The image of cultural heritage is the outcome of the collaborative efforts between the information supply subjects and information demand subjects. Adopting a supply-demand comparison approach, this study gathered online texts on professional projection and user perception from 2002 to 2021. Text mining techniques such as grounded theory, frequency analysis, sentiment analysis, and semantic networks were employed to deconstruct the projection and perception images of Madame Xian in Hainan from the three dimensions of cognition, emotion, and the entirety. This analysis facilitated the identification of the 20-year evolution of the cultural heritage image. The study revealed that the image of Madame Xian cultural heritage can be categorized into three groups, i.e., image of things, image of events and image of people. The emotional tendencies of both the projection and perception images were predominantly positive. The projection image showed a dissemination pattern of "historical figures → cultural heritage → festive activities + tourism performances" while the perception image exhibited a dissemination pattern of "performance activities → historical figures → festive activities". The cultural heritage image elements demonstrated a trend towards tourism, and the emotional tendencies shifted from negative to positive. The projection image network structure exhibited a trend towards centralization, while the perception image network tended to be multi-centered.