Hybrid NLP-Based Knowledge Discovery in Generative AI: A Systematic Literature Review

Research on Generative Artificial Intelligence (GenAI) has accelerated at an unprecedented rate, with an estimated annual growth of 18% in the number of publications across the world making it ever harder for researchers to keep up with new concepts, catch up with new knowledge architectures and predict future research foci. Current knowledge discovery methods mainly utilize statistical topic modelling methods like Latent Dirichlet Allocation (LDA), which provide interpretable topics and limited contextual understanding, or context-sensitive language models like BERT, which provide language model nudity at the cost of lack of interpretability and trend forecasting. This paper presents a novel hybrid approach that combines LDA, LSA and BERT-based contextual embeddings in a PRISMA-guided systematic literature search methodology, covering seven databases. As of now, the literature synthesis reveals a preliminary state of the art of current tools that are fragmented, are good at both statistical coherence and contextual richness, and, most importantly, largely fail to foresee new topics. The proposed framework should enhance the coherence of the topics, facilitate context-aware knowledge discovery, and provide an evidence-based decision-making and technology forecasting automated research-intelligence dashboard in the context of the GenAI space.