Table 1 Overview of prior research in policy recommendation and document Interpretation.

⁶

Fine-tuning, RLHF, Prompt Engineering, RAG

Various datasets for personalized LLM evaluation

Explores various personalization techniques for LLMs to improve user-specific responses

Lack of high-quality user-labeled datasets, cold-start problem, privacy concerns

⁷

RAG, Dense Vector Retrieval (FAISS), Hybrid Search (BM25 + Neural Reranking)

Proprietary unstructured documents

RAG significantly enhances response relevance by incorporating external knowledge

Scaling RAG, reliability issues, hallucination in LLMs

⁸

NER, Transformer-based NLP models (BERT, RoBERTa), OCR

9 Swedish condo insurance policy documents

Automates information extraction, reducing manual effort in insurance policy analysis

Imbalanced dataset, limited positive samples, need for human interpretation

⁹

Rule-based Chatbot, Transformer-based Conversational AI (DialoGPT), Intent Recognition (BERT)

Cornell movie dialogue dataset, custom insurance dataset

Enhances customer support by automating policy-related queries

Limited by dataset scope, lack of real-time learning

¹⁰

Decision Trees, Random Forest, NLP-based Claim Analysis (BERT, LSTM)

88 real-life road traffic accident cases

AI-driven models improve dispute resolution efficiency and fairness

Small dataset, legal language complexity