Sentiment Analysis Engine

Emotion, quantified

The Modulus Sentiment Analysis Engine uses deep-learning neural networks to pull mood and emotion from millions of social media messages as they happen. Within weeks of going live back in 2009, it began surfacing something remarkable: the predominant emotion moving through social media, in effect a real-time read on the world's mood.

Running uninterrupted for well over a decade, the system has amassed multiple petabytes of historic time-series data. Some of the world's top-performing hedge funds, along with governmental organizations, rely on it for forecasting and analysis, and the same engine serves PR, marketing, and brand teams just as well.

Reading the world's mood

Love, joy, disgust, anger, sadness, fear, surprise: the engine analyzes and records the world's emotions in real time, message by message. It draws on deep-learning neural networks, with help from OpenCyc and WordNet, to capture the true meaning, intention, and emotion behind each post.

It works across ten languages, including English, Mandarin Chinese, Spanish, Arabic, Hindi, Bengali, Portuguese, Russian, Japanese, and German, and stores everything in the Modulus Real-Time Data Server. The emotion definitions draw in part on the work of Professor W. Gerrod Parrott.

• Detects love, joy, surprise, anger, sadness, and fear
• Real-time scoring, second by second
• Ten languages, from English to Mandarin
• Backed by deep-learning neural networks
• Meaning extraction via OpenCyc and WordNet
• Storage on the Modulus Real-Time Data Server

Signal the markets feel

The data has shown striking predictive power. Granger causality testing between the engine's output and the Dow Jones Industrial Average has revealed extremely strong predictive relationships, echoing widely cited research that social mood can anticipate market movement days in advance.

Notably, Modulus developed the system a full year before that research was published. The petabytes of historic and real-time data hold time-series for global mood, love, joy, surprise, anger, sadness, fear, and overall sentiment, which quantitative teams can fold directly into their models for equities, futures, options, and other instruments.

Beyond the trading floor

The same engine reads sentiment across Twitter, Facebook, YouTube, LinkedIn, Reddit, Telegram, and other networks, making it just as valuable outside finance. PR and marketing agencies use it to monitor public sentiment in real time for consumer strategy, brand and crisis management, advertising effectiveness, and product-design feedback.

Modulus also offers consulting to support custom social media research, analytics, and reporting, helping organizations predict and track consumer and market trends as they form.

Historical background: SharpeMind

In December 2013, the Modulus SharpeMind project launched as the first financial application built for IBM Watson, delivering real-time analysis of unstructured financial data to mobile devices. Millions of unstructured documents, including analyst reports, news feeds, governmental filings, and social media, fed the system, which gave traders instant access to trading signals and consensus-based buy and sell recommendations.

Watson was never designed for time-series analysis, so Modulus invented and patented a method that let consensus systems process time-series data, allowing Watson to produce forecasts from pre-processed text alone. Modulus is no longer engaged with IBM Watson, having since developed its own natural language processing systems for financial markets and beyond.

Built on a modern stack

What the engine delivers

A sentiment platform in production since 2009, combining deep learning, real-time ingestion, and a deep historic archive.

Inside the platform

The data sources, emotions, and infrastructure behind the engine.