SENTRIX
Connecting...
Neural Intelligence // Local Deployment

SENTRIX

Twitter-RoBERTa v2 // On-Device

Aspect Multilingual Emoji-Aware
// Server Address
On mobile: use your PC's IP (shown in terminal when you run python app.py)
0 Total
0 Positive
0 Negative
0 Neutral
Try:
Input Terminal — twitter_roberta_v2
0/512
// Running neural inference...
// Sentiment
// Language
// Emojis
// Confidence Scores
Positive
0%
Negative
0%
Neutral
0%
// Extracted Aspects
// Reasoning Output
Demo Dataset

PHRASES

Tap any phrase to load & analyze

🟢 Positive
🔴 Negative
🔵 Neutral
🌐 Multilingual
😀 Emoji-Aware
🎯 Aspect-Based
⚖️ Mixed / Tricky
System Info

ABOUT

// Model Specifications
Modeltwitter_roberta_v2
ArchitectureRoBERTa-base
TaskSentiment Classification
FormatSafeTensors
BackendFlask + PyTorch
InferenceCPU (local)
// Capabilities
Aspect-Based Sentiment Analysis
Automatically extracts key noun aspects (camera, battery, price) from the input text and highlights them as chips.

Multilingual Support
Detects and labels French 🇫🇷, Spanish 🇪🇸, German 🇩🇪, and Portuguese 🇧🇷 inputs automatically.

Emoji-Aware Processing
Extracts emojis from input and displays them separately, providing an additional sentiment signal beyond text.

Lexical Override
When model confidence is below 65% but strong negative signals are detected (profanity, threats), the system overrides to NEGATIVE with a warning note.
// NLP Pipeline
Preprocessing → URLs replaced with http, @mentions replaced with @user

Tokenization → RoBERTa BPE tokenizer, max 128 tokens

Inference → Softmax over logits → confidence scores

Post-processing → Language detect, emoji extract, aspect extract, lexical override

Dataset → Trained on Twitter/X sentiment data (Kaggle)
// Mobile Usage
Run python app.py on your PC. The terminal shows two IP addresses:

127.0.0.1:5000 → only works on the PC itself
192.168.x.x:5000 → works from any device on the same WiFi

On this mobile page, enter the 192.168.x.x:5000 address in the Server Address box on the Analyze tab and tap Set.