Study On Tourism Topic Representation And Tourist Preference Learning Based On Online Text

The amount of information in online tourism community is increasing at an astonishing speed.Users switch between the role of user and creator of information repeatedly,which makes online community maintain a good information ecology.Extensive users’ sharing of tourism knowledge,experience and interest has become an important tool in tourism marketing.The acquisition of user preferences from user-generated data is a research hotspot,which has the following difficulties:(1)user-product interaction data is subject to strong sparsity constraints,and it is difficult to learn users’ general preferences?(2)User preferences are not invariable,and there is a dynamic evolution process.At present,few studies pay attention to the characteristics of the establishment and evolution process of user travel preferences at the micro level.In order to solve the above difficulties,this paper identifies the multi-topic dynamic representation of users and points of interest from the massive user-generated data of online tourism community.Based on this,it studies the evolution characteristics of users’ travel preferences in time series,and deeply understands the establishment and development process of users’ preferences to provide more inspirations on tourism marketing.Aiming at the multi-theme representation learning task of tourism entity,firstly,the travel behavior sequence of users is extracted from the travel notes of users,and additional information is extracted from the text of users’ travel notes.Then,a POI theme representation model based on non-negative matrix decomposition algorithm incorporating external information is proposed,and the distribution of POI in the limited tourism theme space is obtained.The reliability of distribution is guaranteed by the group wisdom of users.The experimental results on real data prove the advantages of the model: firstly,the addition of external information can effectively reduce the cost of manual annotation and improve the model effect at the same time? secondly,the model is based on interactive data analysis,which has obvious advantages in the case of sparse user-generated text data.In view of the research on the characteristics of user preference evolution,firstly,a user preference representation algorithm was designed to represent the theme preference of the user travel sequence based on the existing POI themes,then the panel data of the user travel preference time series was constructed and the preference evolution was studied based on it.The individual fixed effect model is used to explore the evolution trend of related indicators over time from the two directions of the central tendency and stability of the distribution of users’ travel preferences respectively,and it is concluded that the preference dispersion degree presents a positive U-shaped relationship over time,which firstly decreases and then increases over time,while the preference stability continues to increase over time and finally maintains a high level in the whole life cycle.Then the management implications for user preference intervention are discussed.The technical contribution of this study lies in that it proposes a new idea to quantitatively measure the general preferences of users based on the multi-topic spatial representation of tourism based on matrix decomposition,and discusses the ways and efficiency of the integration of multiple types of external information.This model has portability under various management scenarios.The management contribution of this paper is that it enriches the existing research on the establishment process and development characteristics of user preferences in the field of tourism by exploring the dynamic evolution law of preferences.The research conclusions also provide useful management enlightenment for consumer preference intervention,travel recommendation and other directions.