Personalized Travel Recommendation System Based On City Portrait And User Portrait

With the development of economy and society,tourism has become an important part of people’s daily life.Whether it is a weekend trip or a holiday trip,people often devote a lot of time and energy in the next travel destination and the corresponding route arrangement,expecting to gain a cost-effective travel experience with a good sense of experience.On the other hand,various online travel strategies and various tourist commodities consume a large amount of users’ time,greatly increasing users’ decision-making costs and reducing users’ pre-travel experience.So how to make better use of online users’ past tour routes,comments,pictures and other multi-dimensional information,after processing,refining and integration,service for existing users?This paper proposes a solution.Based on the user data of the platform "travel with weibo",this paper selects eight target cities including Beijing,Shanghai,Chengdu,Hangzhou,Xi’an,Chongqing,Dalian and Hong Kong,and realizes a personalized travel recommendation system based on "user portrait" and "city portrait" from scratch.First of all,this paper processed the crawled raw data to make it valuable information that can be used by the system.In this stage,"user portrait" and "city portrait" were realized.To construct the user portrait,the first step is to get the "age,gender,constellation,identity attributes" and other population labels by labeling the user’s basic attribute data.In addition,"photography lover/food lover/fashion lover"and other group labels are extracted based on semantic matching rules.The second step is to obtain the user’s travel type label according to the user’s comment text and scenic spot location data published in a certain scenic spot,so as to further enrich the user’s portrait.The construction of city portrait mainly revolves around two dimensions:The first dimension is to get the photo sets of 8 cities according to the pictures of users’ weibo.Through training the image classification model on baidu EasyDL platform,the distribution of images of 8 cities on 9 labels,such as modern elements,traditional elements,natural scenery and delicious food,is predicted to achieve the city portrait of image dimension.The second dimension is from the user text.According to the city aggregation and TextRank keyword extraction technology,the TOP30 keywords of each city on nouns,adjectives and verbs are obtained,so as to realize the city portrait of the text dimension.According to the scenic spots have the comment text under each scenic spot,respectively for attractions keywords,based on semantic matching rules available spots and mapping relationship of different types of tourism scenic spots is obtained by traversing the text heat distribution in the various types of tourism,implement attractions portrait,statistics and the distribution of each city in various types of tourism,for tourism city images of the type dimension.In this process,this paper proposes several reference dimensions to calculate the similarity of cities,scenic spots and users,and realizes them by combining the similarity calculation methods such as Euclidean distance,Jaccard distance and cosine similarity.The details are as follows:1.Calculate the similarity between cities according to the distribution of picture labels,text keywords and the distribution of tourism types of the city’s scenic spots;2.Calculate the similarity between users according to the distribution of user’s population label,ethnic group label and tourism type;3.Calculate the similarity between scenic spots according to the keywords of scenic spots or the heat distribution of each tourism type;Secondly,based on the above work to build a simple personalized recommendation system,the use of information recommendation algorithm for user interactions with the tourism system,through the user’s travel demand and explicit and implicit act of personal data,and then according to the travel constraints,recommend the most suitable for direct user tourist attractions,help users quickly decision making.Specific methods include memory-based collaborative filtering recommendation,content-based recommendation,demographic-based recommendation,tag-based recommendation and knowledge-based recommendation,as well as interest-based keyword search recommendation.This paper explains in detail the recommendation design,application effect,limitations and improvement direction of various recommendation methods.However,due to the limited amount of effective user data after cleaning in this paper,the user-scenic spot matrix is very sparse,and it is difficult to achieve the measurement of recommendation indexes through cross-validation and other methods.However,this paper shows how to make full use of users’ historical travel data in all dimensions to build a personalized travel recommendation system and the corresponding technology realization,and makes a comparative analysis on the goals,advantages,application value and application limitations.In the application and prospect part,this paper proposes to build a small intelligent interactive public platform for tourism recommendation based on the personalized tourism recommendation system,so as to provide feasibility for solving the actual tourism recommendation scenarios.Finally,this article use city portraits and user portraits to analyze the image of the city,from the tourist recommendation extends to construction of tourism destination image of tourist cities.Through content posted by users on influential social media platforms,mining users’ perception of the city image,some Suggestions are put forward for the construction of tourism city image and tourism development.