Sentiment Analysis: Polarity Classification of TripAdvisor's Hotel Reviews

TripAdvisor is a very popular website, in which travellers can post their reviews of accommodations, restaurants and places and read other peers’ feedbacks. Therefore, it can be considered a type of social media that is sometimes referred to as eWOM: this acronym stands for electronic word-of-mouth and puts the emphasis on the fact that ideas are spread among peers using electronic devices (Pollach, 2006). Thus, as already pointed out in the first chapter (see sections 1.1 and 1.8), online reviews have a very powerful role, since they can reach a distant audience, influence other people’s choices and harm or help a facility’s reputation.

As a consequence, the power relationship between vendors and customers has changed (O’Connor, 2008). The characteristic that prompted this change, and that made this kind of websites so popular, is the fact that users are generally free to express their opinions without the influence of tourism stakeholders. Unlike the texts published on private websites that want to advertise a hotel or a location, travellers’ reviews are supposed to mirror reality and to be free from conscious manipulation (Seargeant, 2014). For this reason, up 2010, TripAdvisor’s slogan was “get the truth. then go.”, which was quite effective also thanks to the alliteration of the sounds “g” and “t”.

Nonetheless, this slogan was changed, due to the huge amount of unreliable reviews: not all users are honest travellers who want to share their authentic experiences and who want to help peers and good facilities; on the contrary, some of them publish fake feedbacks in order to promote something or to discredit something else (Seargeant, 2014). These activities are called opinion spamming and the sentiment analysis task that deals with them is called opinion spam detection (Liu, 2012). However, TripAdvisor has not yet developed effective techniques to detect fake reviews.

In spite of this unsolved issue, the reliance on travellers’ reviews continue to grow (Sharma et al. 2015; Werthner and Ricci, 2004 in Claster et al. 2013). Consequently, the language of tourism is also changing and is becoming more similar to the language of social media. While advertising websites use persuasive techniques and rhetorical devices to convince possible customers, reviews are written in a less embellished and more plain, colloquial style, apt for online cooperation. (Seargeant, 2014).
Moreover, TripAdvisor is becoming more and more similar to an online travel agency, since it does not provide only reviews, but also booking possibilities.

This is evident in Picture 23 (next page), which shows an example of hotel review on TripAdvisor with the booking possibilities put on the left of the review: in this example, the price for a room booked on TripAdvisor is €189.
As it can be seen in Picture 23, reviews published on TripAdvisor are structured in a precise way. On the left, there is some information about the reviewer, such as nickname, nationality and/or city of residence, number of reviews written and peers’ feedback, i.e. the number of people who have voted their reviews as helpful. On the right, then, there is the review itself, with its title (in the example “Dated hotel with poor service”), the numeric rating (a number in a range from 1 to 5) assigned to the accommodation facility reviewed (here it is 1), the date in which it has been written (“Reviewed 23 September 2007”), the text, the period in which the reviewer stayed in the facility (“Stayed September 2007”) and numeric values assigned to specific aspects of the facility (e.g “location”, “rooms”).