Sentiment and rules-based equity analysis using customized neural networks in multi-layer, machine learning-based model

A system, comprising:

• a data collection element configured to receive input data comprised of structured data sources and unstructured data sources;
• a multi-layer machine learning-based model configured to analyze a state of a selected equity within a specified future time from the input data, the multi-layer machine learning-based model including:
• a sentiment discovery engine, configured to develop a taxonomy comprising equity-specific keywords and keyword pairings for a selected equity, the taxonomy identifying one or more sentiment predictors for the selected equity from the unstructured data sources, identify discrete-time data points constructed from the structured data sources that define temporal parameters relative to the selected equity for the specific future time, and modify the taxonomy with the temporal parameters to create a set of classified content representing a temporally-relative sentiment for the selected equity,
• a rules discovery engine, configured to develop knowledge-based rules representing specific knowledge relative to the selected equity derived from economic and equity-specific indicators in the input data, and
• a neural network modeling layer configured to:
• map the taxonomy and the knowledge-based rules into one or more neural networks each having a topology comprised of a plurality of nodes and connections that are initiated by threshold activation functions representing the set of classified content and the specific knowledge, and tune the one or more neural networks in at least one deep learning meta network, by identifying additional, temporally-dynamic predictors representing patterns in the taxonomy and the knowledge-based rules that quantify relationships between data points within the set of classified content and the specific knowledge, modify the threshold activation functions for the plurality of nodes and connections of the one or more neural networks, and identify and apply adjusted topologies for the one or more neural networks based on the additional, temporally-dynamic predictors,
• wherein the multi-layer machine learning-based model generates a forecast of the state of the selected equity at the specified future time.

A method, comprising:

• classifying input data comprised of structured data sources and unstructured data sources to develop a data set for a multi-layer machine learning-based model configured to analyze a state of a selected equity within a specified future time, by
• identifying one or more sentiment predictors for the selected equity from the unstructured data sources to develop a taxonomy comprising equity-specific keywords and keyword pairings for a selected equity, and modifying the modifying the taxonomy with temporal parameters relative to the selected equity for the specific future time representing discrete-time data points constructed from the structured data sources, to create a set of classified content representing a temporally-relative sentiment for the selected equity, and
• deriving knowledge-based rules representing specific knowledge relative to the selected equity from economic and equity-specific indicators in the input data;
• modeling the set of classified content and the knowledge-based rules within a neural network modeling layer comprised of one or more neural networks and at least one deep learning meta network, the neural network modeling layer configured to
• represent the set of classified content and the specific knowledge as threshold activation functions for initiating a plurality of nodes and connections comprising a topology of the one or more neural networks to map the taxonomy and the knowledge-based rules into the one or more neural networks, and
• tune the one or more neural networks in the at least one deep learning meta network, by identifying additional, temporally-dynamic predictors representing patterns in the taxonomy and the knowledge-based rules that quantify relationships between data points within the set of classified content and the specific knowledge, to modify the threshold activation functions for the plurality of nodes and connections of the one or more neural networks, and identify and apply adjusted topologies for the one or more neural networks based on the additional, temporally-dynamic predictors;
• and
• performing the multi-layer machine learning-based model to forecast the state of the selected equity at the specified future time.