Following the Great East Japan Earthquake, dark tourism was developed in the Tohoku Region of Japan. Notably, two government-built tsunami memorial parks in Ishinomaki and Rikuzentakata have obtained attention for their profound disaster narratives, iconic disaster sites, and expansive layouts. This study is the first to compare the projected destination image presented by destination management organizations with the perceived destination image held by visitors in these parks, and in dark tourism. Using online text data from both supply and demand sides of dark tourism and text mining analyses such as word frequency analysis, co-occurrence network analysis, and affection tendency examination, we revealed similarities and disparities between these two perspectives. Furthermore, this study concluded dimensions specific to dark tourism sites within the cognitive and affective destination image. Based on the findings, the study provides advice for destination managers to improve these sites, including developing non-dark tourism products and improving infrastructures. Additionally, it suggests fostering visitor engagement in local non-profit and citizen activities to strengthen connections with residents and addressing tourist dissatisfaction to improve park facilities. The findings demonstrate the effectiveness of text mining in comparing projected and perceived destination images in the context of dark tourism sites.