Multifeature fusion for claim scope-aware litigation risk prediction for patent drafts

The ‘claim scope’, or the ‘legal boundaries’ defined by patent claims, has been considered crucial for determining a patent’s value and its associated litigation risk. However, no direct claim semantics-based indicators currently exist to quantify patent claim scope, and existing scope measures are primarily indirect, which limits their ability to capture the semantic nuances of claim text. Additionally, the reliance on post-grant features restricts the applicability of existing litigation prediction models to patent drafts. These limitations complicate the patent drafting process, during which claims are formulated without feedback on scope and litigation risk. This often leads to suboptimal claim articulation, resulting in inadequate protection, increased legal vulnerabilities, or reduced patent grant probability. To address this gap, the hyponym tree score (HTS) is proposed as a novel indicator for quantifying claim scope by analysing hyponym counts, sentence structure, and dependency relations within patent claims. Building on this, early-stage litigation risk prediction has been achieved using a new deep learning model, the Multifeature BERT-Powered Fusion for Author-level Patent Litigation Risk Analysis (MAPRA). The MAPRA model restricts its input features to those available at early stages, such as indicators derived from claim text, inventor information, assignee details, and HTS, ensuring applicability to both draft-stage and granted patents. Despite excluding all post-grant or acquired data, MAPRA achieves a superior area under the receiver operating characteristic curve (AUC) of 0.878, outperforming the most comparable prior study, which reports an AUC of 0.822 using both early-stage and immediate post-grant features. By quantifying claim scope and enabling early-stage litigation risk prediction, this research offers a valuable screening tool for patent drafters, examiners, attorneys, and innovators. It supports informed decision-making during drafting and helps mitigate potential litigation risks. Furthermore, it lays a foundation for future research on claim scope modeling and the development of predictive tools for intellectual property litigation management.