【作者】 贾伟；

【导师】 陈泽华； 王悦环；

【作者基本信息】 太原理工大学 ， 软件工程（专业学位）， 2024， 硕士

【摘要】 近年来,国内旅游业发展迅速,已成为老百姓日常生活和国民经济的重要组成部分。现在自助出游人数增多,虽然网上有很多景点介绍,但用户更需要根据个人喜好进行个性化推荐。因此,构建一个具备个性化和多样化功能的景点推荐系统,对于促进文化旅游产业的蓬勃发展以及为用户提供便捷、贴心的建议,具有举足轻重的意义。利用传统的旅游信息规划旅行时,用户通常需要花费大量时间搜索和比较各种旅游信息做攻略,费时费力,有时效果还不好。因此,本文研究基于深度学习的景点推荐算法,通过构建用户-景点异构图发掘用户景点之间的复杂关系以实现推荐的个性化,并通过增加冷门景点的权重以增加推荐的多样性,从而为用户提供精准和实用的旅游建议,节省用户的时间和精力,提升用户出行品质。本文主要研究内容如下:(1)提出基于图卷积神经网络与多层感知机的景点推荐算法GCN-MP(Graph Convolutional Networks-Multilayer Perceptron,GCN-MP)。首先,构建了用户-景点异构图,利用GCN网络进行消息传递以学习更新嵌入式表示;其次,利用MP学习更复杂的嵌入式表示得到正样本图和负样本图的预测得分;然后,基于景点类别特征,通过预测得分计算损失函数提高长尾项目的权重。最后,对GCN-MP算法进行验证。研究不同参数(如列表长度、卷积层数)对算法性能的影响并结合项目实际情况确定选取的推荐列表参数和卷积层数;利用爬取到的蜂窝网用户评价信息数据集进行实验。通过模型对比分析证明了该算法相较于其他算法命中率具有明显优势,同时覆盖率在推荐数据较多时也有很好的效果。(2)设计并实现了基于GCN-MP算法的景点推荐系统。首先,根据需求分析进行顶层设计;然后,系统采用三层架构实现:表现层进行功能模块展示,逻辑层进行实现业务逻辑,存储层进行数据库交互操作;最后,对系统功能模块设计用例测试验证系统的稳定性。根据实验结果证明,本文所提出的景点推荐系统能够有效地完成个性化和多样化景点推荐任务,并通过贪心算法计算经纬度对推荐列表实现了路径规划。更多还原

【Abstract】 In recent years,domestic tourism has developed rapidly and has become an important part of the daily life of the people and the national economy.Nowadays,the number of self-guided trips has increased,and although there are many spots introduced on the Internet,users are more in need of personalised recommendations based on personal preferences.Therefore,building an spot recommendation system with personalised and diversified functions is of great significance in promoting the flourishing development of the cultural tourism industry as well as providing users with convenient and considerate suggestions.When planning a trip using traditional travel information,users usually need to spend a lot of time searching and comparing various travel information to make a strategy,which is time-consuming,labour-intensive and sometimes ineffective.Therefore,this paper investigates the deep learning-based spot recommendation algorithm,which explores the complex relationship between user spots by constructing a user-spot heterogeneous graph to achieve the personalisation of the recommendation,and increases the weight of the cold spots to increase the diversity of the recommendation,so as to provide the user with accurate and practical travel advice,save the user’s time and energy,and enhance the user’s travel quality.The main research content of this paper is as follows:(1)Graph Convolutional Neural Networks-Multilayer Perceptron(GCN-MP)is proposed as an spots recommendation algorithm based on Graph Convolutional Neural Networks and Multilayer Perceptron(GCN-MP).First,a user-spots heterogeneous graph is constructed,and the GCN network is utilized for message passing in order to learn to update the embedded representations;second,the MP is utilized to learn more complex embedded representations to obtain the prediction scores of the positive sample graphs and the negative sample graphs;and then,based on the spots category features,the loss function is computed by the prediction scores to increase the weight of the long-tail items.Finally,the GCN-MP algorithm is validated.We study the influence of different parameters(e.g.,list length,number of convolutional layers)on the performance of the algorithm and determine the selected parameters of the recommendation list and the number of convolutional layers in combination with the actual situation of the project;and conduct experiments by using the crawled cellular network user evaluation information dataset.The model comparison analysis proves that the algorithm has a significant advantage over other algorithms in terms of hit rate,while the coverage rate also has a good effect when there are more recommendation data.(2)The spot recommendation system based on GCN-MP algorithm is designed and implemented.Firstly,the top-level design is carried out according to the demand analysis;then,the system is implemented using a three-layer architecture:the performance layer for the display of functional modules,the logic layer for the implementation of business logic,and the storage layer for the database interaction operation;finally,the system functional modules are designed to verify the stability of the system by testing the use cases.According to the experimental results,it is proved that the spot recommendation system proposed in this paper can effectively complete the task of personalised and diversified spot recommendation,and the path planning of the recommended list is achieved by calculating the latitude and longitude through the greedy algorithm.更多还原