Changho Son

Technology trajectory prediction from patent corpora supports strategic R&D planning by revealing which technology clusters will gain industrial dominance before market evidence appears. Existing methods are either static—aggregating all years into a single representation—or require domain-expert annotation to identify breakthrough technologies. This paper presents TechTraj-BERT, a fully automated framework that leverages temporally-segmented Sentence-BERT embeddings, density-based clustering, and forward citation counts as proxy ground truth to predict which technology clusters will achieve high impact two years into the future. The framework makes four key contributions: (C1) a temporal embedding scheme that encodes annual patent cohorts and quantifies year-over-year semantic drift; (C2) an automated trajectory tracking algorithm that assigns each cluster one state from {emergence, growth, convergence, decline}; (C3) a forward-citation proxy evaluation protocol that operationalizes technology importance without human labels; and (C4) a Technology Momentum Index (TMI) combining semantic novelty with citation velocity as the central predictor feature. Applied to electrified mobility patents from Google BigQuery patents-public-data (CPC codes H01M10/4x, H01M10/48, G01R31/36–38, H02J7/0x; publication cohorts 2012–2019 with forward citations through 2022), TechTraj-BERT achieves the highest Precision@5 (0.600 vs. 0.400 for all baselines) and citation-weighted F1 (0.462 vs. 0.370 for the next-best baseline) among all evaluated models. All experiments are fully reproducible in Google Colaboratory using only public data and open-source Python libraries.

Chinese characters are an essential means of communication in the East Asian cultural regions. Chinese characters are characterized by many strokes and complex structures, some of which are very similar. However, the misrecognition of messy writing can significantly decrease the accuracy of Optical Character Recognition (OCR) systems. Therefore, filtering the messy writing is critical. We propose a novel federated learning-based deep learning model to classify messy Chinese handwriting, addressing the challenges posed by complex character structures and variations in writing quality. To validate the proposed method, we conducted experiments to compare our approach with global, local, FedAVG, and IPA federated learning methods. We evaluated our approach using the CNN, ResNet50V2, NASNetLarge, EfficientNetV2B0, EfficientNetB0, and swin transformer models. The results showed a 301.2% improvement in the test accuracy and a 241.54% increase in the area under the curve (AUC) score with the proposed method, particularly when EfficientNetB0 was trained using the global method. These findings confirm that the proposed model effectively classifies Chinese messy handwriting. Additionally, when the data were fed into the OCR system, the match rate between the image and the recognized characters improved by up to 7.14%.

This study proposes a framework for selecting and validating data sources for public-based technology foresight. In other words, it finds out which of the many online communities have valuable data sources. Specifically, we evaluate the usefulness of text data from online communities for technology foresight in terms of expertise and diversity. To this end, not only is a bibliographic analysis using metadata conducted, but also, topic modeling techniques for a semantic analysis of texts are utilized. As a case study, we selected 20 candidate communities where discussions and predictions related to technology are made and applied newly proposed metrics. As a contribution of this study, it is expected that it will provide a basis for public participation in technology foresight, not only leaving it to a few experts.

Recently, there have been many attempts to cope with increasingly-diversified and ever-changing customer needs by combining products and services that are critical components of innovation models. Although not only manufacturers, but also service providers, try to integrate products and services, most of the previous studies on Product Service System (PSS) development deal with how to effectively integrate services into products from the product-centric point of view. Services provided by manufacturers’ PSSes, such as delivery services, training services, disposal services, and so on, offer customers ancillary value, whereas products of service providers’ PSSes enrich core value by enhancing the functionality and quality of the service. Thus, designing an effective PSS development process from the service-centric point of view is an important research topic. Accordingly, the purpose of this paper is to propose a service-oriented PSS development process, which consists of four stages: (1) strategic planning; (2) idea generation and selection; (3) service design; and (4) product development. In the proposed approach, the PSS development project is initiated and led by a service provider from a service-centric point of view. From the perspective of methodology, customer needs are converted into product functions according to Quality Function Deployment (QFD), while Analytic Hierarchy Process (AHP) is employed to prioritize the functions. Additionally, this paper illustrates a service-oriented PSS development that demonstrates the application of the proposed process. The proposed process and illustration are expected to serve as a foundation for research on service-oriented PSS development and as a useful guideline for service providers who are considering the development of a service-oriented PSS.