Multi-objective genetic programming strategies for topic-based search with a focus on diversity and global recall

Area of Science:

• Information Retrieval
• Computer Science
• Artificial Intelligence

Background:

• Topic-based search systems aim to retrieve relevant information by understanding user interests.
• Generating diverse, precise, and high-recall queries automatically is a significant challenge.
• Existing Multi-Objective Evolutionary Algorithms (MOEAs) show promise but struggle with result diversity and global recall.

Purpose of the Study:

• To propose novel Multi-Objective Genetic Programming (MOGP) strategies for improved topic-based query generation.
• To address the limitations of MOEAs, specifically loss of diversity and low global recall.
• To develop objective functions that maximize precision and recall while minimizing result similarity.

Main Methods:

• Developed a family of MOGP strategies.
• Defined three novel objective functions utilizing result set similarity and information entropy.
• Conducted extensive experiments to evaluate the proposed strategies.

Main Results:

• Proposed MOGP strategies significantly improve precision within a few generations.
• Some strategies successfully maintain or enhance global recall.
• Comparative analysis shows the MOGP approach outperforms previous MOEAs in precision and global recall.
• MOGP strategies demonstrate superior precision, recall, and F1-score compared to state-of-the-art term-weighting methods.

Conclusions:

• The proposed MOGP strategies offer a superior approach to topic-based query generation.
• The effectiveness of strategies depends on the careful selection of objectives for maximization/minimization based on the specific application.
• This work advances the field of information retrieval by providing more effective tools for topic-based search.