Trust-based Modelling of Multi-criteria Crowdsourced Data

Abstract

As a recommendation technique based on historical user information, collaborative filtering typically predicts the classification of items using a single criterion for a given user. However, many application domains can benefit from the analysis of multiple criteria, e.g. tourists usually rate attractions (hotels, attractions, restaurants, etc.) using multiple criteria. In this paper, we argue that the personalised combination of multi-criteria data together with the creation and application of trust models should not only refine the tourist profile, but also improve the quality of the collaborative recommendations. The main contributions of this work are: (1) a novel profiling approach which takes advantage of the multi-criteria crowdsourced data and builds pairwise trust models and (2) the k-NN prediction of user ratings using trust-based neighbour selection. Significant experimental work has been performed using crowdsourced datasets from the Expedia and TripAdvisor platforms.

Trust and Reputation Modelling for Tourism Recommendations Supported by Crowdsourcing

“…In this context, the method includes: (i ) Profiling; (ii ) T&R Modelling (iii) On-line Rating Prediction; and (iv ) Evaluation. The Profiling includes: (i ) the Personalised Weighted Rating Average (PWRA) proposed by [15]; (ii ) the Trust modelling proposed by [17]; and (iii) two different approaches for reputation modelling -General Reputation and Neighbour-based Reputation -which are the contributions of this paper. For the On-line Rating Prediction we employ the k -NN algorithm and use Pearson correlation to identify the nearest neighbours.…”

“…Trust based approach, which was presented by [17], combines PWRA and trust modelling as described in Equation 3. This approach computes the number of times a user k was identified as a neighbour of a user u and the number of the items selected by u due to neighbour k. Equation 8 displays the rating predictionr Tu,i of item i for a user u employing the trustworthiness T u,k of the n neighbours of user u.…”

“…Since crowdsourced data corresponds to a stream of events, we propose an on-line recommendation approach via incremental updating, i.e., the model is updated whenever a new rating event occurs. In this work, user trustworthiness is based, according to the method [17], on the recommendations selected by the user and the reputation is modelled using the general and the neighbour-based approaches.…”

Analysis and prediction of hotel ratings from crowdsourced data

“…It is based on past interactions and is expected to be used in future interactions between the trustor and the trustee. Concerning trust modeling, Leal, Malheiro, González‐Vélez, and Burguillo () apply Pearson correlation to determine the correlation among users, and, then, build a decentralized trust model based on the recommendations followed by the active user; Wu, Forsman, Yu, and Song () adjust the weight between the active user and his/her neighbors, according to the group or the number of ratings involved.…”

“…The majority of the surveyed rating‐based profiling approaches use entity‐based modeling. In addition, while Wu et al () and Leal, Malheiro, et al () model the trust, the Jøsang et al () and Leal et al () explore the reputation in hotel recommendation supported by crowdsourced ratings.…”