for convenience, but they are information that can be represented in many more dimensions.

[0033] Figure 7(b) shows an example using a single interface, but in order to obtain multiple interior surfaces (for division), multiple interface surfaces are learned using an SVM or similar method. This interface surface learning is performed in advance, and the learned estimation model (interface surface) is provided in the interior surface estimation unit 104.

[0034] Figure 7(c) shows how to obtain features using a neural network (NN). Generally, when time-series data is input into an NN, the compressed time-series data, which is called a feature, is output. In this disclosure, the feature is obtained by inputting speech history into the NN. Here, only speech history is input, but other time-series data or user features may also be input. Furthermore, to solve the problem of predicting the next action, user features (initially random) can also be input into the NN and trained to obtain user features.

[0035] Figure 7(d) shows the compression process using a binative graph of speech and user. By running a GNN (Graph Neural Network) as a link prediction problem, user features can be obtained. These features can then be used for the learning process described above.

[0036] Let me provide further details regarding Figure 7(c). Figure 13 shows the response to a user's utterance via an NN. Normally, as shown in Figure 13(a), the NN (encoder, decoder) is trained to obtain a predetermined utterance (response) by inputting the user's utterance, having an internal encoder output a feature vector, and then feeding this into a decoder. In this disclosure, as shown in Figure 13(b), the NN is further trained to obtain a predetermined response by incorporating the user's features (user features in the figure) into the user's utterance. That is, the input is feature-vectorized within the NN, but by combining and learning user features, user features can be extracted as a byproduct. In Figure 13, the NN is trained based on utterances and responses, but other time-series data can also be used.

[0037] The judgment unit 105 is responsible for determining whether to continue the conversation (behavior to gain trust) or to present an advertisement (behavior to gain money) based on the user's inner thoughts (personality, awareness, cognitive biases) estimated by the inner thoughts estimation unit 104, the trust level measured by the trust level measurement unit 103, the user's most recent actions, and the state of the conversation.

[0038] Figure 8 is a schematic diagram illustrating the behavior of the decision unit 105. As shown in the figure, the DQN (Deep Q Network) can estimate actions from a state through reinforcement learning. The decision unit 105 can estimate (determine) actions from a state using this reinforcement-learned DQN. The DQN is reinforced learning as follows. This reinforcement learning is a generally known process. (i) Defining the state The current state of the dialogue is defined by combining state elements (previous action, RT confidence level, His confidence level, dialogue state, inner thoughts). Alternatively, only the confidence level (His confidence level or RT confidence level) may be used. (ii) Selecting an action The DQN receives the current state as input and selects either an action to gain confidence (small talk) or an action to gain money (advertising). (iii) Calculating the reward Based on the results obtained for the selected action, a reward is calculated. For example, a reward is given if confidence improves through small talk, or if money is obtained through advertising. (iv) Updating the model Based on the obtained rewards, the parameters of the DQN are updated. This will enable us to make better choices in the future.

[0039] As a premise, when the learning mode is ON (initially), the DQN makes random action selections with a very small probability (ε). Conversely, in most cases (with a probability of 1-ε), it makes action selections according to the DQN's selections. The DQN uses the user's emotions (when positive emotions are experienced) resulting from the advertisement as a "reward" during learning. Once a certain amount of results have been accumulated, the DQN performs reinforcement learning. The reward is set to be given when the user's emotions are positive.

[0040] Furthermore, in the initial stages of the service, if DQN's learning is insufficient, the decision unit 105 should make action selections based on the following rules. That is, when the decision unit 105 makes decisions based on rules, it should define the rules so that all patterns (combinations of values ​​that each state can take) can be covered, as follows: Rule 1: Select "Advertisement" if the following conditions are met. The previous action was "Empathy". The state of the (previous) conversation was "Small Talk". The RT confidence level is + (positive). The His confidence level is greater than +1.0. The user's inner state is "Extroverted". Rule 2: Select "Small Talk" if the following conditions are met. The state of the (previous) conversation was "Advertisement". This is a rule to prevent consecutive advertisements. In this way, it is possible to determine whether to engage in small talk or advertisement based on rules.

[0041] The interest acquisition unit 106 is the part that extracts topics or advertisements of interest to the user using the user's past conversations and knowledge graph stored in the speech history storage unit 108.

[0042] Figure 9 is a schematic diagram illustrating the overview of the interest extraction process. As shown in the figure, past utterances are linked to users A to C. This information is stored in the utterance history storage unit 108. The interest acquisition unit 106 then uses past utterances and the knowledge graph to determine topics for casual conversation and advertisements to provide to the users. In Figure 9, one node (word) is linked as a past utterance. This indicates the word that appeared most recently in the utterance.

[0043] A knowledge graph is a data structure that represents information using nodes (points) and edges (lines). Nodes represent specific entities (e.g., "Tokyo" or "Company X"), and edges indicate the relationships between those entities (e.g., "It is the capital" or "It manufactures..."). This makes it clear that it is not just a collection of data, but also the semantic relationships between the data. In this disclosure, this knowledge graph is pre-configured.

[0044] In FIG. 9, the interest acquisition unit 106 acquires a topic (news) or an advertisement as the user's interest by using the past utterances and the knowledge graph for each user. The interest acquisition unit 106 inputs the words included in the most recent utterance among the past topics into the knowledge graph, and uses the nodes output therefrom as keywords. According to the judgment result of the judgment unit 105 (judged action: continue the dialogue or present an advertisement), the interest acquisition unit 106 acquires the keyword as a topic, or acquires the advertisement corresponding to the keyword from the advertisement storage unit 107.

[0045] The advertisement storage unit 107 is a part that assigns and stores keywords for each advertisement. This advertisement storage unit 107 is prepared in advance.

[0046] The utterance history storage unit 108 is a part that stores the history of the user's utterances. This is stored in the utterance history storage unit 108 via the speech recognition unit 101, the emotion recognition unit 102, and the response generation unit 109.

[0047] The action history storage unit 108a is a part that stores the history of the user's actions. This action history information is obtained from a management server (not shown) that collects the user's actions such as the user's location information and purchase information, and is stored in the action history storage unit 108a.

[0048] The response generation unit 109 has an LLM (Large Language Model), and is a part that generates a response sentence using the LLM from the utterance content, the user's interest, and the confidence level. That is, first, the response generation unit 109 acquires the judgment result of whether the judgment unit 105 judges to continue the dialogue (an action to acquire confidence level) or present an advertisement (an action to acquire money), and the acquisition result (advertisement or topic) acquired by the interest acquisition unit 106. Then, the response generation unit 109 generates a prompt for response sentence generation including the utterance content from the user, the judgment result, and the acquisition result, and inputs it into the LLM. The response generation unit 109 performs the response sentence generation process by acquiring the response sentence from the LLM. Note that the LLM may be external, and the response generation unit 109 may transmit the above prompt via a network and obtain the generation result corresponding thereto.

[0049] The speech synthesis unit 110 is the part that generates speech from the response generated by the response generation unit 109.

[0050] Next, the operation of the dialogue server 100 configured in this way will be described. Figure 10 is a flowchart showing the operation of the dialogue server 100.

[0051] The speech recognition unit 101 acquires speech from the user terminal 200 and performs speech recognition on the acquired speech to obtain text information. The emotion recognition unit 102 recognizes the user's emotions from the acquired text information (S101). The confidence level measurement unit 103 then measures the confidence level of the user's dialogue based on the content of the conversation with the user and the user's emotions (S102).

[0052] Next, the internal estimation unit 104 estimates the user's internal state (score) based on the user's speech history (which may also include behavioral history) (S103). Then, the decision unit 105 decides whether to continue the conversation or present an advertisement (S104).

[0053] Here, if the decision unit 105 determines to continue the conversation (S104: conversation continuation), the interest acquisition unit 106 acquires a topic that matches the user's interest (S105). Then, the response generation unit 109 generates a response based on the user's previous utterance and topic (S106). Also, if the decision unit 105 determines to present an advertisement (S104: advertisement presentation), the interest acquisition unit 106 acquires an advertisement that matches the user's interest from the advertisement storage unit 107 (S107). Then, the response generation unit 109 generates a response based on the user's previous utterance and advertisement (S106).

[0054] The speech synthesis unit 110 generates speech based on the response generated by the response generation unit 109 (S108) and transmits the speech to the user terminal (S109).

[0055] In this way, the dialogue server 100 interacts with the user and, in response to the user's response, measures the user's trust in the dialogue server 100. Based on that trust level, it can decide whether to continue the conversation in a way that further increases the trust level or to present an advertisement. Therefore, it can generate and present advertisements at a timing that is easily accepted by the user.

[0056] Next, the effects of this disclosure will be explained. The dialogue server 100 of this disclosure has a speech recognition unit 101 that acquires the user's utterances from the user terminal. The confidence level measurement unit 103 then measures the confidence level of the user's dialogue based on the user's utterances. The response generation unit 109 generates a response based on the confidence level and the information provided in response to the utterances (for example, the user's interests and topics, and advertisements to be offered to the user).

[0057] For example, in the dialogue server 100, the decision unit 105 determines the action to take towards the user according to the level of confidence, and the interest acquisition unit 106 acquires topics or advertisements of interest to the user as information to be provided, in response to the utterance. Then, the response generation unit 109 acquires the advertisements acquired by the interest acquisition unit 106 or the topics, according to the level of confidence.

[0058] According to this disclosure, if the dialogue server 100 gains trust with the user through dialogue, it can obtain appropriate information, such as advertisements, and generate a response based on those advertisements. When a user has gained trust, it means the user is more receptive to advertisements, and in such cases, generating a response based on advertisements is more likely to be accepted by the user. On the other hand, if an advertisement is presented when trust has not been established, it is highly likely to be unpleasant for the user. Therefore, when trust has not been established, the dialogue continues in order to build that trust.

[0059] In this disclosure, the dialogue server 100 and the user terminal 200 communicate via voice, but the communication is not limited to voice; it may also be conducted using text or other means.

[0060] Furthermore, in the dialogue server 100, the emotion recognition unit 102 recognizes the user's emotions based on the user's utterances. The confidence level measurement unit 103 then measures the confidence level based on these emotions.

[0061] According to this disclosure, trust levels can be measured while also considering user emotions. Therefore, it becomes possible to more accurately determine what state is most acceptable to the user and generate responses based on the information presented, such as advertisements, that are appropriate to that state.

[0062] Furthermore, in this disclosure, the speech history storage unit 108 stores the user's speech history information. The confidence level measurement unit 103 measures the confidence level based on the speech history information in addition to the speech itself.

[0063] In this disclosure, it is possible to generate a response using presented information that corresponds to the level of confidence based on past utterance history.

[0064] In this disclosure, confidence can be measured more accurately by using both emotion and speech history, but either emotion or speech history may be used.

[0065] Furthermore, in this disclosure, the dialogue server 100 includes an internal thought estimation unit 104. This internal thought estimation unit 104 estimates the user's internal thought based on the speech history information stored in the speech history storage unit 108. The response generation unit 109 then generates a response based on the internal thought. That is, the decision unit 105 determines an action to take towards the user based on the confidence level and the internal thought. For example, the decision unit 105 decides whether to present an advertisement or continue the dialogue.

[0066] This configuration allows for determining effective actions for users based on their inner characteristics. These inner characteristics include, for example, their personality, awareness, and at least one of their cognitive biases. These influence how receptive users are to advertisements and other communications. Therefore, determining actions for users based on their inner characteristics is crucial.

[0067] In this disclosure, the decision unit 105 includes a DQN, which is a decision model constructed by reinforcement learning, that takes the user's current state as input and determines an action to take on the user. The decision unit 105 uses the DQN to determine the information to be provided to the user as an action to take on the user, based on the user's current state. In this disclosure, the user's current state is at least one of the confidence level based on the most recent utterance or the confidence level based on utterance history information.

[0068] This configuration allows for more appropriate responses by using DQN to determine actions to take towards the user.

[0069] Furthermore, in this disclosure, the dialogue server 100 further includes an action history storage unit 108a that stores user action history information. The inner thoughts estimation unit 104 may estimate the user's inner thoughts based on the action history information.

[0070] In this disclosure, the dialogue server 100 further comprises an interest acquisition unit 106 that acquires the user's interests based on the user's utterances and a knowledge graph showing the relationships between words. The response generation unit 109 generates a response based on the information provided according to the user's prior tastes.

[0071] This configuration allows for the acquisition of topics and advertisements that are of interest to the user, and enables the provision of conversations or advertisements that are easily accepted by the user.

[0072] The apparatus and method described herein have the following configuration.

[0073] [1] A device comprising: a dialogue information acquisition unit that acquires dialogue information from a user; a measurement unit that measures the degree of confidence in the user's dialogue based on the dialogue information; and a response generation unit that generates a response based on the information provided according to the dialogue information, according to the degree of confidence.

[0074] [2] The apparatus according to [1], further comprising: a determination unit that determines an action to take toward the user according to the level of trust; and an interest acquisition unit that acquires topics or advertisements of interest to the user according to the dialogue information as provided information, wherein the response generation unit acquires the level of trust and the advertisements acquired by the interest acquisition unit as provided information, or acquires the topics acquired by the interest acquisition unit as provided information.

[0075] [3] The apparatus according to [1] or [2], further comprising: a recognition unit that recognizes the emotions of the user based on the dialogue information, wherein the measurement unit measures the confidence level based on the emotions.

[0076] [4] The apparatus according to any one of [1] to [3], further comprising: a dialogue history storage unit that stores the user's dialogue history information, wherein the measurement unit measures the reliability based on the dialogue history information in addition to the dialogue information.

[0077] [5] The apparatus according to any one of [1] to [3], further comprising: a dialogue history storage unit that stores dialogue history information relating to the user's dialogue and emotions; and an inner thoughts estimation unit that estimates the user's inner thoughts based on the dialogue history information, wherein the response generation unit further generates the response based on the inner thoughts.

[0078] [6] The apparatus according to [2], comprising a decision model constructed by reinforcement learning that takes the user's current state as input and determines an action to take on the user, wherein the decision unit uses the decision model to determine information to be provided to the user as an action to take on the user based on the user's current state.

[0079] [7] The device according to [6], wherein the current state of the user is at least one of the confidence level based on recent dialogue information or the confidence level based on dialogue history information.

[0080] [8] The apparatus according to [5], further comprising a behavior history storage unit for storing user behavior history information, wherein the internal personality estimation unit estimates the user's internal personality based on the behavior history information.

[0081] [9] The apparatus according to any one of [1] to [7], further comprising an interest acquisition unit that acquires the user's interests based on the user's dialogue information and a knowledge graph showing the relationships between words, and the response generation unit generates a response based on the information provided according to the user's interests.

[0082]

[10] A method for a device that interacts with a user terminal to process a dialogue, comprising: a dialogue acquisition step of acquiring user dialogue information; a measurement step of measuring the user's trustworthiness based on the dialogue information; and a response generation step of generating a response based on the provided information acquired according to the dialogue information, according to the trustworthiness.

[0083] The block diagram used in the description of the above embodiment shows functional units. These functional blocks (components) are realized by any combination of at least one of hardware and software. Furthermore, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one device that is physically or logically coupled, or it may be realized using two or more physically or logically separated devices that are directly or indirectly connected (for example, using wired or wireless connections). A functional block may be realized by combining the one or more devices with software.

[0084] Functions include, but are not limited to, judgment, decision, determination, calculation, calculation, processing, derivation, investigation, exploration, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, assumption, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating (mapping), and assigning. For example, a functional block (configuration part) that enables transmission is called a transmitting unit or transmitter. In all cases, as mentioned above, the method of implementation is not particularly limited.

[0085] For example, the dialogue server 100 and user terminal 200 in one embodiment of the present disclosure may function as a computer that processes the dialogue method of the present disclosure. Figure 14 is a diagram showing an example of the hardware configuration of the dialogue server 100 and user terminal 200 according to one embodiment of the present disclosure. The dialogue server 100 and user terminal 200 described above may be physically configured as a computer device including a processor 1001, memory 1002, storage 1003, communication device 1004, input device 1005, output device 1006, bus 1007, etc.

[0086] In the following explanation, the term "device" can be replaced with "circuit," "device," "unit," etc. The hardware configuration of the dialogue server 100 and the user terminal 200 may include one or more of the devices shown in the figure, or it may be configured to omit some of the devices.

[0087] Each function in the dialogue server 100 and the user terminal 200 is realized by loading predetermined software (programs) onto hardware such as the processor 1001 and memory 1002, which allows the processor 1001 to perform calculations, control communication by the communication device 1004, and control at least one of the reading and writing of data in the memory 1002 and storage 1003.

[0088] The processor 1001 controls the entire computer, for example, by running the operating system. The processor 1001 may be composed of a central processing unit (CPU) that includes interfaces with peripheral devices, control devices, arithmetic units, registers, etc. For example, the above-mentioned speech recognition unit 101, emotion recognition unit 102, confidence measurement unit 103, inner feelings estimation unit 104, judgment unit 105, interest acquisition unit 106, and response generation unit 109 may be implemented by the processor 1001.

[0089] Furthermore, the processor 1001 reads programs (program code), software modules, data, etc., from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes accordingly. The program used is one that causes the computer to execute at least a part of the operations described in the above embodiment. For example, the determination unit 105 may be implemented by a control program stored in the memory 1002 and running on the processor 1001, and other functional blocks may be implemented similarly. The above-described various processes have been explained as being executed by one processor 1001, but they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may also be transmitted from a network via a telecommunications line.

[0090] The memory 1002 is a computer-readable recording medium and may consist of at least one of the following: ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), etc. The memory 1002 may also be called a register, cache, main memory, etc. The memory 1002 can store executable programs (program code), software modules, etc., for implementing the interaction method according to one embodiment of the present disclosure.

[0091] The storage 1003 is a computer-readable recording medium and may consist of at least one of the following: an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disc, a digital multipurpose disc, a Blu-ray® disc), a smart card, flash memory (e.g., a card, a stick, a key drive), a floppy® disk, a magnetic strip, etc. The storage 1003 may also be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, server, or other suitable medium including at least one of memory 1002 and storage 1003.

[0092] The communication device 1004 is hardware (transmitting / receiving device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as a network device, network controller, network card, communication module, etc. The communication device 1004 may be configured to include high-frequency switches, duplexers, filters, frequency synthesizers, etc., in order to implement at least one of frequency division duplex (FDD) and time division duplex (TDD). For example, the part that inputs and outputs sound to the voice recognition unit 101 and the voice synthesis unit 110 described above may be implemented by the communication device 1004. The communication device 1004 may be implemented with physically or logically separated transmitting and receiving units.

[0093] The input device 1005 is an input device that accepts input from an external source (e.g., a keyboard, mouse, microphone, switch, button, sensor, etc.). The output device 1006 is an output device that outputs to an external source (e.g., a display, speaker, LED lamp, etc.). The input device 1005 and the output device 1006 may be configured as an integrated unit (e.g., a touch panel).

[0094] Furthermore, each device, such as the processor 1001 and memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or different buses may be configured for each device.

[0095] Furthermore, the dialogue server 100 and the user terminal 200 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array), and some or all of each functional block may be realized by such hardware. For example, the processor 1001 may be implemented using at least one of these hardware components.

[0096] Information notification is not limited to the embodiments described herein and may be carried out by other means. For example, information notification may be carried out by physical layer signaling (e.g., DCI (Downlink Control Information), UCI (Uplink Control Information)), upper layer signaling (e.g., RRC (Radio Resource Control) signaling, MAC (Medium Access Control) signaling, broadcast information (MIB (Master Information Block), SIB (System Information Block))), other signals, or combinations thereof. RRC signaling may also be called RRC messages, and may be, for example, RRC Connection Setup messages, RRC Connection Reconfiguration messages, etc.

[0097] The processing procedures, sequences, flowcharts, etc., of each aspect / embodiment described in this disclosure may be reordered, provided they do not contradict each other. For example, the methods described in this disclosure present various step elements using exemplary order and are not limited to the specific order presented.

[0098] Input and output information may be stored in a specific location (e.g., memory) or managed using a management table. Input and output information may be overwritten, updated, or appended to. Output information may be deleted. Input information may be transmitted to other devices.

[0099] The determination may be made by a value represented by one bit (0 or 1), by a boolean value (true or false), or by a numerical comparison (for example, a comparison with a predetermined value).

[0100] Each aspect / embodiment described in this disclosure may be used individually, in combination, or switched between as needed during implementation. Furthermore, notification of specific information (e.g., notification that "X is") is not limited to explicit notification, but may also be implicit (e.g., by not providing such notification).

[0101] Although the present disclosure has been described in detail above, it will be clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the intent and scope of the present disclosure as defined by the claims. Accordingly, the descriptions in the present disclosure are illustrative and not intended to be restrictive in any way.

[0102] Software should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, and so on, whether they are called software, firmware, middleware, microcode, hardware description languages, or by any other name.

[0103] Furthermore, software, instructions, information, etc., may be transmitted and received via a transmission medium. For example, if software is transmitted from a website, server, or other remote source using at least one of wired technologies (such as coaxial cable, fiber optic cable, twisted pair, or digital subscriber line (DSL)) and wireless technologies (such as infrared or microwave), then at least one of these wired and wireless technologies is included in the definition of a transmission medium.

[0104] The information, signals, etc. described in this disclosure may be represented using any of the various different techniques. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0105] In addition, terms used in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings. For example, at least one of the channel and symbol may be a signal (signaling). Also, a signal may be a message. Furthermore, a component carrier (CC) may be called a carrier frequency, cell, frequency carrier, etc.

[0106] Furthermore, the information, parameters, etc., described in this disclosure may be expressed using absolute values, relative values ​​from a given value, or other corresponding information. For example, wireless resources may be indicated by an index.

[0107] The names used for the parameters described above are not restrictive in any way. Furthermore, the formulas and other expressions using these parameters may differ from those expressly disclosed in this disclosure. Various channels (e.g., PUCCH, PDCCH, etc.) and information elements can be identified by any suitable name, and therefore, the various names assigned to these various channels and information elements are not restrictive in any way.

[0108] In this disclosure, terms such as "Mobile Station (MS)," "user terminal," "User Equipment (UE)," and "terminal" may be used interchangeably.

[0109] A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other appropriate term.

[0110] As used in this disclosure, the terms “determining” and “determining” may encompass a wide variety of actions. “Determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, or inquiring (e.g., searching in a table, database, or other data structure), or ascertaining. “Determining” may also include, for example, receiving (e.g., receiving information), transmitting (e.g., sending information), inputting, outputting, or accessing (e.g., accessing data in memory). Furthermore, "judgment" and "decision" can include considering something as having been "judged" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. In other words, "judgment" and "decision" can include considering something as having been "judged" or "decided" after some action. Also, "judgment (decision)" can be reinterpreted as "assuming," "expecting," or "considering."

[0111] The terms “connected,” “coupled,” or any variation thereof, mean any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” with each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, “connection” may be reinterpreted as “access.” As used in this disclosure, two elements may be considered to be “connected” or “coupled” with each other using at least one of one or more wires, cables, and printed electrical connections, and, in some non-limiting and non-exclusive examples, electromagnetic energy having wavelengths in the radio frequency domain, microwave domain, and optical (both visible and invisible) domain.

[0112] In this disclosure, the phrase "based on" does not mean "based solely on" unless otherwise specified. In other words, the phrase "based on" means both "based solely on" and "based at least on."

[0113] Any reference to elements using designations such as “first,” “second,” etc., as used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Accordingly, references to first and second elements do not imply that only two elements may be employed, or that the first element must precede the second element in any way.

[0114] Where the terms “include,” “including,” and their variations are used in this disclosure, these terms are intended to be inclusive, as is the term “comprising.” Furthermore, the term “or” as used in this disclosure is not intended to be exclusive OR.

[0115] In this disclosure, if articles are added by translation, such as a, an, and the in English, this disclosure may include the fact that the noun following these articles is plural.

[0116] In this disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "combine" may be interpreted similarly to "different."

[0117] 100...Dialogue server, 200...User terminal, 101...Speech recognition unit, 102...Emotion recognition unit, 103...Confidence level measurement unit, 104...Inner thoughts estimation unit, 105...Judgment unit, 106...Interest acquisition unit, 107...Advertisement memory unit, 108...Speech history memory unit, 108a...Action history memory unit, 109...Response generation unit, 110...Speech synthesis unit.

Claims

1. A device comprising: a dialogue information acquisition unit that acquires dialogue information from a user; a measurement unit that measures the level of confidence in the user's dialogue based on the dialogue information; and a response generation unit that generates a response based on the level of confidence and the information provided according to the dialogue information.

2. The apparatus according to claim 1, further comprising: a determination unit that determines an action to take toward the user according to the level of trust; and an interest acquisition unit that acquires topics or advertisements of interest to the user according to the dialogue information as provided information, wherein the response generation unit acquires the advertisements acquired by the interest acquisition unit as provided information, or acquires the topics acquired by the interest acquisition unit as provided information, according to the action determined by the determination unit.

3. The apparatus according to claim 1, further comprising: a recognition unit that recognizes the emotions of the user based on the dialogue information, wherein the measurement unit measures the reliability based on the emotions.

4. The apparatus according to claim 1, further comprising: a dialogue history storage unit for storing the user's dialogue history information, wherein the measurement unit measures the reliability based on the dialogue history information in addition to the dialogue information.

5. The apparatus according to claim 1, further comprising: a dialogue history storage unit that stores dialogue history information relating to the user's dialogue and emotions; and an inner thoughts estimation unit that estimates the user's inner thoughts based on the dialogue history information, wherein the response generation unit further generates the response based on the inner thoughts.

6. The apparatus according to claim 2, comprising a decision model constructed by reinforcement learning that takes the user's current state as input and determines an action to take on the user, wherein the decision unit uses the decision model to determine an action to take on the user based on the user's current state.

7. The apparatus according to claim 6, wherein the user's current state is at least one of the confidence level based on recent dialogue information or the confidence level based on dialogue history information.

8. The apparatus according to claim 5, further comprising an action history storage unit for storing user action history information, wherein the internal body estimation unit estimates the user's internal body based on the action history information.

9. The apparatus according to claim 1, further comprising an interest acquisition unit that acquires the user's interests based on the user's dialogue information and a knowledge graph showing the relationships between words, wherein the response generation unit generates a response based on the information provided according to the user's interests.

10. A method for processing a dialogue in a device that interacts with a user terminal, comprising: a dialogue acquisition step of acquiring user dialogue information; a measurement step of measuring the user's trustworthiness based on the dialogue information; and a response generation step of generating a response based on the provided information acquired according to the dialogue information, according to the trustworthiness.

Images