# US 6,185,534 B2 · Modeling emotion and personality in a computer user interface

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 6,185,534 B2
- Document ID: US6185534B2
- Status: Granted
- Kind: B2
- Application number: 09/047,160
- Filed: 1998-03-23
- Granted: 2001-02-06
- Earliest claimed: 1998-03-23
- Assignees: Microsoft Corporation
- Inventors: John S. Breese; John Eugene Ball
- Web page: https://patentgrep.com/patent/US6185534B2

## CPC

- H04N 21/466: Learning process for intelligent management, e.g. learning user preferences for recommending movies {(services using the results of monitoring in broadcast systems H04H60/61)}
- G10L 17/26: Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
- H04N 21/4663: {involving probabilistic networks, e.g. Bayesian networks}
- H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

The invention is embodied in a computer user interface including an observer capable of observing user behavior, an agent capable of conveying emotion and personality by exhibiting corresponding behavior to a user, and a network linking user behavior observed by said observer and emotion and personality conveyed by said agent. The network can include an observing network facilitating inferencing user emotional and personality states from the behavior observed by the observer as well as an agent network facilitating inferencing of agent behavior from emotion and personality states to be conveyed by the agent. In addition, a policy module can dictate to the agent network desired emotion and personality states to be conveyed by the agent based upon user emotion and personality states inferred by the observing network. Typically, each network is a stochastic model. Each stochastic model is preferably a Bayesian network, so that the observing network is a first Bayesian network while the agent network is a second Bayesian network. Generally, the first and second Bayesian networks are similar copies of one another. Each of the two Bayesian networks include a first layer of multi-state nodes representing respective emotional and personality variables, and a second layer of multi-state nodes representing respective behavioral variables. Each one of the nodes includes probabilities linking each state in the one node with states of others of the nodes. More specifically, each one of the nodes in the first layer includes probabilities linking the states of the one first layer node to the states of nodes in the second layer. Similarly, each one of the nodes in the second layer include probabilities linking the states of the one second layer node to states of nodes in the first layer.

## Claims

### Claim 1

A computer user interface comprising:
an observer capable of observing user behavior;
an agent capable of conveying emotion and personality by exhibiting corresponding behavior to a user;
a Bayesian network linking user behavior observed by said observer with emotion and personality conveyed by said agent, comprising:
an observing Bayesian network capable of facilitating inferencing user emotion and personality states from said behavior observed by said observer;
an agent network capable of facilitating inferencing of agent behavior from emotion and personality states to be conveyed by said agent.

### Claim 2 (depends on 1)

The user interface of claim 1 further comprising a policy module dictating, to said agent network, said emotion and personality states to be conveyed by said agent based upon user emotion and personality states inferred by said observing network.

### Claim 3 (depends on 1)

The user interface of claim 1 wherein said network comprises a stochastic model.

### Claim 4 (depends on 1)

The user interface of claim 1 wherein said first and second Bayesian networks are generally similar copies of one another.

### Claim 5 (depends on 1)

The user interface of claim 1 wherein said first Bayesian network comprises:
a first layer of multi-state nodes representing respective emotional and personality variables; and
a second layer of multi-state nodes representing respective behavioral variables.

### Claim 6 (depends on 5)

The user interface of claim 5 wherein:
each one of said nodes in said first layer comprise probabilities linking the states of said one first layer node to the states of nodes in said second layer; and
each one of said nodes in said second layer comprise probabilities linking the states of said one second layer node to states of nodes in said first layer.

### Claim 7 (depends on 5)

The user interface of claim 5 wherein said multi-state nodes representing emotional and personality variables comprise at least one of: (a) a valence node, (b) an arousal node, (c) a friendliness node and (d) a dominance node.

### Claim 8 (depends on 7)

The user interface of claim 7 wherein said multi-state nodes representing behavior variables comprise at least one of: (a) a speech attribute node, (b) a facial expression node, (c) a word attribute node.

### Claim 9 (depends on 5)

The user interface of claim 5 wherein said multi-state nodes representing behavior variables comprise a set of word attribute nodes having probabilities relating states of said emotional and personality variables to a set of corresponding word attributes, said interface further comprising:
a language submodel coupled to said word attribute nodes; and
a word node coupled to said language submodel.

### Claim 10 (depends on 9)

The user interface of claim 9 wherein said set of word attributes comprise at least plural ones of: (a) terseness, (b) positiveness, (c) activeness, (d) strength and (e) formality.

### Claim 11 (depends on 9)

The user interface of claim 9 wherein said language submodel comprises:
a word attribute layer relating individual word expressions to probabilities of conveying particular ones of said set of word attributes;
a scoring layer relating states of said word attribute layer for a particular word expression to states of corresponding ones of said set of word attribute nodes of said network.

### Claim 12 (depends on 10)

The user interface of claim 10 wherein said language submodel further comprises a match module selecting a word expression having a winning score computed by said scoring layer.

### Claim 13 (depends on 3)

The user interface of claim 3 wherein said second Bayesian network comprises:
a first layer of multi-state nodes representing respective emotional and personality variables; and
a second layer of multi-state nodes representing respective behavioral variables, each one of said multistate nodes in one of said layers comprising probabilities linking states of said one node with states of multistate nodes in the other layer.

### Claim 14 (depends on 9)

The user interface of claim 9 wherein said multi-state nodes representing emotional and personality variables comprise at least one of: (a) a valence node, (b) an arousal node, (c) a friendliness node and (d) a dominance node.

### Claim 15 (depends on 14)

The user interface of claim 14 wherein said multi-state nodes representing behavior variables comprise at least one of: (a) a speech attribute node, (b) a facial expression node, (c) a word attribute node.

### Claim 16 (depends on 13)

The user interface of claim 13 wherein said multi-state nodes representing behavioral variables comprise a set of word attribute nodes having probabilities relating states of said emotional and personality variables to a set of corresponding word attributes, said interface further comprising:
a language submodel coupled to said word attribute nodes; and
a word node coupled to said language submodel.

### Claim 17 (depends on 16)

The user interface of claim 16 wherein said set of word attributes comprise at least plural ones of: (a) terseness, (b) positiveness, (c) activeness, (d) strength and (e) formality.

### Claim 18 (depends on 16)

The user interface of claim 16 wherein said language submodel comprises:
a word attribute layer relating individual word expressions to probabilities of conveying particular ones of said set of word attributes;
a scoring layer relating states of said word attribute layer for a particular word expression to states of corresponding ones of said set of word attribute nodes of said network.

### Claim 19 (depends on 18)

The user interface of claim 18 wherein said language submodel further comprises a match module selecting a word expression having a winning score computed by said scoring layer.

### Claim 20 (depends on 19)

The user interface of claim 19 wherein said winning score comprises a highest sum of matches between states of said word attribute layer of said language submodel and states of said word attribute nodes of said network.

### Claim 21 (depends on 15)

The user interface of claim 15 wherein said speech attribute node represents one of speech speed, speech pitch, speech volume, reponse promptness.

### Claim 22

A method of operating a computer user interface, comprising:
providing a user model stochastic network;
providing an agent model stochastic network;
inferring, in said user model stochastic network, user emotion and personality states from observed user behavioral states;
dictating, in a policy module, agent emotional and personality states to be conveyed to a user based upon the user emotion and personality states inferred by said stochastic network;
inferring, in said agent model stochastic network, agent behavioral states to be conveyed to the user based upon the agent emotional and personality states dictated by said policy module.

### Claim 23 (depends on 22)

Computer-readable media storing instructions for carrying out the steps of claim 22.

### Claim 24 (depends on 22)

The method of claim 22 wherein the step of providing a user model stochastic network comprises:
providing a first layer of multi-state nodes representing respective emotional and personality variables; and
providing a second layer of multi-state nodes representing respective behavioral variables.

### Claim 25 (depends on 24)

The method of claim 24 wherein:
the step of providing said first layer of nodes comprises providing probabilities linking the states of said one first layer node to the states of nodes in said second layer; and
the step of providing said second layer of nodes comprises providing probabilities linking the states of said one second layer node to states of nodes in said first layer.

### Claim 26 (depends on 24)

The method of claim 24 wherein the step of providing multi-state nodes representing emotional and personality variables comprise providing at least one of: (a) a valence node, (b) an arousal node, (c) a friendliness node and (d) a dominance node.

### Claim 27 (depends on 26)

The method of claim 26 wherein the step of providing multi-state nodes representing behavior variables comprises providing at least one of: (a) a speech attribute node, (b) a facial expression node, (c) a word attribute node.

### Claim 28 (depends on 24)

The method of claim 24 wherein the step of providing multi-state nodes representing behavioral variables comprises providing a set of word attribute nodes having probabilities relating states of said emotional and personality variables to a set of corresponding word attributes, said method further comprising:
providing a language submodel coupled to said word attribute nodes; and
providing a word node coupled to said language submodel.

### Claim 29 (depends on 28)

The method of claim 28 wherein the step of providing a set of word attributes comprises providing at least plural ones of: (a) terseness, (b) positiveness, (c) activeness, (d) strength and (e) formality.

### Claim 30 (depends on 28)

The method of claim 28 wherein the step of providing said language submodel comprises:
providing a word attribute layer relating individual word expressions to probabilities of conveying particular ones of said set of word attributes; and
providing a scoring layer relating states of said word attribute layer for a particular word expression to states of corresponding ones of said set of word attribute nodes of said network.

### Claim 31 (depends on 30)

The method of claim 30 wherein the step of providing said language submodel further comprises providing a match module capable of selecting a word expression having a winning score computed by said scoring layer.

### Claim 32 (depends on 24)

The method of claim 24 wherein the step of providing said agent model stochastic network comprises providing a copy of said user model stochastic network.

### Claim 33 (depends on 22)

The method of claim 22 further comprising:
inferring in said user network a user emotional and personality state from a current observed user behavioral state, and transmitting the inferred user emotional and personality state to said policy module to evoke from said policy module a dictated emotional and personality state to be conveyed to said user;
inferring in said agent network an agent behavioral state from the dictated emotional and personality state evoked from said policy module;
storing the current configuration of said user network based upon the current inferred user emotional and personality state; and
repeating the foregoing inferring steps based upon subsequent observed user behavioral state.

### Claim 34 (depends on 33)

The method of claim 33 wherein the step of inferring a user emotional and personality state is preceded by a step of observing the current user behavioral state.

### Claim 35 (depends on 33)

The method of claim 33 wherein the step of inferring an agent behavioral state is followed by a step of exhibiting in said user interface said agent behavioral state to said user.