Azlo Sentiment
AI trænet på millioner af online samtaler for at afkode den *ægte* tone.
Awaiting Input Stream...
Enter text and click analyze to begin
Skal du scanne store datamængder? Hent API Nøglen.
Kerne Arkitektur
- Backbone Model
- MiniLM-L12-v2
- Antal Parametre
- 33 Million (Quantized)
- Skjulte Dimensioner
- 10-Dimensional Vector
- Attention Heads
- Regression Head
Inference Pipeline
- Runtime Engine
- ONNX Runtime (WASM)
- Graf Optimering
- Constant Folding
- Tokenizer
- WordPiece (30k vocab)
- Gns. Responstid
- ~40ms
Finjusteringsstrategi
- Tabsfunktion (Loss)
- Smooth L1 + Cosine Sim
- Prædiktionshoved
- Linear + LayerNorm
- Kontekstvindue
- Mean Pooling Strategy
- Optimizer
- RobustScaler
About the Azlo Sentiment Engine
10-Dimensional Vector Analysis
The Azlo Sentiment Engine goes beyond basic "positive" or "negative" labels. It utilizes a 10-dimensional output vector trained on millions of data points to map text into a complex psychological coordinate system. This allows the model to distinguish between 150+ distinct states, such as the difference between "Loneliness" and "Solitude", or "Joy" and "Catharsis".
Complex State Detection
Most AI models are trained on basic emotions. Azlo is trained on nuances. It can detect specific phenomena like Cognitive Dissonance (conflicting beliefs), Hiraeth (nostalgic longing), Schadenfreude, and Deep Work states. This makes it uniquely suited for analyzing literary text, brand sentiment, and complex social interactions.
Mathematical Precision
By using Cosine Similarity and Smooth L1 Loss during training, the engine minimizes the distance between predicted emotional vectors and the ground truth. The result is a highly accurate, "distance-based" understanding of emotion, where the model calculates how close your text is to a specific feeling, rather than just guessing a category.