Translate patent vacancies into human-readable texts: Identifying technology opportunities with text embedding inversion

Abstract

Many attempts have been made to identify technology opportunities by finding vacancies in patent maps, but the inherent limitation of being unable to clearly interpret the technological content of the discovered patent vacancies still remains unresolved. This study proposes a novel generative approach to uncovering technology opportunities from patent maps using machine learning techniques. The proposed approach employs the embedding inversion technique, which restores high-dimensional embeddings to their original data form, to translate patent vacancies into human-readable texts. The process involves five steps: 1) transforming patent abstracts into high-dimensional vectors using a text embedding model, 2) training an autoencoder to project high-dimensional embeddings into a two-dimensional space and enable bidirectional mapping, 3) constructing a grid-based patent map using kernel density estimation, 4) identifying vacant cells and their coordinates as patent vacancies, and 5) reconstructing the vacancy coordinates into high-dimensional embedding vectors using the decoder, and generating human-readable texts using vec2text. To demonstrate the proposed approach, a case study on LiDAR technology was conducted using 17,616 collected patents. The results verified that the proposed approach can successfully identify patent vacancies and translate them into human-readable texts, which indicates its potential as a highly practical and useful instrument for technology opportunity analysis.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 6 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.