Patent Keyword Extraction Algorithm Based on Distributed Representation for Patent Classification.

Area of Science:

• Computer Science
• Information Science
• Artificial Intelligence

Background:

• Keyword extraction is crucial for text mining tasks like retrieval and summarization.
• Existing methods often rely on discrete bag-of-words representations.
• There is a need for effective keyword extraction in patent analysis.

Purpose of the Study:

• To propose a novel patent keyword extraction algorithm (PKEA).
• To leverage distributed Skip-gram models for keyword extraction.
• To develop quantitative measures for evaluating keyword extraction without human annotations.

Main Methods:

• Developed the Patent Keyword Extraction Algorithm (PKEA) using the Skip-gram model.
• Created quantitative performance measures based on information gain and cross-validation with Support Vector Machine (SVM) classification.
• Evaluated PKEA on a benchmark dataset and a custom dataset of 2500 autonomous car-related patents.

Main Results:

• PKEA demonstrated promising performance in extracting keywords from patent texts.
• The algorithm was compared against Frequency, TF-IDF, TextRank, and RAKE.
• The proposed quantitative measures proved valuable for evaluation in the absence of human-annotated keywords.

Conclusions:

• The proposed PKEA offers an effective approach for keyword extraction in patent classification.
• The Skip-gram model provides a robust foundation for representing patent text for keyword extraction.
• The developed evaluation metrics are suitable for assessing keyword extraction algorithms when human annotations are limited.