Leveraging Pattern Semantics for Extracting Entities in Enterprises.

Area of Science:

• Information Extraction
• Natural Language Processing
• Data Mining

Background:

• Enterprise entity extraction is crucial for applications like search and recommendation, but faces challenges unique to corporate data.
• Existing web-based systems struggle with sparse enterprise data, lack of redundancy, and semantic drift, leading to low recall or high noise.
• Internal enterprise entities often lack public web presence, necessitating methods that utilize sparse internal signals effectively.

Purpose of the Study:

• To propose an end-to-end framework for accurate and complete entity extraction in enterprise domains.
• To address the challenges of sparse data, semantic drift, and limited web presence of internal entities.
• To improve enterprise efficiency through enhanced entity identification and collection.

Main Methods:

• Developed an end-to-end framework accepting an enterprise corpus and limited seeds for entity extraction.
• Introduced the Semantic Pattern Graph (SPG) to leverage public signals for understanding lexical pattern semantics.
• Reinforced pattern evaluation using mined semantics from the SPG to improve entity extraction accuracy and completeness.

Main Results:

• The proposed framework effectively extracts entities from enterprise data, outperforming existing methods in accuracy and recall.
• Experiments on Microsoft enterprise data demonstrated the framework's capability to handle sparse data and semantic drift.
• The Semantic Pattern Graph significantly enhanced the understanding and utilization of lexical patterns for entity discovery.

Conclusions:

• The novel framework and Semantic Pattern Graph provide a robust solution for enterprise entity extraction challenges.
• This approach leads to a high-quality entity collection, enhancing enterprise efficiency and data utilization.
• The method is effective in discovering sparse and semantically drifted entities within enterprise domains.