SimCog has Developed an Investment Strategy by Means of Self-Learning Sentiment Algorithms
A novel prediction technique based on structured and unstructured data is used to operate a fully automated stock price prediction model. SimCog utilizes web crawlers to search for company relevant news in newsfeeds, Twitter, social networks, forums and blogs and analyzes it. Latest SimCog algorithms detect unknown patterns and relations in this data.
Text data (unstructured data) as well as relevant structured data is collected on a daily basis. After screening, the data is split into several subject areas. Furthermore, sentiment analyses are performed. Finally, the machine learning methods for pattern recognition are trained on a daily basis.
The SimCog model is following a successful market neutral trading strategy, which is UCITS compatible.
Fraudulent activity can lead to a substantial loss for companies. Even as anti-fraud measures are put in place, fraud continues to grow worldwide. However, sophisticated data technologies allow installation of an early warning system for accounting and credit card fraud as well as money laundering.
The supervised method for fraud detection requires classified training data sets. Thereby, a “fraud” and “non-fraud” data set is needed. For good applicability a large number of past frauds is beneficial.
In unsupervised methods no prior classification of fraudulent or non-fraudulent activities exists. Therefore, the analysis first involves the detection of outliers in known distributions. Furthermore, behaviour modification is detected when clusters of subgroups suddenly behave differently. Finally, Newcomb Benford’s Law is applied.
Based on the issue at hand, SimCog is using “supervised” or “unsupervised” learning methods.