Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From PARIS to LE-PARIS: Toward Patent Response Automation with Recommender Systems and Collaborative Large Language Models

Jung-Mei Chu, Hao-Cheng Lo, Jieh Hsiang, Chun-Chieh Cho

TL;DR

This paper tackles the challenge of automating Office Action responses in patent prosecution by introducing two integrated systems, PARIS and its LLM-enhanced successor LE-PARIS, which blend topic-aware templates, hybrid recommender systems, and LLM-based generation to support patent attorneys. Through a multi-paradigm evaluation spanning $2018$ to $2023$, including topic modeling, Delphi-based consensus, recommender experiments, generation quality assessments, and user studies with hundreds of attorneys, the authors demonstrate substantial improvements in template relevance, recommendation accuracy, and generated response quality. LE-PARIS, in particular, shows stronger semantic representations and higher practical impact, driven by LLM embeddings and a generative component that integrates external resources with attorney input. The work delivers a scalable OA response workflow, a large OA Topics Database of $132$ topics, $15{,}571$ templates, and a cascade hybrid recommender, establishing a foundation for practical, confidential patent prosecution automation and guiding future research on domain-specific LLM deployment and human-AI collaboration.

Abstract

In patent prosecution, timely and effective responses to Office Actions (OAs) are crucial for securing patents. However, past automation and artificial intelligence research have largely overlooked this aspect. To bridge this gap, our study introduces the Patent Office Action Response Intelligence System (PARIS) and its advanced version, the Large Language Model (LLM) Enhanced PARIS (LE-PARIS). These systems are designed to enhance the efficiency of patent attorneys in handling OA responses through collaboration with AI. The systems' key features include the construction of an OA Topics Database, development of Response Templates, and implementation of Recommender Systems and LLM-based Response Generation. To validate the effectiveness of the systems, we have employed a multi-paradigm analysis using the USPTO Office Action database and longitudinal data based on attorney interactions with our systems over six years. Through five studies, we have examined the constructiveness of OA topics (studies 1 and 2) using topic modeling and our proposed Delphi process, the efficacy of our proposed hybrid LLM-based recommender system tailored for OA responses (study 3), the quality of generated responses (study 4), and the systems' practical value in real-world scenarios through user studies (study 5). The results indicate that both PARIS and LE-PARIS significantly achieve key metrics and have a positive impact on attorney performance.

From PARIS to LE-PARIS: Toward Patent Response Automation with Recommender Systems and Collaborative Large Language Models

TL;DR

This paper tackles the challenge of automating Office Action responses in patent prosecution by introducing two integrated systems, PARIS and its LLM-enhanced successor LE-PARIS, which blend topic-aware templates, hybrid recommender systems, and LLM-based generation to support patent attorneys. Through a multi-paradigm evaluation spanning to , including topic modeling, Delphi-based consensus, recommender experiments, generation quality assessments, and user studies with hundreds of attorneys, the authors demonstrate substantial improvements in template relevance, recommendation accuracy, and generated response quality. LE-PARIS, in particular, shows stronger semantic representations and higher practical impact, driven by LLM embeddings and a generative component that integrates external resources with attorney input. The work delivers a scalable OA response workflow, a large OA Topics Database of topics, templates, and a cascade hybrid recommender, establishing a foundation for practical, confidential patent prosecution automation and guiding future research on domain-specific LLM deployment and human-AI collaboration.

Abstract

In patent prosecution, timely and effective responses to Office Actions (OAs) are crucial for securing patents. However, past automation and artificial intelligence research have largely overlooked this aspect. To bridge this gap, our study introduces the Patent Office Action Response Intelligence System (PARIS) and its advanced version, the Large Language Model (LLM) Enhanced PARIS (LE-PARIS). These systems are designed to enhance the efficiency of patent attorneys in handling OA responses through collaboration with AI. The systems' key features include the construction of an OA Topics Database, development of Response Templates, and implementation of Recommender Systems and LLM-based Response Generation. To validate the effectiveness of the systems, we have employed a multi-paradigm analysis using the USPTO Office Action database and longitudinal data based on attorney interactions with our systems over six years. Through five studies, we have examined the constructiveness of OA topics (studies 1 and 2) using topic modeling and our proposed Delphi process, the efficacy of our proposed hybrid LLM-based recommender system tailored for OA responses (study 3), the quality of generated responses (study 4), and the systems' practical value in real-world scenarios through user studies (study 5). The results indicate that both PARIS and LE-PARIS significantly achieve key metrics and have a positive impact on attorney performance.
Paper Structure (24 sections, 1 equation, 6 figures, 3 tables, 1 algorithm)

This paper contains 24 sections, 1 equation, 6 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries and Background
  3. System Overview
  4. Study 1: Topic Modeling
  5. Data Acquisition
  6. Data Prepocessing
  7. Latent Dirichlet Allocation (LDA)
  8. Results
  9. Study 2: Panel Consensus
  10. Participants and Measurements
  11. Convergent Delphi Procedure
  12. Results
  13. Study 3: Recommendation System
  14. Proposed Hybrid Recommender System
  15. Experimental Setup
  16. ...and 9 more sections

Figures (6)

  • Figure 1: The Overview of PARIS and LE-PARIS. (a) Construction of an OA Topics and Legal Keywords (L-keywords) Database. (b) Development of Response Templates Based on OA Topics and Legal Keywords. (c) Integration of the Response Template Recommendation System. (d) Patent Attorney Workflow for Responding to Office Actions. (T-keywords is short for technical keywords) (e) Semi-Autocompletion Feature in OA Response System (f) Generative Components in LE-PARIS System. (g) User Data Recording and Feedback Mechanism in System Optimization.
  • Figure 2: Visualization of Results Topic Modeling. (a) Intertopic Distance Map. (b) Word Cloud for a specific topic (which could involve 35 U.S. Code § 103).
  • Figure 3: The Convergent Delphi Process Across Different Periods for OA Topics and Legal Keywords.
  • Figure 4: Visualization Results of Our Hybrid Recommender Systems. (a) The evolution of recommender performance over time, measured by median Precision@10 and median Recall@10. Each line represents a combination of CB and CF techniques used in the recommendation process. The points plotted on each line are coordinates representing (median Recall@10, median Precision@10), with annotations indicating the year of analysis (e.g., '19' for the year 2019). (b) Three dimensional embedding space visualizations, representing the mere CB aspect of the system. The orange points mark the current OA accessed by a user, while the 'x' markers identify the top 10 recommended templates based on their proximity as the nearest neighbors in the embedding space. (b-i) corresponds to the PARIS system, which adopts Sentence-BERT embedding, while (b-ii) corresponds to the LE-PARIS system, which adopts LLM embedding.
  • Figure 5: Generative Results from Different LLMs
  • ...and 1 more figures