Deep Learning Based Travel Product Recommendation System

In the era of big data,recommendation systems are becoming increasingly important.As data volume increases,people need to spend more time and effort searching for content or products they like,while recommendation systems can help users quickly find interesting content and products.With the popularity of e-commerce,online platforms selling travel-related products have become increasingly diverse.In addition to traditional items such as airline tickets,train tickets,and hotels,there are now also homestays,travel routes,scenic area tickets,and special activities at tourist attractions.How to recommend the most suitable travel products to users has become a research problem.Therefore,designing an excellent recommendation model is crucial for travel platforms’recommendation systems.This article first introduces the research background and significance of recommendation systems for travel-related goods and elaborates on the development process of recommendation systems as well as the recommended algorithms used in this process.The recommended algorithms are divided into two categories:traditional recommended algorithms and deep learning-based recommended algorithms.The mainstream algorithm in each category is analyzed in detail with its working principle explained along with its advantages and disadvantages.Then we introduce DeepFM model based on deep learning technology and random forest model based on Bagging algorithm idea respectively.In DeepFM part we focus on analyzing improvements made by DeepFM compared to other deep learning models;In random forest part we analyze what attribute randomization means in Random Forest Algorithm along with data randomization meaning&characteristics of Random Forest Algorithm itself.Based on this analysis a method called DFRF which combines both models were proposed by connecting hidden layers from DeepFM model with output layer from Random Forest Model followed by final prediction through output layer resulting in improved predictive performance&robustness but may increase computational complexity¶meter count requiring more fine-tuning against overfitting or underfitting issues.Finally using dataset provided by Alibaba’s Fliggy Travel platform it was verified that DFRF outperforms NeuralCF Model,Wide&Deep Model and DeepFM Model in terms of recommendation effectiveness&stability.