SE-MSLC: Semantic Entropy-Driven Keyword Analysis and Multi-Stage Logical Combination Recall for Search Engine

Area of Science:

• Computer Science
• Information Science

Background:

• Information retrieval is crucial for accessing data.
• Existing methods struggle with accuracy in vertical domain search engines.

Purpose of the Study:

• To propose an information retrieval framework (SE-MSLC) that enhances retrieval effectiveness using information theory.
• To improve the quality of search results in intelligent vertical domain search engines.

Main Methods:

• Developed a semantic entropy-driven keyword importance analysis (SE-KIA) to dynamically weight query keywords based on logs, corpus, and semantic entropy.
• Implemented a hybrid recall strategy (HRS-MSLC) combining multi-stage and logical combination strategies for multi-granularity word segmentation and "AND"/"OR" logic.
• Managed retrieval uncertainty by prioritizing keywords with high information content.

Main Results:

• The SE-MSLC framework improved Hit Rate@1 by 7.3% and Hit Rate@3 by 6.6%.
• Effectively addressed and resolved problematic cases within the vertical domain search engine.
• Demonstrated a superior balance between retrieval result quantity and relevance.

Conclusions:

• The proposed SE-MSLC framework significantly enhances information retrieval effectiveness in vertical domains.
• Semantic entropy and advanced recall strategies are key to improving search engine performance and user satisfaction.