FlowMaster Logo
FlowMasterLet's Innovate
Deterministic | Computational Geometry | Research-Backed

GEOHARM: Hate Speech Detection System

Deterministic Multimodal Geometric Framework for Hate Speech Detection.

GEOHARM is a non-ML, non-DL deterministic framework based on Convex Hull and Computational Geometry. It accepts text, image, and audio inputs, transforms each into deterministic numerical/geometric descriptors, and uses robust convex regions to classify hate speech. Built for resource-constrained environments, it provides explainable geometric decisions with low computational cost.

GEOHARM: Hate Speech Detection System

87.6%

Detection Accuracy

Text + Image + Audio

Modalities

< 50 ms

Inference Time

Yes

CPU Only

System Profile

Multimodal Geometric Framework Architecture

From raw input to geometric decision — a complete deterministic pipeline based on Convex Hull and Computational Geometry.

System Profile 1
1

Deterministic Feature Extraction

Lexical, structural, visual, and acoustic features without ML/DL

2

Geometric Space Construction

Low-dimensional feature spaces for text, image, and audio

3

Convex Hull & Decision Engine

Hate Hull, Safe Hull, category-specific hulls, and geometric fusion

Core Technology

Deterministic Geometric Framework

Deterministic Text Features

Lexical (offensive density, target-group density), structural (punctuation, capitalization), and statistical (entropy, diversity) features.

Deterministic Image Features

Color entropy, edge density, texture statistics, OCR word count, and text density from embedded text.

Deterministic Audio Features

RMS energy, pitch statistics, spectral centroid, spectral flux, speech activity ratio, and MFCC statistics.

Convex Hull Classification

Hate Hull, Safe Hull, and category-specific hulls (racial, religious, gender, political, threatening, dehumanizing).

Multimodal Geometric Fusion

Geometric fusion of text, image, and audio evidence with cross-modal agreement analysis.

Explainable AI (XAI)

Visualizes geometric boundaries, input points, distances, modality contributions, and final geometric decision.

Technical Deep-Dive

Core Technologies

A complete geometric framework using Convex Hull for deterministic hate speech detection across multiple modalities.

Key Features

  • Convex Hull (SciPy spatial.ConvexHull / QuickHull)
  • Deterministic feature engineering without ML/DL
  • Category-specific geometric regions
  • Outlier-resistant hull strategies

Specifications

Hull Types

Hate Hull, Safe Hull, Category Hulls

Modalities

Text + Image + Audio

Geometric Intelligence

Geometric Intelligence Engine

A complete deterministic pipeline from raw input to geometric decision using Convex Hull and Computational Geometry.

📝

Deterministic Text Feature Engine

Extracts lexical, structural, and statistical features from text without ML/DL. Features: offensive density, target-group density, entropy, lexical diversity, punctuation density.

LexicalStructuralStatistical
👁️

Deterministic Visual Feature Engine

Extracts color, edge, texture, and OCR features from images without ML/DL. Features: color entropy, edge density, texture statistics, OCR word count, text density.

ColorEdgeOCRTexture
🎤

Deterministic Acoustic Feature Engine

Extracts RMS energy, pitch, spectral, and speech features from audio without ML/DL. Features: RMS energy, pitch statistics, spectral centroid, speech activity ratio.

AcousticSpectralSpeech
🔷

Convex Hull Generator

Constructs Hate Hull, Safe Hull, and category-specific hulls (racial, religious, gender, political, threatening, dehumanizing) from training data.

Hate HullSafe HullCategory Hulls
🔗

Multimodal Geometric Fusion

Fuses text, image, and audio evidence geometrically. Calculates modality agreement and identifies which modality drives the decision.

Geometric FusionModality AgreementEvidence
🧠

Explainability & Visualization

Visualizes geometric boundaries, input points, distances, modality contributions, and final decision with interactive plots.

VisualizationExplainabilityGeometric Plots
Data Sheet

Technical Specifications

Lexical Featuresword count, offensive density, target-group density, repeated-word ratio
Structural Featurespunctuation density, capitalization ratio, URL/mention/hashtag density
Statistical Featuresentropy, lexical diversity, n-gram frequency
Sentiment Proxiesnegative-word ratio, aggressive-word ratio, threatening-word ratio
Market Opportunity

Applications & Target Audience

Social Media Content Moderation
Community Guidelines Enforcement
Hate Speech Research & Analysis
Educational & Academic Research
Low-Resource Edge Deployment
Government & Policy Monitoring
Why Choose Us

The GEOHARM Advantage

Aspect
ML/DL Approach
Geometric Framework
Detection Method
ML/DL (Probabilistic, Black-Box)
Computational Geometry (Deterministic, Transparent)
Computational Cost
High (GPU, Large Models)
Low (CPU only, Lightweight)
Explainability
Low (Attention/Activation Maps)
High (Geometric Boundaries, Distances)
Modality Fusion
High-Dimensional Concatenation
Geometric Fusion with Modality Agreement
Edge Deployment
Limited (GPU Required)
Excellent (CPU, Minimal Dependencies)
Inference Time
100-500 ms
< 50 ms
Innovation Showcase

Novel Technology & Features

Novel Framework

Deterministic Multimodal Geometry

First computational geometry framework for multimodal hate speech detection. Non-ML, non-DL, fully deterministic.

Explainable AI

Geometric Explainability

Visualizes hull boundaries, input points, and distances. Explains why a sample is classified as hate/non-hate geometrically.

Efficient

Low Computational Cost

Designed for CPU-only inference with < 50 ms latency. No GPU, no large models, minimal dependencies.

Product Imagery

Visual Reference

GEOHARM Geometric Framework Dashboard

GEOHARM Geometric Framework Dashboard

Convex Hull Visualization & Decision

Convex Hull Visualization & Decision

Multimodal Geometric Fusion Analysis

Multimodal Geometric Fusion Analysis

Ready to deploy GEOHARM for deterministic hate speech detection?

We respond within 24 hours