An evaluation of standard retrieval algorithms and a binary neural approach

Area of Science:

• Computer Science
• Information Retrieval
• Machine Learning

Background:

• Traditional data structures like inverted file lists and hash tables are widely used in information retrieval.
• Evaluating the efficiency and capabilities of various data retrieval algorithms is crucial for optimizing information access.

Purpose of the Study:

• To evaluate and compare standard data structures with a novel binary neural network for information retrieval.
• To assess algorithms based on storage efficiency, retrieval speed, and partial matching capabilities.

Main Methods:

• Utilized a large Information Retrieval dataset for evaluation.
• Implemented and analyzed standard data structures (inverted file lists, hash tables).
• Developed and tested a novel binary neural network featuring single-epoch training, superimposed coding, and associative matching.

Main Results:

• The novel binary neural network demonstrated superior performance in training speed.
• The neural network approach significantly outperformed traditional methods in partial match retrieval time.
• Identified specific strengths and weaknesses of each evaluated data retrieval approach.

Conclusions:

• The novel binary neural network approach is highly effective for information retrieval, particularly for tasks requiring fast partial matching.
• Recommendations are provided for the optimal application of this advanced neural network technique in information retrieval systems.