Elsevier

Neurocomputing

Volume 173, Part 3, 15 January 2016, Pages 789-798
Neurocomputing

Attraction recommendation: Towards personalized tourism via collective intelligence

https://doi.org/10.1016/j.neucom.2015.08.030Get rights and content

Highlights

  • This paper proposes a framework of personalized attraction recommendation in tourism.

  • It mines collective intelligence from heterogeneous travel multimedia on social media.

  • PAS-model is employed to recommend similar attractions with explicit interaction.

  • We can fuse heterogeneous collective intelligence with weight-adaptation.

  • Context information is considered to refine the final results for freshness and surprise.

Abstract

Travel recommendation systems can tackle the problem of information overload and recommend proper attractions on the basis of users׳ preferences. Most existing travel recommendation systems utilized travel history, yet neglected the low frequency of tourism and the flexible styles of attractions in different cities, which will cause the inaccuracy in both collaborative filtering recommendation and content-based recommendation. To deal with this issue, we propose a novel personalized travel recommendation framework by leveraging explicit user interaction and multi-modality travel information. As far as we known, it is the first time that attractions are recommended by user interaction and collective intelligence in a unified framework. Specifically, we first collect heterogeneous travel information by multi-user sharing, which is regarded as collective intelligence to provide reliable references by other travelers. Second, valuable knowledge is mined from collective intelligence in order to filter out the noisy data and make travel information structured. Then, personalized attraction similarity (PAS) model is designed to suggest attractions through fusing heterogeneous information with weighted adaptation and simultaneously considering explicit user interaction. Finally, context information such as the user׳s location is well adopted to refine the recommendation that may influence the user׳s choice at a particular moment. Experimental results on pseudo-relevance data and real-world data demonstrate that our method gains promising performance in terms of effectiveness as well as efficiency.

Introduction

With attraction the improvement of people׳s daily life, tourism has become more and more popular. Moreover, with the rapid development of Internet technology and the rise of social media, the users׳ requirements for the quality of travel service have become more and more high. However, it is challenging that valuable information can be quickly and correctly picked out from massive travel information. For generic, most existing commercial tourism websites show the must-see attractions in the location city on the basis of user ratings to travelers. When travelers wish to explore the places where they have not previously been to, it is rather difficult for them to schedule a perfect trip in view of personal interests and characteristic tourism. Consequently, in order to satisfy users׳ personal requirements and content-based recommendation can be introduced for personalized attraction recommendation. Meanwhile, when the attractions are so appealing, travelers will take souvenir photos, write comments and make scores. Thus, the heterogeneous information uploaded by travelers can be considered as their travel preferences and experiences, namely collective intelligence. Moreover, considering massive travel information, an intelligent website or system should take advantage of collective intelligence for content-based personalized attraction recommendation. Therefore, it is more desirable to mine knowledge from heterogeneous collective intelligence and combine personalization in the coming intelligent travel recommendation system.

The traditional dominant travel recommendation approaches are roughly divided into two categories: collaborative filtering recommendation [1], [2] and content-based recommendation [3], [4]. Travelers may travel once or twice a year even less on average, so travel-based user data is very sparse. Although collaborative filtering recommendation is much easier to implement than content-based recommendation, it will cause cold-start problem which depends on productive users׳ behaviors and profiles. Consequently, collaborative filtering approaches [5], [6] are not appropriate for sparse data in travel situation. By comparison, content-based approaches can handle the sparse user data but only can cope with single-modality information instead of heterogeneous information. And some works [7], [8] focus on visual information classification and the visual-based classification problem is well-suited for the attraction recommendation, because visual information can vividly represent the attractions. In reality, travel information contains massive heterogeneous information [9], including text, image and numerical value. In the case of travel recommendation, the content-based recommendation should be adopted in order to solve the problem with heterogeneous information from social media. Aiming to make full use of abundant heterogeneous resources in social media, the knowledge of collective intelligence will be explored to address the cold-start problem.

To further establish the personalized recommendation, the user information is collected either explicitly or implicitly. Explicit collection [10] means that user providing information actively, where the users should answer the questions provided by the interface for interaction. Implicit collection [11] is a passive way to mine users׳ interests from user historical behavior and context information. However, implicit way is not available for travel recommendation, because different cities have a diversity of architectural styles so that travel histories are not the best way as prior knowledge. For example, there are many different styles of parks around the world. The visual characteristics are not uniform in parks in different cities or even in the same city. Obviously, implicit information, such as user history, is not suitable for travel recommendation.

According to above analysis, due to the intrinsic characteristics of tourism, content-based recommendation with user explicit feedback, inevitably, is more suitable for travel personalized attraction recommendation. Therefore, we formulate a novel framework of travel attraction recommendation with personalization which consists of four principal modules, such as collective intelligence collection, knowledge extraction, PAS-model and user interaction. As shown in Fig. 1, in order to learn experiences of other travelers, we first collect travel heterogeneous information as collective intelligence from various travel-related websites on the Internet. Photos with metadata are crawled from Flickr [12] which are searched by the GPS location of each attraction. Meanwhile, official travelogues from Wikitravel and comments from Tripadvisor [13] are searched by the name of the attraction. In the same manner, ratings by travelers from Tripadvisor are tailed up. Then, knowledge is multi-modality descriptions of attractions, which is extracted from collective intelligence in different aspects, i.e., content-based, semantic-based and social-based. And then, given user interaction to avoid data sparsity and cold-start problem, the personalized attraction similarity model (PAS-model) is established with a combination of knowledge fusion to recommend attraction in a comprehensive view. In the model, each aspect of knowledge can construct graph-links between attractions with appropriate similarity measure. To realize personal recommendation, the user can choose favorite and unfavorite attractions as positive and negative labels in an explicit way, and the recommendation problem is considered as graph-based classification. Candidate recommended attractions are classified by graph-based multi-modality attraction information fusion in the way of weighted adaptation. Finally, users׳ current situation is utilized properly as context information to optimize the candidate attractions that can influence the user׳s choice under a particular condition. To give an intuitive way, personalized attraction recommendation is shown as an attraction ranking list in our system.

Our contributions are summarized as: (a) personalized attraction recommendation with explicit interaction is first composed in the personalized travel recommendation by analyzing collective intelligence from social media; (b) PAS-model is designed in a unified way to solve the recommendation problem which can mine the intrinsic links between modalities of heterogeneous information and fuse heterogeneous collective intelligence with weight adaptation; (c) context information are considered to refine the recommended attractions to simulate the particular situation to predict user׳s favors. The rest of the paper is organized as follows. Section 2 briefs related work. Then, in Section 3, we introduce our personalized attraction recommendation framework. The details of our framework are introduced in Section 3. Experiments and discussions are presented in Section 4. The final is conclusion in Section 5.

Section snippets

Related work

In an unfamiliar city, travelers want to visit both popular and favorite attractions. Travel attraction recommendation infers what the users׳ preferences and shows interesting and popular attractions for users to plan trips for them. Intelligent attraction recommendation on the basis of travel information and user׳s information is a hot topic. In general, dominant methods are classified as collaborative-filter (CF)-based, content-based methods.

On one hand, users׳ traveling histories, groups of

Problem definition

The purpose of our method is to fuse multi-modality of collective intelligence to recommend attractions based on explicit user feedback. We have the assumption that the recommended attractions are similar to user feedbacks. And travel attraction recommendation system is developed as input a set of famous attractions in each city and collective intelligence from social media. First, heterogeneous information is collected from social media as collective intelligence. Second, content-based,

Evaluation methodology

For our travel attraction recommendation framework, we are intended to evaluate the framework from five aspects, including accuracy, surprise, computational efficiency and freshness . Following sections will introduce the details.

Conclusion

It is desperately needed to recommend relevant attractions to users with the consideration of travel information and user interaction. In this paper, we present a novel framework of personalized travel attraction recommendation by leveraging explicit user interaction with the system and heterogeneous travel information. Aiming to learn from tourists who had just been there, collective intelligence is first gathered from large amount of user-generated content in social media. Second, different

Acknowledgment

This work is supported by the National High Technology Research and Development Program of China (2013AA01A602), the National Natural Science Foundation of China (Nos. 61125204 and 61432014), the Program for New Century Excellent Talents in University (NCET-12-0917), the Fundamental Research Funds for the Central Universities (Grant Nos. BDZ021403 and K5051302019), the Key Science and Technology Program of Shaanxi Province, China (2014K05-16) and Program for Changjiang Scholars and Innovative

Junge Shen received the B.Sc. degree in electronic engineering from Xidian University in 2010. Currently she is pursuing her Ph.D. degree at School of Electronic Engineering, Xidian University. Her current research mainly focuses on multimedia and computer vision.

References (29)

  • A.K. Jain et al.

    Image retrieval using color and shape

    Pattern Recognit.

    (1996)
  • B. Sarwar, G. Karypis, J. Konstan, J. Riedl, Item-based collaborative filtering recommendation algorithms, in:...
  • J.L. Herlocker, J.A. Konstan, J. Riedl, Explaining collaborative filtering recommendations, In: Proceedings of the 2000...
  • C.D. Manning et al.
    (2008)
  • D. Billsus et al.

    User modeling for adaptive news access

    User Model. User-Adapt. Interact.

    (2000)
  • S.K.L, Al Mamunur Rashid, G. Karypis, J. Riedl, Clustknn: a highly scalable hybrid model memory-based cf algorithm,...
  • B. Sarwar, G. Karypis, J. Konstan, J. Riedl, Application of dimensionality reduction in recommender system—a case...
  • F. Zhu et al.

    Weakly-supervised cross-domain dictionary learning for visual recognition

    Int. J. Comput. Vis.

    (2014)
  • L. Shao et al.

    Feature learning for image classification via multiobjective genetic programming

    IEEE Trans. Neural Netw. Learn. Syst.

    (2014)
  • T. Kurashima, T. Iwata, G. Irie, K. Fujimura, Travel route recommendation using geotags in photo sharing sites, In:...
  • G. Jawaheer, M. Szomszor, P. Kostkova, Comparison of implicit and explicit feedback from an online music recommendation...
  • M. Balabanovic, An interface for learning multi-topic user profiles from implicit feedback, In: AAAI-98 Workshop on...
  • Flickr...
  • Tripadvisor...
  • Cited by (55)

    • Towards comprehensive approaches for the rating prediction phase in memory-based collaborative filtering recommender systems

      2022, Information Sciences
      Citation Excerpt :

      For better results, e-commerce needs to invest resources to overcome this obstacle. One useful solution is to personalize item lists to suit users [6]. The personalized item lists are automatically generated by systems called recommender systems [4,7].

    • How smart is e-tourism? A systematic review of smart tourism recommendation system applying data management

      2021, Computer Science Review
      Citation Excerpt :

      Furthermore, suggesting a travel plan using the recommended places can enhance the place recommendations for tourists [114]. The current framework should be tested on mobile gadgets which could speed up the interaction between user and system [101]. Improving system accuracy can enhance the efficiency of personalised tourism information service framework [69].

    View all citing articles on Scopus

    Junge Shen received the B.Sc. degree in electronic engineering from Xidian University in 2010. Currently she is pursuing her Ph.D. degree at School of Electronic Engineering, Xidian University. Her current research mainly focuses on multimedia and computer vision.

    Cheng Deng received the B.Sc., M.Sc., and Ph.D. degrees in signal and information processing from Xidian University, Xi׳an, China. He is currently a Full Professor with the School of Electronic Engineering at Xidian University. His research interests include computer vision, multimedia processing and analysis, and information hiding. He is the author and coauthor of more than 40 scientific articles at top venues, including IEEE T-MM, T-SMC, T-IP, Signal Processing, Neurocomputing, ICCV, CVPR and AAAI.

    Xinbo Gao received the B.Eng., M.Sc. and Ph.D. degrees in signal and information processing from Xidian University, China, in 1994, 1997 and 1999 respectively. From 1997 to 1998, he was a research fellow in the Department of Computer Science at Shizuoka University, Japan. From 2000 to 2001, he was a postdoctoral research fellow in the Department of Information Engineering at the Chinese University of Hong Kong. Since 2001, he joined the School of Electronic Engineering at Xidian University. Currently, he is a Professor of Pattern Recognition and Intelligent System, and Director of the VIPS Lab, Xidian University. His research interests are computational intelligence, machine learning, computer vision, pattern recognition and wireless communications. In these areas, he has published 5 books and around 150 technical articles in refereed journals and proceedings including IEEE T-IP, T-CSVT, T-NN, T-SMC, Pattern Recognition etc.

    He is on the editorial boards of several journals including Signal Processing (Elsevier), and Neurocomputing (Elsevier). He served as general chair/co-chair or program committee chair/co-chair or PC member for around 30 major international conferences. Now, he is a Fellow of IET and Senior Member of IEEE.

    View full text