Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated Neural Patent Landscaping in the Small Data Regime

Tisa Islam Erana, Mark A. Finlayson

TL;DR

The paper tackles automated patent landscaping in the small-data regime by combining active learning with a multi-stream neural architecture to compensate for scarce labeled data. It extends seed/anti-seed data collection with an annotation loop to obtain hard-boundary examples and augments text, CPC-embedding, and citation features, including a BERT-for-Patents backbone. The results show substantial improvements over prior systems, particularly in low-data settings, achieving around $F_1$ values of $0.75$ with only $24$ labeled patents and demonstrating that abstract-claims-CPC fusion and CPC-based citation features are effective. The work provides a data-and-methods blueprint for scalable, domain-adaptive patent landscaping and releases code and dataset to enable reproducibility and further research.

Abstract

Patent landscaping is the process of identifying all patents related to a particular technological area, and is important for assessing various aspects of the intellectual property context. Traditionally, constructing patent landscapes is intensely laborious and expensive, and the rapid expansion of patenting activity in recent decades has driven an increasing need for efficient and effective automated patent landscaping approaches. In particular, it is critical that we be able to construct patent landscapes using a minimal number of labeled examples, as labeling patents for a narrow technology area requires highly specialized (and hence expensive) technical knowledge. We present an automated neural patent landscaping system that demonstrates significantly improved performance on difficult examples (0.69 $F_1$ on 'hard' examples, versus 0.6 for previously reported systems), and also significant improvements with much less training data (overall 0.75 $F_1$ on as few as 24 examples). Furthermore, in evaluating such automated landscaping systems, acquiring good data is challenge; we demonstrate a higher-quality training data generation procedure by merging Abood and Feltenberger's (2018) "seed/anti-seed" approach with active learning to collect difficult labeled examples near the decision boundary. Using this procedure we created a new dataset of labeled AI patents for training and testing. As in prior work we compare our approach with a number of baseline systems, and we release our code and data for others to build upon.

Automated Neural Patent Landscaping in the Small Data Regime

TL;DR

The paper tackles automated patent landscaping in the small-data regime by combining active learning with a multi-stream neural architecture to compensate for scarce labeled data. It extends seed/anti-seed data collection with an annotation loop to obtain hard-boundary examples and augments text, CPC-embedding, and citation features, including a BERT-for-Patents backbone. The results show substantial improvements over prior systems, particularly in low-data settings, achieving around values of with only labeled patents and demonstrating that abstract-claims-CPC fusion and CPC-based citation features are effective. The work provides a data-and-methods blueprint for scalable, domain-adaptive patent landscaping and releases code and dataset to enable reproducibility and further research.

Abstract

Patent landscaping is the process of identifying all patents related to a particular technological area, and is important for assessing various aspects of the intellectual property context. Traditionally, constructing patent landscapes is intensely laborious and expensive, and the rapid expansion of patenting activity in recent decades has driven an increasing need for efficient and effective automated patent landscaping approaches. In particular, it is critical that we be able to construct patent landscapes using a minimal number of labeled examples, as labeling patents for a narrow technology area requires highly specialized (and hence expensive) technical knowledge. We present an automated neural patent landscaping system that demonstrates significantly improved performance on difficult examples (0.69 on 'hard' examples, versus 0.6 for previously reported systems), and also significant improvements with much less training data (overall 0.75 on as few as 24 examples). Furthermore, in evaluating such automated landscaping systems, acquiring good data is challenge; we demonstrate a higher-quality training data generation procedure by merging Abood and Feltenberger's (2018) "seed/anti-seed" approach with active learning to collect difficult labeled examples near the decision boundary. Using this procedure we created a new dataset of labeled AI patents for training and testing. As in prior work we compare our approach with a number of baseline systems, and we release our code and data for others to build upon.
Paper Structure (17 sections, 4 figures, 5 tables)

This paper contains 17 sections, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Patent Classification
  4. Automated Patent Landscaping
  5. Data
  6. Annotation using Active Learning
  7. Dataset
  8. Patent Landscaping Methods
  9. Our Neural Architecture
  10. Textual Information (Streams 1--3)
  11. Citation Networks (Stream 4)
  12. CPC Embeddings (Stream 5)
  13. Models Tested
  14. Results
  15. Discussion
  16. ...and 2 more sections

Figures (4)

  • Figure 1: A section of the CPC code hierarchy showing CPC class, subclass, main group and subgroups
  • Figure 2: Screenshot of the Patentify Annotation Tool
  • Figure 3: Neural Architecture. We experimented with a number of variants. In particular, model variants included or excluded various streams (numbered 1--5) with various settings, as described in Table \ref{['tab:fullmodels']}.
  • Figure 4: Learning curve for 400, 200, 100, 48, and 24 examples on the two best performing models (CPCAvg and B4P-All) compared with Abood2018AutomatedPL's model (A&F) and the best performing baseline (SVM/tfidf).