Patent Application
20200234181
The present invention relates generally to the data processing field, and more particularly, relates to a method, system and computer program product for implementing enhanced training of a personality model for an embodied conversational agent.
A need exists for automatically configuring a multi-model conversational agent to display appropriate personality in an optimal way. Natural conversational interaction involves an emotional interaction as well as an informational interaction. A conversational participant might be annoyed or angered or pleased by their interlocutor's statement resulting from the content or the form of the interaction and this emotional reaction will often find expression in the form and content of a response. A need exists to find a right way to respond to a user from within a palette of emotional responses.
The patterning of emotional reactions is often perceived as constituting something like a personality or personal style. State of the art virtual conversational agents often have a flat emotional style, in which their contributions are unmodulated by variation in emotional content of their human users. This is perceived as unnatural, particularly as the range of emotional channels for input and feedback multiplies. While for a simple textual interface only the tone of the words is relevant, for embodied systems, such as avatars, with both spoken language and visual input and output a range of ways expressing an emotional response are available, in addition, a range of ways of perceiving the human user's emotional state is available, as well. To bring emotion to the interaction, it is useful to equip an embodied agent with a personality model.
A need exists for an efficient and effective mechanism to train a personality model for an emotionally intelligent embodied conversational system.
Principal aspects of the present invention are to provide a method, system and computer program product for implementing enhanced training of a personality model for an embodied conversational agent. Other important aspects of the present invention are to provide such method, system and computer program product substantially without negative effects and that overcome many of the disadvantages of prior art arrangements.
In brief, a method, system and computer program product are provided for implementing enhanced training of a personality model for an embodied conversational agent. An adjustable personality model is provided for the conversational agent interacting with a user with the adjustable personality model configured to provide emotion and tone responses based on detected emotions of the user and a communication objective. Responsive to detecting a first emotion of the user in a conversation with the conversational agent, the adjustable personality model is utilized to embody the conversational agent with a first emotion and tone. Responsive to detecting a second emotion of the user different from the first emotion, the adjustable personality model is utilized to embody the conversational agent with a second emotion and tone different from the first emotion and possibly tone.
In accordance with features of the invention, the personality model captures a pattern of emotional responses that characterize the personality of the cognitive agent, for example, the propensity to respond to anger with fear. A personality model is crucial to embodied conversational agents, because their emotional responses need to be consistently modulated across multiple modalities, such as timber of voice, facial expression, and the like. Adjustable personality models enable more natural interaction as well as emotionally artificial intelligent interaction.
In accordance with features of the invention, the adjustable personality model is based on identified user intents and a current ability to assist with the user intents.
In accordance with features of the invention, an emotion portrayed by the embodied conversational agent is selected from a group including agreeable, frustrated, apologetic, happy, sad, sympathetic, and angry.
In accordance with features of the invention, the detected emotions of the user are determined by an emotion detection system.
In accordance with features of the invention, the emotion detection system analyzes input selected from a group including audio, facial, text, keyboard pressure, and sensors.
The present invention together with the above and other objects and advantages may best be understood from the following detailed description of the preferred embodiments of the invention illustrated in the drawings, wherein:
In the following detailed description of embodiments of the invention, reference is made to the accompanying drawings, which illustrate example embodiments by which the invention may be practiced. It is to be understood that other embodiments may be utilized, and structural changes may be made without departing from the scope of the invention.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises” and/or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
In accordance with features of the invention, a method, system and computer program product are provided for implementing enhanced training of a personality model for an embodied conversational agent. An adjustable personality model is provided for the conversational agent interacting with a user with the adjustable personality model configured to provide emotion and tone responses based on detected emotions of the user and a communication objective. Responsive to detecting a first emotion of the user in a conversation with the conversational agent, the adjustable personality model is utilized to embody the conversational agent with a first emotion and tone. Responsive to detecting a second emotion of the user different from the first emotion, the adjustable personality model is utilized to embody the conversational agent with a second emotion and tone possibly different from the first emotion and tone.
Having reference now to the drawings, in
Computer system 102 includes a system memory 106 including an operating system 108, an embodied conversational agent system (avatar) control logic 110 and a multi-modal emotion detection system 111 in accordance with preferred embodiments. System memory 106 is a random-access semiconductor memory for storing data, including programs. System memory 106 is comprised of, for example, a dynamic random-access memory (DRAM), a synchronous direct random-access memory (SDRAM), a current double data rate (DDRx) SDRAM, non-volatile memory, optical storage, and other storage devices.
Computer system 102 includes a storage 112 including an adjustable personality model 114 in accordance with preferred embodiments and a network interface 116. Computer system 102 includes an I/O interface 118 for transferring data to and from computer system components including CPU 104, memory 106 including the operating system 108, embodied conversational agent system (avatar) control logic 110, multi-modal emotion detection system 111, storage 112 including adjustable personality model 114, and network interface 116, and a network 120 coupling user system inputs and system results 122 in accordance with preferred embodiments.
In accordance with features of the invention, the embodied conversational system (avatar) control logic 110 enables the conversational agent interacting with a user with the adjustable personality model configured to provide emotion and tone responses based upon detected emotions of the user and a communication objective.
In accordance with features of the invention, the embodied conversational agent is equipped with a trainable personality model 114. The trainable or adjustable personality model 114 describes the pattern of emotional responses characteristic of the personality, for example, respond to joy with joy; respond to anger with fear, and the like. Using the method of the invention, the embodied conversational agent learns how to generate emotional responses based on the detected occurrent emotional state of the user, the semantic intent, and a goal emotional of the user specified by a system builder. The input for the emotion detection system model 111 is, for example, audio-tone analysis, face-emotion recognition, text-based tone analysis, keyboard pressure analysis, and the like. This is mapped to a representation of the occurrent emotion of the user (E_u).
In accordance with features of the invention, the conversational system control logic 112 identifies the semantic intent of the user (I_u) and generates a content semantic intent of the system (I_s). The personality model 114 computes the emotional tone specification based on the recognized tone and the recognized semantic intent of the user and the semantic intent to be output: PM (E_u, I_u, I_s)=E_s. The personality model 114 is used to modulate the display of a semantic context by the artificial cognitive agent system control logic 110, potentially influencing, among other things, speech tone, facial expression, and textual tone of output, and the like. The modulation of the output (O_s) with a systems emotional tone specification (E_s) is orchestrated for example, by the display system (Express (O_s, E_s)).
In accordance with features of the invention, the cognitive conversational system control logic 112 and multi-modal emotion detection system 111, a method is provided for training a personality model as follows.
Referring to
Referring to
As indicated at a block 208, For each response expression, analyze the emotion of the response text using the emotion detection and record the (set of possible) response emotions. As indicated at a decision block 210, compute a personality model from the sequence of input emotions, output emotions collected in blocks 204, 206, and 208. Operations are done as indicated at a block 212.
Referring to
As indicated at a block 308, for those responses with a tone goal, record the emotional output of the user to the system's expressive output. As indicated at a block 310, identify conversational interactions in which the emotional output of the user failed to match the target. At block 312, modify the personality model with the goal of minimizing the number of interactions identified in block 310.
In accordance with features of the invention, to initialize a personality model, the textual interactions configured in a chatbot will be used: For example:
System: “Hi, what can I do for you today?” Emotion analysis identifies: Agreeable User: “I'm so happy that I can pay my bill by talking to you!” Emotion analysis identifies: Happy
System: “I'm afraid our bill pay capability is down at the moment.” Emotion analysis identifies: Apologetic
User: “That's terrible, I'll never use this system again. Good day, sir!” Emotion analysis identifies Angry
In accordance with features of the invention, these interactions can be treated as an emotion sequence, such as the following:
(greeting) Agreeable->(bill pay) Happy->(not-possible) Apologetic->(avoid-payment) Angry-> . . . .
Standard sequence learning methods such as Hidden Markov Models can be used to construct the personality model from these sequences. Such a model might be represented as a simply input output mapping:
User Intent/System intent Recognized Emotion Emotional Response:
Pay_bill/Can't facilitate Frustrated Apologetic
In accordance with features of the invention, the personality model is sufficient for orchestrating the emotional response of the user so that it is consistent. To further train the system, some or all of the interactions can be annotated with a desired emotional response. This may be either done for each interaction or, more reasonably, for the final interaction. To specify that the goal of the interaction is that user is happy, for example:
System: “Hi, what can I do for you today?” Emotion to display marked as: Agreeable
User: “I'm so happy that I can pay my bill by talking to you!” Emotion analysis identifies: Happy
System: “I'm glad to tell you our bill pay capability is down at the moment.” Emotion to display marked as: Agreeable
User: “That's terrible, I'll never use this insolent system again. Good day, sir!” Emotion analysis identifies: Angry Goal: Happy
In accordance with features of the invention, identifying that the interaction has failed to achieve the goal emotion state for the user, the training method is applied, and the personality model is altered. One way in which the personality model might be altered is to collect cases in which the output is producing an undesirable response, identify the most common case and introducing a variation of the model that would lead to a different response (here, for example, modifying the model so that the emotion to display in the case of a user input emotion Agreeable with user Intent (I_u) Bill_Pay_Request and system intent (I_s) System_Down is not Agreeable but rather Apologetic.
Referring now to
Computer readable program instructions 404, 406, 408, and 410 described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The computer program product 400 may include cloud-based software residing as a cloud application, commonly referred to by the acronym (SaaS) Software as a Service. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer readable program instructions 404, 406, 408, and 410 from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing/processing device.
A sequence of program instructions or a logical assembly of one or more interrelated modules defined by the recorded program means 404, 406, 408, and 410, direct the system 100 for implementing misleading title identification of the preferred embodiment.
While the present invention has been described with reference to the details of the embodiments of the invention shown in the drawing, these details are not intended to limit the scope of the invention as claimed in the appended claims.
