Granular neural networks for numerical-linguistic data fusion and knowledge discovery

Area of Science:

• Artificial Intelligence
• Data Mining
• Computational Intelligence

Background:

• Traditional data mining methods struggle with heterogeneous data types (numerical and linguistic).
• Integrating diverse data sources is crucial for comprehensive knowledge discovery.
• Existing neural network models often lack the capacity to process both numerical and linguistic information simultaneously.

Purpose of the Study:

• To develop a novel methodology for knowledge discovery and data mining (KDDM) that handles both numerical and linguistic data.
• To design a granular neural network (GNN) capable of processing, fusing, and discovering knowledge from mixed-type databases.
• To enable the prediction of missing data using discovered granular knowledge.

Main Methods:

• A granular neural network (GNN) was designed, integrating principles of granular computing, neural computing, fuzzy computing, linguistic computing, and pattern recognition.
• The GNN processes granular data, performs numerical-linguistic data fusion, and discovers granular knowledge.
• The methodology addresses challenges in converting fuzzy linguistic data to numerical features and fusing heterogeneous data.

Main Results:

• The GNN successfully processes granular data and makes decisions based on fused granular inputs.
• It learns internal granular relationships within numerical-linguistic datasets and predicts new relations.
• The GNN demonstrates capability in compressing low-level granular data into high-level granular knowledge with a quantifiable compression rate.

Conclusions:

• The proposed GNN offers a robust framework for knowledge discovery and data mining with mixed numerical and linguistic data.
• This approach facilitates effective data fusion and granular knowledge extraction.
• Future work will explore parallel and distributed GNN architectures for handling large-scale KDDM challenges.