Applied Soft Computing Journal 97 (2020) 106768 Contents lists available at ScienceDirect Applied Soft Computing Journal journal homepage: www.elsevier.com/locate/asoc A T1OWA and aspect-based model for customizing recommendations on eCommerce Jesus Serrano-Guerrero ∗ , Jose A. Olivas, Francisco P. Romero Department of Technologies and Information Systems, Escuela Superior de Informática, University of Castilla-La Mancha, Ciudad Real, Spain article info Article history: Received 17 May 2020 Received in revised form 4 September 2020 Accepted 29 September 2020 Available online 5 October 2020 Keywords: Sentiment analysis Recommender systems T1OWA aggregation operator Fuzzy linguistic representation abstract Online reviews have a significant impact on the decisions of consumers, providing valuable information which must be managed from two different perspectives: that of the user who reads the review and the people who gave those opinions. These two perspectives are the basis of the novel fuzzy aspect- based sentiment analysis approach described in this paper to recommend the most suitable products for a specific user. This approach consists of a T1OWA-based mechanism to characterize the user profile, which is able to model whether the user can be more influenced by negative opinions or positive opinions, a mechanism for determining their preferences, and a variation coefficient method for weighting the importance of the aspects of the product reviews. Combining these ideas, our model outperforms other well-known methods for ranking products, while also having the advantage of being adaptable to the preferences and characteristics of a specific user. © 2020 Elsevier B.V. All rights reserved. 1. Introduction With the rapid development of eCommerce services, an in- creasing number of people prefer to purchase products online and exchange opinions about them on the Internet. Technically, the expression ‘electronic word-of-mouth’ (eWOM), has been coined to refer to these opinions or reviews, which are becoming a major source of information for potential buyers on the Web. Studies have revealed that reviews of products have important effects on consumers’ decision-making [1,2]. These effects are sometimes positive, and sometimes negative, for instance, be- cause of noise or the misleading information available, an exam- ple of which might be the detection of inconsistencies [3]. For that reason, new applications based on Opinion Mining are currently arising to exploit the information contained in opinions, such as opinion filters, product recommenders, opinion summarizers, among many others. As stated in the studies previously mentioned, it is important to pay attention to the content of opinions, but it is also necessary to understand the user who is reading those opinions. Each user may behave in a different way as a consumer, but it is known that there are common attitudes shared by most consumers. The prospective theory states, for example, that negative opinions have a higher impact on user decisions than positive or that user perception of ratings like ‘‘good’’ and ‘‘very good’’ may be ‘‘interchangeable’’, that is, they are not ‘‘very’’ different [4]. ∗ Corresponding author. E-mail addresses: jesus.serrano@uclm.es (J. Serrano-Guerrero), joseangel.olivas@uclm.es (J.A. Olivas), franciscop.romero@uclm.es (F.P. Romero). For example, analyzing opinions from well-known online plat- forms like TripAdvisor, it is possible to observe there are users who rate an aspect from a hotel by using 5 stars, but when reading their textual comments, the word used to describe the service is ‘‘good’’, several examples about it are described in [5]. Hence, when ranking products, it is important to consider the user who is receiving the final list of recommended prod- ucts. It is thus necessary to model those aspects related to the consumer, but also to model those features related to the users who are giving their opinions on a product, as they influence the consumer. There are many approaches to ranking products based on Fuzzy Logic theory, as it provides a flexible mechanism to rep- resent information about the polarity of opinions, for example, linguistic distributions, as used in [6] or intuitionistic fuzzy num- bers modeled by calculating how positive, neutral or negative the opinions are, in [7]. Nevertheless, most of these studies try to model their solutions based on Multicriteria Decision Making techniques, giving solutions for most problems and treating the opinion holders as experts who must reach a level of consensus degree for a global ranking, as studied in other works [8–10]. However, most of these approaches do not take advantage of Fuzzy Logic to model characteristics related to users and opinion holders. All previous approaches model the aspects of a product as criteria and use Fuzzy Logic representations to assess the degree of usefulness of each criterion according to a set of ‘‘experts’’ or a collection of reviews collected from the Internet. Our proposal, in contrast, first analyzes the opinions according to their con- tent, it then analyzes the user in terms of what he/she tends to https://doi.org/10.1016/j.asoc.2020.106768 1568-4946/© 2020 Elsevier B.V. All rights reserved.