The rapid growth of social media has generated vast data with significant business value, making the study of user geographical locations recognition crucial for understanding user behavior and supporting various applications. To address the problem of mining interactions between users and geographical locations, this paper proposes a RoBERTa-based ULA-Rel entity relation extraction model focused on the interactions between the current user and their location. The model tackles issues such as relation overlap and the low accuracy of traditional methods when dealing with pronouns like “I” and “we”, which refer to the current user. The ULA-Rel model transforms the traditional entity relation extraction, or triplet extraction, problem into three binary classification tasks. Compared to traditional triplet extraction methods, this approach more effectively identifies the user’s past, present, and potential future locations, enriching user profiles and providing insights into the characteristics and areas of interest of users in different regions.