Improving tourism recommender system through quantifying reviewer credibility

Pao Yue-kong Library Electronic Theses Database

Improving tourism recommender system through quantifying reviewer credibility


Author: Wang, Yuanyuan
Title: Improving tourism recommender system through quantifying reviewer credibility
Degree: Ph.D.
Year: 2015
Subject: Recommender systems (Information filtering)
Tourism -- Information services.
Hong Kong Polytechnic University -- Dissertations
Department: Dept. of Computing
Pages: xxii, 123 pages : illustrations ; 30 cm
Language: English
InnoPac Record:
Abstract: With the growing interconnectedness of the world and advances in transportation and communication, an increasing number of people are travelling as independent tourists, putting together their own itineraries and activities from information researched from social media. Moreover, a growing number of travellers post reviews and give ratings online to share experiences and opinions, which have become one important source of information. However, the explosive growth of travel information and the proliferation of uninformative, biased or false information make it very time-consuming and challenging for travellers to find helpful and credible information.Recommender systems can assist travellers in managing the information available and facilitate their travel decisions. There have been some recommender systems developed in the tourism domain. However, these systems usually apply collabo-rative filtering-based, content-based or knowledge-based approaches, which require historical ratings, description of items, or extra knowledge about users' needs. It is difficult for them to generate reliable recommendation when the ratings, description as well as the knowledge are insufficient, and they cannot make recommendations when there is no such information. In the tourism domain, ratings, description and knowledge available are much fewer than the equivalent for books or movies. Therefore, they usually suffer from the sparseness and cold-start issues.Addressing the sparseness issue, we apply rating inference method to augment ratings for rating-based recommender systems. We investigate several clustering approaches to do sentiment analysis on travel reviews to generate numerical ratings. The clustering methods include K-means, co-clustering, hierarchical co-clustering, and six state-of-the-art traditional hierarchical clustering algorithms. Moreover, we compare different features extracted from reviews to choose more suitable features for each clustering method.Experimental results show that hierarchical algorithms (traditional hierarchical clustering and hierarchical co-clustering) with stepwise exploiting strategy lead to more accurate clustering results than non-hierarchical algorithms (K-means and co-clustering). Especially, hierarchical co-clustering method gets better clusters than all of the other clustering methods, no matter what features it uses. From the investigation, we also found that it is difficult to get very accurate multi-rating clusters by using these unsupervised approaches.Rating inference on reviews can augment ratings for recommendation, but it is not helpful for solving the cold-start issue, since a new item or traveller has no review. Therefore, a demographic recommender system is applied to the recommendation of attractions to overcome the cold-start problem. Our system categorizes travellers using their demographic information and then makes recommendations based on demographic classes. Its advantage is that the history of ratings, description of attractions and extra knowledge are not necessary, so even a new traveller can obtain recommendations. Different machine learning methods are adopted to produce prediction of ratings, so as to determine whether these approaches and demographic information are suitable for providing recommendations. Our preliminary results show that these machine learning methods and demographic information can be used to predict travellers' ratings on attractions. But only limited accuracy can be achieved using demographic information alone.
Although recommender systems are able to provide travellers with recom-mendations, most of them make recommendations based on existing ratings or reviews, which may contain uninformative, biased or even false information.Recommendations will be not so helpful or reliable if the recommender systems generate them based on these unreliable information. For instance, TripAdvisor, the world's largest travel community supplies a recommender system which can ranks reviews on an attraction for travellers based on posting dates or ratings. Travellers can then read some top ranked reviews on the attraction. However, there may be some incredible information involved in the top ranked reviews. Hence, it is critical to help travellers seek credible information from such amounts of travel information. Most current work applies mainly qualitative approaches to investigate the credibility of reviews or reviewers without quantitative evaluation.This thesis presents a method that quantifies the credibility of reviewers, to help travellers find more credible information. We propose an Impact Index to quantify the credibility of reviewers by simultaneously evaluating the expertise and trustworthiness based on the number of reviews posted by reviewers and the number of helpful votes received by those reviews. Furthermore, Impact Index is enhanced into the Exposure-Impact Index by considering in addition the number of destinations on which the reviewer posted reviews. Our experimental results show that both methods perform better than the state-of-the-art method in discovering credible reviewers. To further examine the effectiveness and applicability of Impact Index and Exposure-Impact Index, we evaluate them on the data sets collected from two rather different online travel communities: TripAdvisor, the world's largest travel community, and Qunar, one of the most popular travel communities in China, by taking into consideration the differences between these two travel communities, such as different languages, scales and data distributions. Experimental results show that both Impact Index and Exposure-Impact Index lead to results more consistent with human judgments. They can not only discover more credible reviewers, but also provide better ranking of reviewers, which manifest their effectiveness and applicability across diverse travel communities.

Files in this item

Files Size Format
b28163898.pdf 2.823Mb PDF
Copyright Undertaking
As a bona fide Library user, I declare that:
  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.


Quick Search


More Information