Semantic Similarity Matching for Patent Documents Using Ensemble BERT-related Model and Novel Text Processing Method
Liqiang Yu, Bo Liu, Qunwei Lin, Xinyu Zhao, Chang Che
TL;DR
The paper tackles the challenge of measuring semantic similarity between patent phrases within the Cooperative Patent Classification framework, a task complicated by language barriers and the intricacies of patent text. It introduces an ensemble approach that combines four BERT-related models (including DeBERTaV3 variants, BERT-for-Patents, and ELECTRA) and a novel V3 text preprocessing method that structures anchor-context pairs and trains with per-token scores using BCELoss. Key contributions include the weighted ensemble with validation-driven weights, the V3 preprocessing producing target and score lists, and comprehensive evaluation on the U.S. Patent Phrase-to-Phrase Matching dataset, achieving a final ensemble Pearson correlation of 0.8534. The work demonstrates improved CPC-based patent document analysis performance, with potential to enhance CPC classification accuracy and search capabilities in patent analytics.
Abstract
In the realm of patent document analysis, assessing semantic similarity between phrases presents a significant challenge, notably amplifying the inherent complexities of Cooperative Patent Classification (CPC) research. Firstly, this study addresses these challenges, recognizing early CPC work while acknowledging past struggles with language barriers and document intricacy. Secondly, it underscores the persisting difficulties of CPC research. To overcome these challenges and bolster the CPC system, This paper presents two key innovations. Firstly, it introduces an ensemble approach that incorporates four BERT-related models, enhancing semantic similarity accuracy through weighted averaging. Secondly, a novel text preprocessing method tailored for patent documents is introduced, featuring a distinctive input structure with token scoring that aids in capturing semantic relationships during CPC context training, utilizing BCELoss. Our experimental findings conclusively establish the effectiveness of both our Ensemble Model and novel text processing strategies when deployed on the U.S. Patent Phrase to Phrase Matching dataset.