That's Classified! Inventing a New Patent Taxonomy

Journal: Industrial and Corporate Change

Date: 2021

Author: Stephen D. Billington, Alan J. Hanna

Abstract:
Innovation researchers currently make use of various patent classification schemas, which are hard to replicate. Using machine learning techniques, we construct a transparent, replicable and adaptable patent taxonomy, and a new automated methodology for classifying patents. We contrast our new schema with existing ones using a long-run historical patent dataset. We find quantitative analyses of patent characteristics are sensitive to the choice of classification; our interpretation of regression coefficients is schema dependent. We suggest much of the innovation literature should be carefully interpreted in light of our findings.

Link: Google Scholar

Background and Context

Research Problem

Innovation researchers currently use various inconsistent patent classification systems which are difficult to replicate and compare across studies.

Study Approach

The researchers develop a new transparent and replicable patent taxonomy using machine learning techniques to consistently classify patents based on their industry of final use.

Data Analyzed

The study examines British patents granted between 1700-1850 to test how different classification systems affect the interpretation of patent characteristics.

Frequency of Patent Classes Across Historical Taxonomies

Shows which patent classes appear most consistently across historical classification systems
Machinery and Textiles are the most common classes, appearing in 83% of taxonomies
Demonstrates the core technology categories that persist across different classification approaches

Evolution of Patent Classification Methods Over Time

Traces how patent classification systems have evolved from very detailed to more consolidated approaches
Shows reduction from 246 classes in Woodcroft's 1860 system to 20 classes in the new taxonomy
Reflects the trend toward more manageable and comparable classification systems

Impact of Classification Choice on Patent Value Assessment

Demonstrates how different classification systems lead to varying interpretations of patent value
Shows significant differences in citation impact coefficients between classification methods
Highlights the importance of consistent classification for accurate patent analysis

Classification Divergence in Capital-Saving Patents

Illustrates how classification choice affects the identification of capital-saving innovations
Shows contradictory results between classification systems for the same patent categories
Emphasizes the need for standardized classification methods in historical patent analysis

Topic Analysis Success Rates Across Countries

Shows the effectiveness of the new classification system across different countries
Demonstrates the international applicability of the machine learning approach
Validates the robustness of the new taxonomy across different patent systems

Contribution and Implications

Provides the first standardized, replicable patent classification system that can be consistently applied across different time periods and countries
Demonstrates that choice of classification system significantly impacts research findings and interpretations of historical innovation patterns
Offers a new machine learning methodology that reduces subjectivity in patent classification and improves comparability across studies

Data Sources

Class Frequency Chart: Based on Table 3 showing frequency of patent classes across historical taxonomies
Evolution Chart: Constructed from historical classification systems discussed in Tables 1 and 2
Citation Impact Chart: Derived from Table 8 showing negative binomial regression results
Capital-Saving Chart: Based on Table 9 showing probit regression results
Topic Analysis Chart: Constructed from Table 5 showing topic analysis results across countries