Connecting the Knowledge Dots: Retrieval-augmented Knowledge Connection for Commonsense Reasoning

While large language models (LLMs) have achieved remarkable performance across various natural language processing (NLP) tasks, LLMs exhibit a limited understanding of commonsense reasoning due to the necessity of implicit knowledge that is rarely expressed in text. Recently, retrieval-augmented language models (RALMs) have enhanced their commonsense reasoning ability by incorporating background knowledge from external corpora. However, previous RALMs overlook the implicit nature of commonsense knowledge, potentially leading to the retrieved documents not directly contain information needed to answer questions. In this paper, we propose Retrieval-augmented knowledge Connection, RECONNECT, which transforms indirectly relevant documents into a direct explanation to answer the given question. To this end, we extract relevant knowledge from various retrieved document subsets and aggregate them into a direct explanation. Experimental results show that RECONNECT outperforms state-of-the-art (SOTA) baselines, achieving improvements of +2.0% and +4.6% average accuracy on in-domain (ID) and outof-domain (OOD) benchmarks, respectively 1 . edge into LLMs to complement their commonsense reasoning capabilities. To enhance the reasoning capability of LLMs, RALMs have been introduced to incorporate relevant information from external corpora into the reasoning process (Su et al., 2024; Wang et al., 2025) . Recent studies employ a variety of external knowledge sources, such as textual documents (Yu et al., 2022) or exemplars of QA (Molfese et al., 2024) , to supplement LLMs with the contextual grounding they often lack. These approaches have yielded notable performance gains in commonsense reasoning tasks (Yu et al., 2022; Molfese et al., 2024) . However, previous RALMs have two challenges that arise from overlooking the nature of implicit commonsense knowledge. First, the commonsense question usually does not explicitly represent the required knowledge. For example, in Figure 1 , while understanding concepts like air resistance or net forces is essential to answer the given question,