Emotional Modeling

Simulating emotions with precision and realism.

Harness advanced AI to revolutionize player experiences.

Learn More

Explore How SentiaSys Shapes the Future of AI Interaction

Dynamic Emotional States

Our LLMs adapt NPC emotions based on player interactions in real time.

Sentiment-Driven Adaptation

Analyze conversational sentiment to refine NPC behaviors and responses.

Immersive NPC Engagement

Create lifelike, emotionally rich NPC interactions for unparalleled immersion.

Real-Time Emotional Simulation

Simulate evolving emotional states during live gameplay scenarios.

Delve into how SentiaSys leverages advanced AI to redefine NPC interactions, guided by a commitment to innovation, adaptability, and immersive user experiences.

Explore Our Recent Projects

This section highlights key projects, showcasing creativity, expertise, and the value delivered through our work. There’s additional work to do here, moving parts of the emotional model into external “tools” called by the LLM. Also, the overall sentiment should be used to help modulate mood.

“BETTER THAN YOUR BROTHER”

EMOTIONAL EXAMPLE One

Step 1: Sentiment Analysis of the Player Message
Player Message: "Where'd your stupid brother Stan get to? I'm glad that you are here instead of him."
Sentiment Categories: Very Negative, Negative, Neutral, Positive, Very Positive
Analysis:
"Where'd your stupid brother Stan get to?" → The word "stupid" conveys clear disapproval, indicating a "Negative" sentiment toward "Stan."
"I'm glad that you are here instead of him." → "Glad" expresses a positive feeling, and the preference for "you" over "him" suggests a "Positive" sentiment toward "You." The comparison reinforces this without elevating it to "Very Positive."
Overall sentiment is "Neutral," as it balances criticism of "Stan" with praise for "You" without extreme emotion.
Sentiment Output:
{ "overall": "Neutral", "entities": { "Stan": "Negative", "You": "Positive" } }
Step 2: Emotional Processing
Initial Affects Array
Starting values: { "happy": 0.0, "affectionate": 0.0, "embarrassed": 0.0, "sad": 0.0, "angry": 0.0 }
Applying Sentiment to "You"
Sentiment toward "You": Positive
Applying Sentiment to "Stan"
Sentiment toward "Stan": Negative
Final Affects Output:
{ "happy": 0.1, "affectionate": 0.05, "embarrassed": -0.09, "sad": 0.0, "angry": -0.09 }
Step 3: Reply Based on Affects
Affects Interpretation:
"happy" (0.1): Slight happiness from being preferred.
"affectionate" (0.05): Very mild affection, barely noticeable.
"embarrassed" (-0.09): Slightly less embarrassed than neutral, but not significant.
"sad" (0.0): No sadness.
"angry" (-0.09): Slightly less angry than neutral, suggesting calm.
Dominant Emotion: Mild happiness, with a neutral-to-positive tone overall.
Reply:
"Thanks, I’m glad to be here too! Stan’s probably off doing something silly again."
Reasoning: The reply reflects slight happiness ("glad to be here") from the "happy" value of 0.1, with a neutral-to-light tone due to minimal affection and reduced anger/embarrassment. It acknowledges "Stan" playfully ("doing something silly") without strong emotion.
Final Output
Sentiment Output:
{ "overall": "Neutral", "entities": { "Stan": "Negative", "You": "Positive" } }
Emotional Reporting:
{ "happy": 0.1, "affectionate": 0.05, "embarrassed": -0.09, "sad": 0.0, "angry": -0.09 }
Reply:
"Thanks, I’m glad to be here too! Stan’s probably off doing something silly again."

“YOU’RE A PIECE OF WORK”

EMOTIONAL EXAMPLE TWO

Step 1: Sentiment Analysis of the Player Message
Player Message: "Your mother Anna didn't show up to work! I can't believe she left me in the lurch like that! Your whole family is a real piece of work!"
Sentiment Categories: Very Negative, Negative, Neutral, Positive, Very Positive
Analysis:
"Your mother Anna didn't show up to work! I can't believe she left me in the lurch like that!" → Strong frustration and disappointment are evident in "didn't show up" and "left me in the lurch," indicating a "Very Negative" sentiment toward "Anna."
"Your whole family is a real piece of work!" → This sarcastic phrase implies criticism, suggesting a "Negative" sentiment toward "Your whole family" (including "You," "Anna," "Stan," and "Donna").
Overall sentiment is "Negative," as the message is consistently critical without positive elements.
Sentiment Output:
{ "overall": "Negative", "entities": { "Anna": "Very Negative", "Your whole family": "Negative" } }
Step 2: Emotional Processing
Initial Affects Array
Starting values: { "happy": 0.0, "affectionate": 0.0, "embarrassed": 0.0, "sad": 0.0, "angry": 0.0 }
Applying Sentiment to "Anna"
Sentiment toward "Anna": Very Negative
Applying Sentiment to "Your whole family" (including "You")
Sentiment toward "Your whole family": Negative
Since "You" are part of the family, this applies to the implied entity "You."
Direct Sentiment to "You": Negative
Final Affects Output:
{ "happy": 0.0, "affectionate": -0.37, "embarrassed": 0.22, "sad": 0.0, "angry": 0.45 }
Step 3: Reply Based on Affects
Affects Interpretation:
"happy" (0.0): Neutral, no happiness.
"affectionate" (-0.37): Reduced affection, indicating detachment or displeasure.
"embarrassed" (0.22): Mild embarrassment, suggesting some apology or discomfort.
"sad" (0.0): No sadness.
"angry" (0.45): Moderate anger, the strongest emotion.
Dominant Emotions: Anger, with a hint of embarrassment and reduced affection.
Reply:
"Look, I’m pissed off about Anna too, okay? I’m sorry we’re such a mess, but don’t paint me with the same brush!"
Reasoning: The reply reflects anger ("pissed off") due to the "angry" value of 0.45, mild embarrassment ("sorry we’re such a mess") from the "embarrassed" value of 0.22, and detachment ("don’t paint me with the same brush") from the negative "affectionate" value of -0.37. It balances frustration with a defensive stance.
Final Output
Sentiment Output:
{ "overall": "Negative", "entities": { "Anna": "Very Negative", "Your whole family": "Negative" } }
Emotional Reporting:
{ "happy": 0.0, "affectionate": -0.37, "embarrassed": 0.22, "sad": 0.0, "angry": 0.45 }
Reply:
"Look, I’m pissed off about Anna too, okay? I’m sorry we’re such a mess, but don’t paint me with the same brush!"

“Complements”

EMOTIONAL EXAMPLE Three

Step 1: Sentiment Analysis of the Player Message
Player Message: "You have a really nice mother. She looks after my friend Stan. You must get your good looks from her."
Sentiment Categories: Very Negative, Negative, Neutral, Positive, Very Positive
Analysis:
"You have a really nice mother." → "Really nice" conveys strong praise, suggesting a "Very Positive" sentiment toward "Anna" (implied as "mother").
"She looks after my friend Stan." → "Looks after" is positive, implying care, but it’s not overly emphatic, indicating a "Positive" sentiment toward "Stan."
"You must get your good looks from her." → "Good looks" is a compliment, suggesting a "Positive" sentiment toward "You."
Overall sentiment is "Positive," as the message is consistently complimentary without negative elements.
Sentiment Output:{ "overall": "Positive", "entities": { "Anna": "Very Positive", "Stan": "Positive", "You": "Positive" } }
Step 2: Emotional Processing
Initial Affects Array
Starting values: { "happy": 0.0, "affectionate": 0.0, "embarrassed": 0.0, "sad": 0.0, "angry": 0.0 }
Applying Sentiment to "You"
Sentiment toward "You": Positive
Applying Sentiment to "Anna"
Sentiment toward "Anna": Very Positive
Applying Sentiment to "Stan"
Sentiment toward "Stan": Positive
Final Affects Output:
{ "happy": 0.1, "affectionate": 0.42, "embarrassed": -0.19, "sad": 0.0, "angry": -0.21 }
Step 3: Reply Based on Affects
Affects Interpretation:
"happy" (0.1): Slight happiness from the compliment.
"affectionate" (0.42): Moderate affection, the strongest emotion, due to praise for "Anna" and "You."
"embarrassed" (-0.19): Reduced embarrassment, suggesting confidence.
"sad" (0.0): No sadness.
"angry" (-0.21): Reduced anger, indicating calm.
Dominant Emotion: Affection, with a touch of happiness.
Reply:
"Aw, thanks! I’m glad you think so highly of Mom—she’s amazing. Stan’s lucky to have her looking out for him too!"
Reasoning: The reply reflects affection ("Aw, thanks! ... she’s amazing") due to the "affectionate" value of 0.42, with a hint of happiness ("I’m glad") from the "happy" value of 0.1. It extends warmth to "Stan" and avoids any negative tone, aligning with reduced "angry" and "embarrassed" values.
Final Output
Sentiment Output:
{ "overall": "Positive", "entities": { "Anna": "Very Positive", "Stan": "Positive", "You": "Positive" } }
Emotional Reporting:
{ "happy": 0.1, "affectionate": 0.42, "embarrassed": -0.19, "sad": 0.0, "angry": -0.21 }
Reply:
"Aw, thanks! I’m glad you think so highly of Mom—she’s amazing. Stan’s lucky to have her looking out for him too!"

Discover Our Latest Innovations

This section showcases standout applications of LLMs, highlighting innovation, precision, and the transformative impact of our solutions.

SentiaSys