Perplexity

Stability (learning theory)

Machine learning

Data mining

World Wide Web

Language model

Artificial intelligence

Isolation (microbiology)

Index (typography)

Computer science

Selection (genetic algorithm)

Biology

Microbiology

Entropy

This study constructs a comprehensive index to effectively judge the optimal number of topics in the LDA topic model. Based on the requirements for selecting the number of topics, a comprehensive judgment index of perplexity, isolation, stability, and coincidence is constructed to select the number of topics. This method provides four advantages to selecting the optimal number of topics: (1) good predictive ability, (2) high isolation between topics, (3) no duplicate topics, and (4) repeatability. First, we use three general datasets to compare our proposed method with existing methods, and the results show that the optimal topic number selection method has better selection results. Then, we collected the patent policies of various provinces and cities in China (excluding Hong Kong, Macao, and Taiwan) as datasets. By using the optimal topic number selection method proposed in this study, we can classify patent policies well.

Selection of the Optimal Number of Topics for LDA Topic Model—Taking Patent Policy Analysis as an Example