Multi-tourism Scenario Recommendation Methods With Heterogeneous Information Fusion

With the rapid economic development and the continuous improvement of people’s living standards,tourism has become an important way of leisure and entertainment.In view of different travel scenarios,users have diverse needs for travel recommendation systems.For instance,some users want Online Travel Agencies(OTA)to provide personalized travel products and services.Some users hope that travel product recommended by OTA can be interpreted and take user dynamic preferences into account.Some users visit unfamiliar Places-of-Interest(POI)in foreign cities and want travel and want tour recommender to help them plan a personalized itinerary based on their trip constraints(e.g.,time limits,start and end points).Most tourists prefer a trip with friends or family.Group itineraries should maintain a balance between the group preferences and the given temporal and spatial constraints.These tourism scenes contain a large amount of heterogeneous information,which is reflect in:(ⅰ)The types of tourism data include continuous or discrete,can be static or dynamic,can be structured or semi-structured,may be complete or missing.(ⅱ)Travel data includes user-side,item-side,and global.(ⅲ)Travel subject can be a user or a group.How to mine users’ online behavior characteristics,geographic location information and offline trajectory data,and have learn the diverse user requirements of travel in the complex tourism scenarios,is an important problem for the travel recommendation system.In response to these issues,the research work mainly includes:(1)We propose a travel product recommendation model based on matrix factorization and information fusion.Existing studies only consider the user-specific and item-specific features,but ignore the global features,and simply regard the unobserved values as the negative feedback.In view of this,we consider user-specific features,item-specific features and global features comprehensively,and fuses the probabilistic matrix factorization on the user-item interaction matrix with the linear regression on a suite of features constructed by the multiple auxiliary information.Meanwhile,the model is built by a whole-data based learning approach which utilizes unobserved data to increase the coupling between probabilistic matrix factorization and linear regression.Also,the importance of features is examined to reveal the crucial auxiliary information having a great impact on the adoption of travel products.(2)We propose a travel product recommendation model based on user dynamic preference and intention recognition.User preferences changes over time,user interests in the stream will dynamically change over time.In addition to the clicked travel products,the user’s purchase intention will also be identified through other supervision signals.In view of this,we regard the title of travel products as additional supervision signals to learn the common intentions contained in similar products,and propose a multi-task learning method to recommend travel products based on users’ dynamic preferences.Modeling combined with the keyword generation task of identifying user intent can simultaneously improve recommendation performance and improve the interpretability of recommendation results.At the same time,the model considers the long-term and short-term preferences of users in both travel product recommendation and keyword generation tasks.(3)We propose a personalized itinerary recommendation based on deep and collaborative learning with textual information.Existing studies usually ignore the temporal and spatial constraints of personalized itineraries or only model user preferences via POI categories,which will cause a cold start problem.In view of this,we employ an unsupervised deep learning model to embed the POI textual contents,and then propose a DCC model,which seamlessly integrates the embedded POI textual contents with the traditional and widely used user-POI visits and POI categories,to predict the user interests as well as the visit durations.Then,after formulating itinerary construction as a variant of the Orienteering Problem,an Iterated Local Search based algorithm is proposed to calculate the visit sequence with maximized satisfaction consists of multiple POIs and personalized POI visit durations,i.e.,the optimal itinerary.(4)We present an attentive multi-task learning-based group itinerary recommendation.Most traditional group itinerary recommendation methods adopt predefined preference aggregate strategies without considering the group members’ distinctive characteristics and inner relations.Besides,POI textual information is beneficial to capture overall group preferences but is rarely considered.Toward this end,this paper proposes an attentive multi-task learning-based group itinerary recommendation framework,which can dynamically learn the inner relation between group members and obtain consensus group preferences via the attention mechanism.Meanwhile,this method integrates POI categories and POI textual information via another attention network.Finally,the group preferences are used in a variant of the orienteering problem to recommend group itineraries.