The tourism industry is undergoing a profound transformation, with tourists’ demands and preferences showing a diversified trend. Accurately identifying tourists’ preferences and satisfaction levels is of crucial importance. Taking the Longji Terraced Fields Scenic Area in Guilin as a case study, deep learning technology and cluster analysis methods were employed to extract the themes of tourists’ service preferences from online reviews and identify groups of tourists with similar preferences. Combined with questionnaire survey data, the service satisfaction was quantitatively evaluated. The research finds that tourists’ service preferences cover accommodation, catering, transportation and information services. The improvement space for information services and accommodation services is the largest, while transportation services and catering services also have room for improvement. Based on this, corresponding suggestions are put forward.