A system and method for delivering domain or use case switching suggestions for ongoing conversations.
The system addresses user confusion in dialogue managers by implementing confidence scoring and user confirmation for seamless transitions between use cases and domains, improving efficiency and user experience.
Patent Information
- Application Number
- JP2023579291
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2022-06-23
- Publication Date
- 2026-11-19
- Estimated Expiration
- 2042-06-23
AI Technical Summary
Existing dialogue managers struggle to handle interruptions, cancellations, and switches between use cases and domains in ongoing conversations, leading to user confusion and inefficiency.
A system and method for determining domain or use case switching suggestions in human-computer conversations, involving confidence scoring and user confirmation to ensure alignment with intended use cases and domains.
Reduces ambiguity and enhances user experience by efficiently managing transitions between use cases and domains, making the interface more user-friendly and cost-effective.
Smart Images

Figure 0007953917000001 
Figure 0007953917000002 
Figure 0007953917000003
Abstract
Description
TECHNICAL FIELD
[0001] The present invention relates generally to systems and methods for dialogue management, and more specifically, the present invention relates to determining transitions from a use case based on an updated use case for interaction between a user and a system, and providing services to the user. BACKGROUND ART
[0002] The field of human-machine interaction has seen several significant improvements over the past few years with new breakthroughs in the artificial intelligence field, forming fundamental impacts and thereby transforming all sectors of the economy. As the number of online services, such as customer service, and information management tasks including retrieving specific information from vast amounts of data increases, various systems are configured to assist users in performing tasks. There are various modes by which systems enable users to retrieve information as desired by the user via various services. These services typically operate in association with the organization to which they belong, and often include specialized agents designed to perform various functions. These agents are either human agents in the system or dialogue managers. Over time, the use of dialogue managers that provide users with services similar to those provided by human agents has been increasing. However, dialogue managers have limitations, and these dialogue managers are also expected to be able to engage in interactions with users to perform tasks as efficiently as human agents do.
[0003] As users are increasingly able to interact with dialogue managers in natural language conversations, existing dialogue managers often find it difficult to handle interruptions, cancellations, and switches between use cases and domains in ongoing conversations.
[0004] A domain can represent a specific content area of a relevant conversation between a user and a conversation manager. Examples of domains, but not limited to, include search, banking, shopping, movies, music, restaurants, flights, and call centers.
[0005] A use case can represent activities associated with a domain to fulfill an intent. In the present invention, the goal of a use case is to communicate the most important aspects of user-system interaction by focusing on user intent. Examples of use cases, but not limited to these, include activities and interactions between a user and a dialogue manager in the active domain of banking transactions, such as transferring money to an account, paying credit card debt, or updating credentials.
[0006] An intent represents the action or purpose that a user wants to achieve by participating in a conversation with the conversation manager. Examples of intents include, but are not limited to, user requests such as initiating an action, finding content, restarting a process, going back, or finding information. An example of an intent to initiate an action would be a user making a request to transfer money from their account within the corresponding use case.
[0007] For example, in a scenario where a user and a conversation manager are conversing within a specific use case in a domain (e.g., querying an account balance), the user may transition to a different use case midway through the conversation (e.g., updating account holder information). As a result, the conversation manager may not be able to immediately recognize and smoothly handle the use case switch. The conversation manager may need additional information to anticipate and respond to the user's transition. In that case, the user may want to transition to a completely different domain (e.g., a bicycle configurator), which the conversation manager may find difficult to follow accurately.
[0008] Currently, conversation managers have limitations in recognizing user inconsistencies with the active use case or domain based on analysis of ongoing conversations. As a result, the system cannot mitigate uncertainty and / or confusion regarding user input, and may provide output that may not align with the intended use case. Consequently, users may become reluctant to use it in the future, feeling unable to find a user-friendly interface and feeling constrained by use cases and domains they do not wish to adhere to. [Overview of the project]
[0009] This summary is provided to introduce, in a simplified form, a selection of concepts that will be further explained in the detailed description below. This summary is not intended to identify any significant or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.
[0010] Embodiments of the present invention relate to a method and system for dialogue management in human-computer conversations, which involves determining a transition between domains or use cases and initiating a domain or use case switching suggestion to a user. The switching suggestion method includes determining a first domain or use case in a human-computer conversation and assigning a first confidence score to the first domain or use case. Next, the switching proposal method includes determining a second domain or use case in a human-computer conversation and assigning a second confidence score to the second domain or use case. Furthermore, the switching proposal method includes deciding whether to switch the domain or use case to the second domain or use case based on the first and second confidence scores.
[0011] As a result, the present invention can be used to measure and reduce uncertainty or confusion surrounding user input that may not align with the intended use case within the system. A user (who does not want to interrupt their use case activity / is unaware of the possibility of interruption) is asked by the system whether they wish to follow a given use case when the system is uncertain, thereby adjusting the system to extract information intended for / from the user. As a result, ambiguity in domain or use case detection is reduced. This makes the interface more efficient, user-friendly, and easy to use. Furthermore, it saves time and costs for both system operation and user use. [Brief explanation of the drawing]
[0012] [Figure 1A] This block diagram shows the data flow between the user and the system for determining use case switching suggestions. [Figure 1B] This block diagram shows the data flow between the user and the system for determining use case switching suggestions, excluding the speech feature extractor. [Figure 1C] This is a Venn diagram showing the various levels that the system can switch between. [Figure 2] This is a detailed block diagram showing an exemplary switching proposal module. [Figure 3] This flowchart shows a computing system configured to score a single use case in a human-computer conversation. [Figure 4] This flowchart shows a computing system configured to score multiple use cases in human-computer conversations. [Figure 5] This flowchart shows exemplary scenarios for user-triggered and system-triggered use case switching. [Figure 6]This flowchart shows an exemplary scenario of a system-triggered use case switch. [Modes for carrying out the invention]
[0013] This specification describes systems and methods for determining domain or use case switching suggestions in human-computer conversations. These systems and methods are illustrated with reference to figures, which are intended to be illustrative rather than limiting, in order to facilitate the description of exemplary systems and methods relating to embodiments of the present invention.
[0014] The foregoing description of specific embodiments is intended to fully illustrate the general nature of the embodiments herein. Others may readily modify and / or adapt such specific embodiments to various uses without departing from the general concept by applying current knowledge, and such adaptations and modifications should and are intended to be understood within the meaning and scope of the equivalents of the disclosed embodiments.
[0015] Furthermore, it should be noted that embodiments may be described as processes shown as flowcharts, flow diagrams, data flow diagrams, structural diagrams, or block diagrams. While flowcharts can describe operation as a series of processes, many of these operations may be performed in parallel or simultaneously. Moreover, the order of operations may be reordered. A process terminates when its operations are complete, but it may have additional steps not shown in the diagram. A process may correspond to a method, function, procedure, subroutine, subprogram, etc. If a process corresponds to a function, its termination corresponds to the function's return to the calling function or main function.
[0016] It should be understood that any expressions or terms used herein are for illustrative purposes only and not for limitation. Therefore, while embodiments herein have been described in relation to preferred embodiments, those skilled in the art will recognize that embodiments herein can be implemented with modifications within the spirit and scope of the appended claims.
[0017] As used herein, the term “network” refers to any form of communication network used to carry data and connect communication devices (e.g., telephones, smartphones, computers, servers) to one another. According to embodiments of the present invention, data includes at least one of processed data and raw data. Such data includes data obtained by automated data processing, manual data processing, or raw data.
[0018] As used herein, the term "artificial intelligence" refers to a set of executable instructions stored on a server and generated using machine learning techniques.
[0019] As used herein, the term “communication control module” refers to an electronic device comprising a processor, which may be connected to other electronic / computing devices, displays, and sensors via a network.
[0020] In the following explanation, terms such as "first," "second," etc., are used to describe various elements, but these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, without deviating from the scope of the various examples provided, the first intent may be called the second intent, and similarly, the second intent may be called the first intent.
[0021] FIG. 1A shows a conceptual diagram illustrating an exemplary framework of data flow between a human and an exemplary computer communication system 100 for determining a handoff proposal, in accordance with one or more aspects of the present invention. In this example, the disclosed system includes a communication control module 104, a speech feature extractor 105, a dialogue engine 106, a natural language understanding module 107 (hereinafter referred to as NLU module 107), a handoff proposal module 108, a user model 109, and a conversation history module 110.
[0022] Referring back to FIG. 1A, a user 101 uses a client device 102 to initiate a call to the system 100. The call is transmitted via the telephony gateway 103 of the carrier network, and further routed to the communication control module 104. The client device 102 corresponds to a wide variety of electronic devices. According to an embodiment of the present invention, the client device 102 is any telecommunication device such as a smartphone or a feature phone, or a regular landline telephone. The client device 102 functions as a service request unit for inputting a user request.
[0023] The user 101 uses the client device 102 via a call to receive information and input service requests, such as initiating various processes that may be part of the service request. The communication control module 104 provides a plurality of communication links for managing the overall functions of the system 100, and functions as an entry point of the user 101 to the system 100. Calls received by the communication control module 104 are further routed to the dialogue engine 106. After that, the dialogue engine 106 receives the user's conversational input.
[0024] The communication control module 104 is further connected to the speech feature extractor 105 and the dialogue engine 106.
[0025] When the telephone gateway 103 receives a call via the telephone network, it converts the conversation with user 101 into a digital audio file / stream that can be transmitted over the data network and received by the communication control module 104, for example, when a request is entered. The communication control module 104 controls the input and output audio and routes it to the speech feature extractor 105 and the dialogue engine 106. The speech feature extractor 105 can receive and listen to the audio, and the dialogue engine 106 can convert the audio into machine-readable text using one or more methods, including automatic speech recognition (ASR). The communication control module 104 can convert the audio signal received from user 101 into a digital format for further processing, such as an audio file. The communication control module 104 can also detect speech activation and the end / start of an utterance. The speech feature extractor 105 is configured to receive real-time audio data input transmitted from the communication control module 104. The voice feature extractor 105 is further configured to extract one or more predetermined voice features from the voice data when it receives voice data input from the communication control module 104. The voice features may include, for example, pitch, waveform, etc., which may be required by the switching suggestion module 108 for further speech processing.
[0026] The communication control module 104 can also receive speech features extracted from the speech feature extractor 105. The communication control module then transmits the extracted speech features to the dialogue engine 106. The dialogue engine 106 and the NLU module 107 then use one or more techniques for further processing and to transmit the extracted speech features further to the switching suggestion module 108.
[0027] In yet another embodiment of the present invention, the speech feature extractor 105 may be omitted. Figure 1B shows a conceptual diagram illustrating an exemplary framework system 150 for such an embodiment. For illustrative purposes only, in this embodiment, the chatbot connector 111 sends and receives text data between the chatbot app / web page on the client device 102 and the dialogue engine 106. The chatbot connector 111 may be, for example, a communication platform that connects a user 101 to the dialogue engine 106.
[0028] Referring back to Figure 1A, when user 101 inputs a voice service request to the dialogue engine 106 using client device 102, the user's input voice is processed using a method and system for speech-to-text conversion of the user's incoming voice. In this example, artificial intelligence technology is used to collect data from several users or individual users and extract key terms that indicate user preferences.
[0029] The dialogue engine 106 provides a user interface between the user and the service, primarily by engaging in natural language dialogue with the user. The dialogue may include questions requesting one or more aspects of a particular service, such as asking for information. In this way, the system 100 can also receive general conversational queries and engage in ongoing conversations with the user via the dialogue engine 106. The dialogue engine 106 can further switch domains and use cases by recognizing new intents / use cases / domains from the user during the conversation and adding feedback on previously generated switching suggestions.
[0030] The dialogue engine 106 further connects to the NLU module 107 and the switching suggestion module 108. The NLU module 107 receives input from the dialogue engine 106 and converts the natural language input into machine-readable information. The NLU module 107 determines and generates the rewritten context, intent, use case, entity, and metadata of the conversation with user 101. The NLU module 107 uses natural language processing to determine the use case from the user's conversation input. The switching suggestion module 108 determines and generates a use case threshold score and a use case switching suggestion based on the use case threshold score as output. The use case threshold may be determined automatically. The switching suggestion module 108 can also generate scores associated with use cases corresponding to active domains in the conversation. For example, the switching suggestion module 108 determines a first use case in the conversation from the user to system 100 and assigns a first use case score to the first use case. Next, the switching suggestion module 108 determines a second use case in the user's conversation with system 100 and assigns a second use case score to the second use case.
[0031] The use case score indicates the degree of confidence (e.g., probability or other likelihood) that the purpose of the dialogue between user 101 and dialogue engine 106 satisfies user 101's intent. The switching suggestion module 108 may initiate a use case switching suggestion or refrain from initiating a use case switching suggestion in the conversation depending on whether the use case score meets a predetermined use case score threshold. For example, if the use case score is less than a predetermined use case score threshold (e.g., 50%), the switching suggestion module 108 refrains from initiating a use case switching suggestion based on the use case. If the use case score exceeds or meets the predetermined use case score threshold, the switching suggestion module 108 initiates a use case switching suggestion based on the use case.
[0032] Venn diagrams or other techniques may be used to define or visualize the level of switching in a conversation between the user and system 100. Figure 1C shows the use of a Venn diagram for this purpose. Set S161 represents the domain of the conversation. According to an exemplary embodiment of the present invention, the domain may represent a content area of the conversation, such as search, banking, shopping, movies, music, restaurants, or flights.
[0033] Set T162 illustrates a conversational use case. According to exemplary embodiments of the present invention, a use case may represent interactions and activities associated with a domain to satisfy an intent. Examples of use cases, but not limited to these, include activities and interactions between a user and system 100 corresponding to an active domain of banking transactions, such as transferring money to an account, paying credit card debt, or updating credentials.
[0034] Set U163 indicates the intent of the conversation. According to an exemplary embodiment of the present invention, the intent represents an action or objective that the user wishes to achieve by participating in a conversation with system 100. Examples of intents include, but are not limited to, user requests such as initiating an action, finding content, redoing a process, going back, and finding information. An example of an intent to initiate an action includes the user making a request to transfer money from their account within a corresponding use case.
[0035] Set 164 includes three sets: Set U163, Set T162, and Set S161. Set 164 shows the overlapping nature of Set U163 with Set T162 and Set T162 with Set S161. Set 164 represents the various levels at which System 100 can initiate and perform switching during an ongoing conversation. Set 164 further shows that the intents indicated by Set U163 are a subset of Use Case Set T162, which indicates that intents are a part and, for the present invention, a key point or ultimate goal of the use cases. Furthermore, Set 164 also indicates that its sets U163 and T162, i.e., intents and use cases are a subset of Set S161, which indicates that intents and use cases are domain-specific.
[0036] In an exemplary scenario, user 101 is in a conversation with system 100 in the active “banking” domain and “transfer” use case to an account. Next, user 101 may suddenly transition to a completely different use case, such as “recipient management” for their own account. Next, the switch suggestion module 108 recognizes that the confidence score for the “recipient management” use case has exceeded a predetermined threshold and / or corresponds to the highest confidence score among other identified use cases. Next, the switch suggestion module 108 may initiate a use case switch suggestion to user 101 via the dialogue engine 106 based on the “recipient management” use case.
[0037] In the present invention, the switching suggestion module 108 can further initiate a domain switching suggestion in a conversation with user 101 if the score determined for a domain is above the highest and / or predetermined scoring threshold for that domain. For example, user 101, who is conversing with system 100 under the active "banking transactions" domain and "transfer" use case, may suddenly switch to a completely different domain such as "bicycle configuration". The switching suggestion module 108 then recognizes that the confidence score for the "bicycle configuration" domain has exceeded a predetermined threshold and / or has the highest confidence score among the other confidence scores corresponding to multiple identified domains. The switching suggestion module 108 then may initiate a domain switching suggestion to user 101 via the dialogue engine 106 based on the "bicycle configuration" domain.
[0038] In this invention, the switching suggestion module 108 can switch between various levels during a conversation with the user. For example, user 101, who is in a conversation with system 100 in the active domain regarding "banking transactions," may suddenly switch to a completely different domain, such as "bicycle configuration," and then to a use case corresponding to the intention of "purchasing a specific bicycle configurator." The switching suggestion module 108 then recognizes the confidence scores of each domain and use case accordingly and initiates a switching suggestion to user 101 via the dialogue engine 106 based on the highest confidence score.
[0039] In yet another embodiment of the present invention, the user model 109 is connected to the dialogue engine 106 and the switching suggestion module 108. The user model 109 may include a data structure or other data unit configured to store information about the user for providing personalized services. The user model 109 may include various information about the user, such as user 101. For example, the user model 109 may include one or more identifiers for user 101, such as a user ID, username, and other self-identifying information of the user. The user model 109 may also include user demographic information, such as the user's age, gender, occupation, educational information (e.g., education level, degree, etc.), or one or more places associated with the user (e.g., the user's home, the user's workplace, places the user frequently visits, etc.). The user model 109 may be generated over one or more interactions with user 101. For example, each time the system 100 in Figure 1A interacts with user 101, the system 100 may learn new information about user 101. The system 100 may then store the new information in the user model 109 for use in future interactions. Therefore, the dialogue engine 106 and the switching suggestion module 108 utilize the user model 109 to make decisions about the user based on the use case of the active domain. The user's conversational input and user contextual characteristics are stored and retrieved remotely from at least one of the following: the user model, local storage media, a web interface, an application programming interface (API), and a command-line interface or console.
[0040] According to yet another embodiment of the present invention, the system 100 provides personalized services to user 101 based on the use cases of the active domain and various user-related information data stored in user model 109.
[0041] The switching suggestion module 108 further connects to the conversation history module 110. The conversation history module 110 stores data from past conversations associated with one or more users within the system 100. The data includes various pieces of information, such as contextual characteristics that can help improve domain / use case / intent detection. The data from the conversation history module 110 is used by the switching suggestion module 108 to determine and generate refined domains / use cases in order to further determine the user's more accurate intent in the conversation. The switching suggestion module 108 can provide a first intent and a second intent corresponding to a first use case and a second use case, respectively. The first and second intents are determined by analyzing the user's conversation input with a machine learning classifier 201.
[0042] Figure 2 is a detailed block diagram showing an exemplary switching suggestion module 108 according to one or more embodiments of the present invention. Figure 2 includes a switching suggestion module 108 as an example of the switching suggestion module 108 of Figure 1A.
[0043] The switching proposal module 108, described in detail below in the context of system 100 in Figure 1A, includes a machine learning classifier 201 connected to a text and dialogue-based feature module 202, a speech-based feature module 203, a combined training data module 206, and a confidence score module 207. The combined training data module 206 is further connected to the text and dialogue-based training data module 204 and the speech-based training data module 205.
[0044] Figure 2 shows only one specific example of the switching proposal module 108; many other examples of the switching proposal module 108 may be used in other examples and may include a subset of the components included in the examples. The switching proposal module 108 may include additional components not shown in Figure 2.
[0045] The switching suggestion module 108 is a module that determines whether or not to suggest a use case switch from a first use case to a second use case during a conversation between user 101 and system 100, based on relevant scores determined from information received through voice and utterance data. The switching suggestion module 108 can provide a first intent and a second intent corresponding to the first and second use cases, respectively. The first and second intents are determined by analyzing the user's conversation input using a machine learning classifier 201.
[0046] The switching proposal module 108 can also initiate a domain switching proposal in a conversation with user 101 if it is determined that the domain score is the highest and / or is above a predetermined scoring threshold for the use case.
[0047] Referring back to Figure 2, the machine learning classifier 201 uses text and dialogue-based feature data from the text and dialogue-based feature module 202, along with speech-based feature data from the speech-based feature module 203 and combined training data from the combined training data module 206, to classify multiple use cases and help determine and assign confidence scores for each use case. Although not shown herein, the machine learning classifier 201 of the switching proposal module 108 connects to the dialogue engine 106 of system 100 shown in Figure 1A.
[0048] In yet another embodiment of the present invention, the machine learning classifier 201 uses only text and dialogue-based feature data as well as combined training data to help classify multiple use cases and determine and assign confidence scores for each use case.
[0049] The text and dialogue-based feature module 202 stores text and dialogue-based feature data from user conversations, including, for example, term frequency inverse document frequency (TF-IDF), sentiment analysis, word embeddings, user timeouts, and recognition failures. This text and dialogue-based feature data can be extracted by the text and dialogue-based feature module 202 from the NLU module 107 and the conversation history module 110, as shown in Figure 1A.
[0050] The speech-based feature module 203 stores speech-based feature data from user conversations received from the speech feature extractor 105, including, for example, waveforms (zero crosses, extreme values), root mean square energy, Mel-frequency cepstrum coefficients (MFCCs), and pitch. Speech-based feature data from the speech-based feature module 203 of the last user input, such as user 101, and from previous user inputs (if available), are received by the machine classifier 201. The machine learning classifier 201 uses the data to help determine confidence scores for use case switching proposals and predict user intent.
[0051] The combined training data 206 receives training data from text and dialogue-based training data modules 204 and voice-based training data modules 205 to define use cases that may indicate specific intentions of user 101 in conversations with system 100, when correlated with specific use cases in the active domain. The training data can be further used to build statistical models or policies and train the dialogue engine 106.
[0052] The machine learning classifier 201 analyzes the combined training data from the combined training data 206 and may use deep learning on the information received from the data to train, develop, and update machine learning models to predict actions, and therefore intentions related to services, queries, and context. The machine learning classifier 201 determines the relevant context to use from conversational input from the user.
[0053] The machine learning classifier 201 maps each use case to its corresponding confidence score, as shown in the confidence score module 207. In this example, the confidence score module 207 represents multiple use cases and their corresponding confidence scores. The machine learning classifier 201 analyzes the user's conversation input to determine a first and second use case from the multiple use cases. Then, based on the confidence scores, the machine learning classifier 201 decides whether to initiate a use case switching suggestion from the first use case to the second use case to a user such as user 101 from the multiple use cases in the active domain during the conversation. The confidence scores may provide a numerical measure of multiple concepts, including, for example, conversational coordination, balance, and emphasis.
[0054] Figure 3 is a flowchart illustrating the steps of predicting and scoring use cases in order to perform use case switching suggestions in a human-computer conversation, according to one or more aspects of the present disclosure.
[0055] Figure 3 illustrates, as needed, a method for predicting and executing switching suggestions in the context of the system 100 in Figure 1A and the switching suggestion module 108 in Figure 2. For example, the switching suggestion module 108 further performs steps 301 to 316 for predicting and executing switching suggestions as needed, according to one or more embodiments of the present invention. In step 301, when user 101 calls system 100 and initiates a dialogue, corresponding data related to voice, text, dialogue, user model, and conversation history is sent to and from the switching suggestion module 108.
[0056] In the next step in step 302, the machine learning classifier 201 of the switching proposal module 108 extracts relevant feature data from incoming data that can define a use case.
[0057] In the next step in step 303, the machine learning classifier 201 of the switching proposal module 108 loads a machine learning model. The machine learning model may be trained early using relevant training data received from the combined training data store 206. The machine learning model may be used to determine, predict, and score one or more use cases that can be identified in the running use cases in the conversation with user 101.
[0058] In the next step in step 304, the machine learning classifier 201 predicts intent and scores the use cases identified in the active domain during the conversation with user 101. Then, confidence scores are assigned to the use cases, as shown in Figure 2.
[0059] In the next step in step 305, the machine learning classifier 201 determines whether the received use case confidence score meets or exceeds the minimum scoring threshold. Next, the switching suggestion module 108 provides the dialogue engine 106 with information on whether to send a use case switching suggestion in response to the user 101. For example, if the confidence score of the use case is less than the minimum scoring threshold (e.g., 50%), the switching suggestion module 108 may refrain from initiating a use case switching suggestion based on the user 101's intent. If the score is equal to or greater than the minimum scoring threshold, the switching suggestion module 108 may initiate a use case switching suggestion based on the user 101's intent. The minimum scoring threshold may be determined automatically or predetermined by the administrator of system 100.
[0060] In yet another embodiment of the present invention, in step 305, the dialogue engine 106 determines whether the received use case confidence score meets or exceeds a minimum scoring threshold, and whether to initiate a use case switching suggestion to the user based on the use case confidence score.
[0061] In the next step in step 307, if it is determined that the use case compliance confidence score does not meet or exceed the minimum scoring threshold, the dialogue engine 106 generates a response to help user 101 maintain the association with the previous use case in the active domain of the conversation. The switching suggestion module 108 106 may further confirm that it does not generate a use case switching suggestion.
[0062] In the next step in step 308, if the received use case compliance confidence score meets or exceeds the minimum scoring threshold, the dialogue engine 106 generates a use case switching proposal for the response to user 101.
[0063] In the next step in step 309, the dialogue engine 106 sends a response to the user 101 associated with a use case switching suggestion for identified use cases that exceed a threshold.
[0064] In the next step in step 310, user 101 needs to confirm with the dialogue engine 106 whether or not the user agrees to the use case switching proposal. Based on the user's approval in the conversation, the dialogue engine 106 can perform a use case switch from the first use case to the second use case. Approval includes requesting the user to confirm the use case switch.
[0065] In the next step in step 311, if user 101 does not agree to the use case switching proposal, the conversational engine 106 then continues the conversation with user 101 associated with the previous use case in the active domain.
[0066] In the next step in step 313, the dialogue engine 106 decides whether process 300 allows multiple use case switching suggestions in the active domain of the conversation with user 101.
[0067] In the next step in step 314, the dialogue engine 106 determines that process 300 will allow further use case switching suggestions if they are identified by the switching suggestion module 108 in the active domain within the conversation with user 101. The dialogue engine 106 then continues the conversation with user 101, and steps 301 to 310 are repeated for the conversation with user 101.
[0068] In the next step in step 315, the dialogue engine 106 determines that multiple use case switching suggestions are not permitted in the active domain of the conversation with user 101. Subsequently, the dialogue engine 106 continues the conversation with user 101 without further use case switching suggestions.
[0069] In the next step in step 312, user 101 agrees to the use case switching proposal. Subsequently, the dialogue engine 106 switches the use case in the active domain to the one that use case user 101 has agreed to through the conversation.
[0070] In the next step in step 316, after the use case switch, the dialogue engine 106 starts a new conversation with user 101 associated with use case user 101. The machine learning classifier 201 resets the constructs of the use case in the active domain within the new conversation with user 101.
[0071] If the dialogue engine 106 is identified by the switch proposal module 108 in the active domain within a new conversation with user 101, it determines whether process 300 is permitted to allow a further use case switch proposal. If permitted, the dialogue engine 106 continues the conversation with user 101, and steps 301 to 310 are repeated for the conversation with user 101.
[0072] If the dialogue engine 106 determines that multiple use case switching proposals are not permitted in the active domain of the new conversation with user 101, the dialogue engine 106 continues the conversation with user 101 without further use case switching proposals.
[0073] Figure 4 is a flowchart illustrating a method for predicting and scoring multiple use cases to perform use case switching suggestions in a human-computer conversation, according to one or more aspects of the present disclosure.
[0074] Figure 4 is described below in the context of the system 100 in Figure 1A and the switching suggestion module 108 in Figure 2. For example, the switching suggestion module 108 further performs steps 401 to 411 to predict and execute switching suggestions as needed, according to one or more embodiments of the present invention.
[0075] In step 401, the extracted features may include, for example, data related to speech, text, dialogue, user models, and conversation history sent to and received by the machine learning classifier 201 of the switching suggestion module 108.
[0076] In the next step in step 402, the machine learning classifier 201 of the switching proposal module 108 loads a machine learning model. The machine learning model is used to determine and score multiple use cases that may be identified in the running use cases in the conversation with user 101.
[0077] In the next step in step 403, the machine learning classifier 201 predicts one or more intentions and scores the multiple use cases identified in the active domain within the conversation with user 101. The multiple use cases are assigned confidence scores, as shown in Figure 2. For example, the first and second use cases may be selected from multiple use cases identified during the conversation.
[0078] In the next step in step 405, the machine learning classifier 201 may determine multiple use cases along with their corresponding confidence scores and predict new possible intentions. Next, the machine learning classifier 201 determines whether any of the corresponding confidence scores of the received multiple use cases meet or exceed a minimum scoring threshold.
[0079] In yet another embodiment, in step 405, the dialogue engine 106 may be used to determine whether any of the multiple confidence scores received for the corresponding use case meet or exceed a minimum threshold score. The dialogue engine 106 may then determine whether to send a use case switching suggestion in response to the user 101.
[0080] In the next step in step 406, if it is determined that none of the multiple use case support confidence scores meet or exceed the minimum scoring threshold, the switching suggestion module 108 may confirm that it does not generate a use case switching suggestion. Subsequently, if no use case switching suggestion is initiated by the switching suggestion module 108, the dialogue engine 106 continues the conversation with user 101 associated with the previous use case in the active domain.
[0081] In the next step in step 407, if any of the received use case confidence scores meet or exceed the minimum scoring threshold, a use case switching proposal is determined and initiated. Subsequently, the dialogue engine 106 generates a use case switching proposal for the use case with the highest confidence score and / or that exceeds the minimum scoring threshold, in order to respond to the user 101.
[0082] In the next step in step 408, the dialogue engine 106 sends the user 101 the response associated with the use case switching suggestion for the identified use case that has the highest score and / or exceeds the threshold.
[0083] If user 101 does not agree to the use case switching suggestion in the next step in step 411, the dialogue engine 106 continues the conversation with user 101 associated with the previous use case in the active domain within the conversation.
[0084] In the next step in step 410, if user 101 agrees to the switching proposal, a use case switch is performed. Subsequently, the dialogue engine 106 switches the use cases in the active domain to the use case associated with the highest confidence score, and user 101 further agrees.
[0085] Figure 5 is a flowchart illustrating an exemplary scenario of process 500, executed by an exemplary computing system configured for user-triggered use case switching.
[0086] Figure 5 is described below in the context of system 100 in Figure 1A.
[0087] In step 501, user 101 invokes system 100 via communication control module 104 with a query relating to a specific use case, such as a bicycle configurator. In the next step in step 502, the dialogue engine 106 asks user 101 questions that clarify one or more aspects of a specific request related to the use case, such as asking for missing information about the bicycle configurator to collect relevant data to guide user 101. In the next step in step 503, user 101 answers the questions asked by the dialogue engine 106. In the next step in step 504, user 101 realizes that they may have changed their mind about their previous request relating to the use case related to the bicycle configurator and have a query about a second use case, such as purchasing and configuring an electric bicycle instead. In the next step in step 505, user 101 explicitly commands the dialogue engine 106 that they want to purchase and configure an electric bicycle instead. In the next step in step 506, the dialogue engine 106 then switches the use case from the bicycle configurator to the electric bicycle configurator and guides the user 101 accordingly with the relevant data. In the next step 507, the configuration of the newly generated use case is resumed.
[0088] Figure 6 is a flowchart showing an exemplary scenario of process 600 executed by an exemplary computing system configured for system-triggered use case switching.
[0089] Figure 6 is described below in the context of system 100 in Figure 1A and the switching proposal module 108 in Figure 2. Operations 601-608 are performed in a different order than those shown in Figure 5, and using additional or fewer operations than those shown.
[0090] In step 601, user 101 invokes system 100 via communication control module 104 using a query related to a specific use case, such as a bicycle configurator. In the next step in step 602, the dialogue engine 106 asks user 101 questions that clarify one or more aspects of a specific request related to the use case, such as asking for missing information about the bicycle configurator to collect relevant data to guide user 101. In the next step in step 603, user 101 answers the questions asked by the dialogue engine 106. In the next step in step 604, the switching suggestion module 108 determines that the confidence score corresponding to the electric bicycle configurator use case is the highest and s exceeds the minimum scoring threshold. Next, the switching suggestion module 108 initiates a use case switching suggestion for the electric bicycle configurator use case. In the next step in step 605, the dialogue engine 106 sends a response to user 101 suggesting that the electric bicycle be better configured and purchased. In the next step in step 606, user 101 agrees to the proposal made by the dialogue engine 106. In the next step in step 607, the dialogue engine 106 switches the use case from the bicycle configurator to the electric bicycle configurator and guides user 101 accordingly with the relevant data. In the next step in step 608, the configuration of the newly generated use case is resumed.
[0091] In yet another exemplary scenario of the present invention, during a conversation in a use case in an active domain with a user, the dialogue engine 106 may be unable to collect the information necessary to satisfy the user's intent. For example, the dialogue engine 106 may be bound by a situation in which either user 101 has not provided the necessary information or the information provided is incomprehensible to the dialogue engine 106. In such a situation, the dialogue engine 106 may suggest to user 101 that the conversation be transferred to a human agent. According to one embodiment of the present invention, in such a situation, if the user agrees to the switching suggestion, the user is rerouted to a human agent to fulfill their intent. The use case in this situation is switched from the perspective of system 100, not from the user's perspective, since the user's intent remains the same. In this example, the use case switching is viewed from the system's perspective. According to embodiments of the present invention, in a method for determining an updated use case and providing a use case switching suggestion in a human-computer conversation, the conversation takes place over a telecommunications network. Furthermore, in the case of a system (100) for determining updated use cases and providing use case switching suggestions to both humans and computers, the conversation takes place over a telecommunications network.
[0092] The system and method for determining updated use cases according to the present invention and providing use case switching suggestions in human-computer conversations include use cases relating to services from at least one of the following: primary sector services, secondary sector services and tertiary sector services, legal services, banking or accounting services, restaurant services, health and fitness services, business and management services, medical / physical services, travel services, call center services, e-commerce services, education services, transportation services and government services.
[0093] The foregoing description of specific embodiments is intended to fully illustrate the general nature of the embodiments herein, so that others may readily modify and / or adapt such specific embodiments to various uses without departing from the general concept by applying their current knowledge, and such adaptations and modifications should and are intended to be understood within the meaning and scope of the equivalents of the disclosed embodiments. It should be understood that any expressions or terms used herein are for illustrative purposes only and not for limitation. Thus, although the embodiments herein have been described in relation to preferred embodiments, those skilled in the art will recognize that the embodiments herein can be modified and implemented within the spirit and scope of the appended claims.
[0094] The embodiments and examples discussed herein are for illustrative purposes only. It will be apparent that numerous other forms of the invention can be conceived by those skilled in the art without departing from the scope of the invention. The claims and embodiments are intended to encompass all such forms and modifications within the claims, and the description of the embodiments should not be construed as limiting the scope of the description.
Claims
1. A method for determining an updated use case between a user and a computer system and for providing a use case switching suggestion in a human-computer conversation, wherein the method is: a. A step of determining a first use case within the active domain in the conversation between the human and the computer, b. The step of assigning a first use case score to the first use case, c. A step of determining a second use case having the highest confidence score from among a plurality of use cases identified during the conversation in the active domain in the conversation between the human and the computer, d. The step of assigning a second use case score to the second use case, e. The process includes the step of determining whether to propose switching to the second use case by comparing the second use case score with a predetermined use case score threshold based on the first use case score and the second use case score, and initiating a proposal to switch to the second use case if the second use case score is equal to or greater than the predetermined use case score threshold, The domain represents a specific content area of a relevant conversation between the user and the conversation manager, the use case represents an activity associated with the domain to fulfill an intent, and the intent represents an action or purpose that the user wants to achieve by participating in a conversation with the conversation manager.
2. The step of determining the use case switching proposal to the second use case based on the first use case score and the second use case score is: a. A method for determining an updated use case and providing a use case switching proposal in a human-computer conversation, comprising the step of determining the use case score threshold and the use case switching proposal based on the use case score threshold, according to claim 1.
3. a. Further comprising initiating the use case switching proposal when the second use case score meets or exceeds a predetermined threshold, A method for determining an updated use case as described in claim 1 and for providing a use case switching suggestion in a human-computer conversation.
4. Determining the first use case and the second use case is a. The step of receiving the user's conversation input, b. A method for determining an updated use case according to claim 1 and providing a use case switching suggestion in a human-computer conversation, comprising the step of analyzing the user's conversation input using a machine learning classifier.
5. A method for determining an updated use case according to claim 4 and for providing a use case switching suggestion in a human-computer conversation, further comprising the step of providing a first intent and a second intent corresponding to the first use case and the second use case, respectively, wherein the first intent and the second intent are determined by analyzing the user's conversational input.
6. a. The step of performing a use case switch from the first use case to the second use case based on the user's approval in the conversation, wherein the approval includes requesting the user to confirm the use case switch. A method for determining an updated use case as described in claim 1 and for providing a use case switching suggestion in a human-computer conversation.
7. A method for determining an updated use case according to claim 1, wherein the first use case and the second use case are selected from a plurality of use cases identified in the conversation, and for providing a use case switching suggestion in a human-computer conversation.
8. A method for determining an updated use case according to claim 1, and for providing use case switching suggestions in a human-computer conversation, wherein the use case score provides a numerical measure of a plurality of concepts including at least one of conversational adjustment, balance, and emphasis.
9. A method for determining an updated use case according to claim 1, wherein the use case score threshold is automatically determined, and for providing a use case switching suggestion in a human-computer conversation.
10. a. When a use case switch is performed, the process further includes a step in which the machine learning classifier resets its constituent features for the new use case. A method for determining the updated use case described in claim 6 and for providing a use case switching suggestion in a human-computer conversation.
11. A method for determining an updated use case according to claim 5 and providing a use case switching suggestion in a human-computer conversation, wherein if the information necessary to satisfy one of the intents cannot be collected, the user's intent is not changed, and the use case switching is performed from a system perspective.
12. A method for determining an updated use case according to claim 1 and providing a use case switching suggestion in a human-computer conversation, wherein the user's conversational input and the user's contextual characteristics are remotely stored and retrieved from at least one of a user model, a local storage medium, a web interface, an application programming interface (API), a command-line interface, and a console.
13. The step of analyzing the user's conversation input is a. A step to detect voice activation, b. Steps to detect the end / beginning of an utterance in a conversation, c. A method for determining an updated use case and providing a use case switching suggestion in a human-computer conversation, further comprising the step of determining a use case from the user's conversation input using natural language processing.
14. A system (100) for determining an updated use case between a user and a computer system, and for providing a use case switching suggestion in a conversation between a human and a computer, Dialogue engine (106), Natural language understanding module (107), It includes a switching suggestion module (108), The switching proposal module 108 determines a first use case in the active domain in the conversation from the user to the system 100 and assigns a first use case score to the first use case; the switching proposal module 108 determines a second use case from among a plurality of use cases identified in the conversation between the user and the system 100 in the active domain that has the highest confidence score among them and assigns a second use case score to the second use case; further, based on the first use case score and the second use case score, it determines whether to propose a use case switching to the second use case by comparing the second use case score with a predetermined use case score threshold, and if the second use case score exceeds or satisfies the predetermined use case score threshold, the switching proposal module 108 initiates a use case switching proposal. The system (100) wherein the domain represents a specific content area of a relevant conversation between a user and a conversation manager, the use case represents an activity associated with the domain to fulfill an intent, and the intent represents an action or purpose that the user wants to achieve by participating in a conversation with the conversation manager.
15. A system (100) for providing a use case switching proposal in a human-computer conversation, wherein the switching proposal module (108) determines a use case threshold score, and the use case switching proposal determines an updated use case according to claim 14 based on the use case threshold score.
16. A system (100) for determining an updated use case and providing a use case switching suggestion in a human-computer conversation, wherein the switching suggestion module (108) determines a first use case in the human-computer conversation, assigns the first use case score to the first use case, determines a second use case in the human-computer conversation, and assigns the second use case score to the second use case.
17. A system (100) for determining an updated use case and providing a use case switching suggestion in a human-computer conversation, wherein the dialogue engine (106) receives user conversation input, and the switching suggestion module (108) uses a machine learning classifier (201) to analyze the user conversation input and determine the first and second use cases.
18. A system (100) for determining an updated use case and providing use case switching suggestions in a human-computer conversation, wherein the switching suggestion module (108) can further provide a first intent and a second intent corresponding to the first use case and the second use case, respectively, and the first intent and the second intent are determined by analyzing the user's conversation input by a machine learning classifier (201).
19. A system (100) for determining an updated use case and providing a use case switching suggestion in a human-computer conversation, wherein the dialogue engine (106) can perform a use case switching from the first use case to the second use case based on an acknowledgment from the user in the conversation, and the acknowledgment includes requesting confirmation from the user regarding the use case switching.
20. A system (100) for determining an updated use case according to claim 16, wherein the first use case and the second use case are selected from a plurality of use cases identified during the conversation, and for providing a use case switching suggestion in a human-computer conversation.
21. A system (100) for determining an updated use case according to claim 14, wherein the use case score includes a numerical measure of a plurality of concepts, including at least one of conversational adjustment, balance, and emphasis, and for providing use case switching suggestions in human-computer conversation.
22. A system (100) for determining a use case switching suggestion in a human-computer conversation according to claim 21, wherein the use case score is determined based on information received from the user's conversation input.
23. A system (100) for determining a use case switching suggestion in a human-computer conversation according to claim 15, wherein the use case score threshold is automatically determined.
24. A system (100) for determining a use case switching proposal in a human-computer conversation according to claim 17, wherein the machine learning classifier is reset for a new use case when a use case switching is performed.
25. A system (100) for determining an updated use case according to claim 18, wherein if the dialogue engine 106 is unable to collect the information necessary to satisfy the user's intent, the user is rerouted to a human agent, and the use case in this situation is switched from the perspective of the system 100 rather than the perspective of the user, since the user's intent remains the same, and for providing a use case switching suggestion in a human-computer conversation.
26. A system (100) for determining a use case switching suggestion in a human-computer conversation according to claim 14, wherein the user's conversational input and the user's contextual characteristics are remotely stored and retrieved from at least one of a user model, a local storage medium, a web interface, an application programming interface (API), a command line interface, and a console.
27. The aforementioned switching proposal module (108) The communication control module (104) and the natural language understanding module (107) are used to analyze the user's conversation input, and the communication control module (104) a. Detect voice activation, b. The natural language understanding module (107) detects the end / beginning of an utterance in a conversation and determines a use case from the user's conversation input. A system (100) for determining an updated use case as described in claim 14 and for providing a use case switching suggestion in a human-computer conversation.
28. A method for determining an updated use case according to claim 1, wherein the conversation takes place over a telecommunications network, and for providing a use case switching suggestion in a human-computer conversation.
29. A system (100) for determining an updated use case according to claim 14, wherein the conversation takes place via a telecommunications network, and for providing a use case switching suggestion in a human-computer conversation.
Citation Information
Patent Citations
Interaction support apparatus and interactive apparatus
JP2018142280A
Call center assisting apparatus, call center assisting system, call center assisting method, and call center assisting program
JP2019028783A
Dialogue system and domain determining method
JP2019070957A
Dialogue control method, dialogue control program, dialogue control device, information presentation method, and information presentation device
JP2020087352A
User-Programmable Automated Assistant
JP2020535452A
