Integrating Semantic Clustering-Based Recommendation Models and RAG Generative AI Chatbots for Regional Innovation Analytics
TL;DR
To the authors' knowledge, this is the first platform integrating clustering based recommendation with RAG based querying for regional innovation monitoring, offering a replicable design for similar settings.
Abstract
Regional research and innovation agencies are responsible for monitoring large volumes of innovation data submitted by local government units. Non-standardized submissions, manually matched collaboration opportunities, and a growing data request burden have left this data underutilized for policymaking. This study aimed to design, implement, and evaluate an integrated platform for regional innovation monitoring at the Regional Research and Innovation Agency (BRIDA) of East Java Province, using a design and development research approach. The platform combines a clustering based recommendation model that groups semantically similar innovations using multilingual sentence embeddings, an AI chatbot built on Retrieval-Augmented Generation (RAG) for querying the innovation database, and descriptive analytical reports on innovation maturity. Tested on 643 innovation records, the platform achieved a 100 percent functional pass rate across 35 test scenarios, while the chatbot achieved mean faithfulness and relevance scores of 4.0 out of 5.0 under an LLM as a judge evaluation. These results suggest that embedding based clustering can reveal previously hidden collaboration opportunities, while a generative chatbot can meaningfully reduce manual search workload. To the authors' knowledge, this is the first platform integrating clustering based recommendation with RAG based querying for regional innovation monitoring, offering a replicable design for similar settings.