[Objective]In response to the issue of low accuracy and efficiency in automatic extraction of patent technology triplets, this article explores a model for patented technology triplet extraction to improve the accuracy of personalized, fine-grained, multi-dimensional deep extraction and semantic association.

[Methods]This article proposes an extraction method based on the WeakLabel-Bert-BiGRU-CRF model for four technical thematic dimensions: problems, solutions, functions, and effects. The model is evaluated using indicators such as macro average.

[Results] Patents in the field of graphene energy storage applications were selected as the dataset. The experimental results showed that compared to the Bert-BiGRU-CRF model, the proposed method achieved a macro average of over 0.8 for triplet extraction, further reducing the workload of data annotation and achieving better extraction results.

[Limitations] The model proposed in this article requires the joint participation of domain experts and patent intelligence analysts in data annotation, and differences in annotation quality can have an impact on application effectiveness.

[Conclusions] Based on this model, this article has developed a corresponding prototype system for further use and promotion of the patent technology triplet extraction method in the future, which also has a broad application prospect in the field of scientific and technological literature knowledge mining.