Tourism Destination Recommendation based on bag of Visual Word Combined with SVM Classification

Abstract

In recent years, people's living standards have gradually improved, and more and more people plan to travel. In response to the low accuracy of user travel destination recommendation, this study proposes a travel destination recommendation method combining bag of visual word with support vector machine classification. Firstly, the study introduces a convolutional neural network extractor to improve bag of visual word. Subsequently, domain adaptation is introduced to address the distribution differences between the target feature data and the source domain feature data. Finally, by improving the support vector machine, the scenic spots that users are interested in are classified. Similarity calculation is adopted to achieve tourism destination recommendation. In the results, the accuracy of the collaborative filtering algorithm started to increase from 34.5%, and then reached 36.6% when the nearest neighbor users were 50 and tended to stabilize. The recall of collaborative filtering algorithm was 42.1% when the nearest neighbor users were 20, and 43.6% when the nearest neighbor users were 50. The proposed recommendation algorithm’s accuracy was 42.6% when the nearest neighbor users were 20, and 44.9% when the nearest neighbor users were 50. The recall started to increase from 43.8% and stabilized at around 44.8% when the nearest neighbor users reached 50. Overall, the designed tourism destination recommendation algorithm has strong practical applicability. This algorithm provides a strong recommendation strategy for many users to travel and helps them efficiently choose their desired attractions.