Retrieval Augmented Generation (RAG) has been pivotal in the utilization of Large Language Models (LLM) to improve the factuality of long-form question answering systems in industrial settings. Knowledge graphs (KG) represent a linking of disparate information sources that potentially yield useful information for mitigating the issues of insufficient knowledge and hallucination within the LLM-RAG pipeline. However, the creation of domain-specific KG is costly and usually requires a domain expert. To alleviate the above challenges, this work proposes QuARK, a novel domain-specific question answering framework to enhance the knowledge capabilities of LLM by integrating structured KG, thereby significantly reducing the reliance on the “generic” latent knowledge of LLMs. Here, we showcase how LLMs can be deployed to not only act in dynamic information retrieval and in answer generating frameworks, but also as flexible agents to automatically extract relevant entities and relations for the automated construction of domain-specific KGs. Crucially we propose how the pairing of question decomposition and semantic triplet retrieval within RAG can enable optimal subgraph retrieval. Experimental evaluations of our framework on financial domain public dataset, demonstrate that it enables a robust pipeline incorporating schema-free KG within a RAG framework to improve the overall accuracy by nearly 13%.

Aspect Based Sentiment Analysis is a dominant research area with potential applications in social media analytics, business, finance, and health. Prior works in this area are primarily based on supervised methods, with a few techniques using weak supervision limited to predicting a single aspect category per review sentence. In this paper, we present an extremely weakly supervised multi-label Aspect Category Sentiment Analysis framework which does not use any labelled data. We only rely on a single word per class as an initial indicative information. We further propose an automatic word selection technique to choose these seed categories and sentiment words. We explore unsupervised language model post-training to improve the overall performance, and propose a multi-label generator model to generate multiple aspect category-sentiment pairs per review sentence. Experiments conducted on four benchmark datasets showcase our method to outperform other weakly supervised baselines by a significant margin.