Social media has become a “public arena of citizenship”. The diverse customer activities on social media, such as debating social issues or punishing companies’ unethical actions, create a vast volume of dialog data that is valuable for improving business practices. This study examines corporate social responsibility (CSR) dialogs on Twitter and reveals how rich dialog data contributes to corporate social innovation (CSI). Drawing on value co-creation literature, a framework is proposed for text mining of Twitter dialogs and is tested using a U.K. retailer case study. The central assumption of the framework is that through comparing the worldviews of a company and its stakeholders embedded in the dialogs, the cognitive distance can be reduced (Lusch & Nambisan 2015), thus, allowing the company to design CSR programs in line with stakeholders’ expectation. In Twitter dialog data, three CSI contexts were identified in the company’s posts, namely exploring, co-creating and communicating. Based on these CSI contexts, customer knowledge regarding “know-what”, “know-who”, “know-how” and “know-why” (Mirvis et al. 2016) was extracted using text mining. We further investigated whether customer tweets contain CSI ideas through experts’ collective labeling, and we analyzed how customer knowledge helps predict crucial CSI ideas using decision tree models. The models achieve high performance in classifying non-ideas and CSI ideas, which assure the practical implications of the proposed dialog mining framework. Importantly, the framework helps to bridge cognitive distance by discovering the knowledge in the customer domain and promoting knowledge internalization in the company domain. Though the findings of the case study have difficulties in generalization, the mind-set and methodology is applicable, not limited to CSI, but to new product development and service innovation.