Variational Bayesian Semi-Supervised Keyword Extraction.

Area of Science:

• Natural Language Processing
• Information Retrieval
• Machine Learning

Background:

• The increasing volume of textual data necessitates efficient information extraction methods.
• Existing keyword extraction techniques often lack effectiveness in semi-supervised settings, particularly when using partial keyword lists.

Purpose of the Study:

• To enhance the effectiveness and scalability of semi-supervised keyword extraction.
• To address the under-exploration of leveraging partial keyword lists in keyword extraction.

Main Methods:

• A novel variational Bayesian semi-supervised (VBSS) method is proposed.
• The VBSS method utilizes variational inference and data augmentation to replace computationally expensive posterior sampling.
• This approach enables closed-form updates and significant speedups, particularly for long documents.

Main Results:

• The VBSS method demonstrates improved performance on longer texts compared to state-of-the-art techniques.
• The proposed method offers better control over false discovery rates.
• Numerical results validate the effectiveness and scalability of the VBSS approach.

Conclusions:

• The VBSS method provides an effective and scalable solution for semi-supervised keyword extraction.
• This advancement is particularly beneficial for processing large volumes of text and leveraging partial keyword information.
• The VBSS method offers a practical alternative to existing keyword extraction techniques.