Convolutional neural encoding of online reviews for the identification of travel group type topics on TripAdvisor
Previous studies have concluded that there are significant differences in travelers’ preferences depending on the trip type. The problem of extracting users’ preferences from a corpus of text can be solved by using traditional clustering algorithms, which work quite well when there is no predefined...
| Authors: | , , |
|---|---|
| Format: | article |
| Status: | Versión aceptada para publicación |
| Publication Date: | 2021 |
| Country: | España |
| Institution: | Universidad de Sevilla (US) |
| Repository: | idUS. Depósito de Investigación de la Universidad de Sevilla |
| OAI Identifier: | oai:idus.us.es:11441/154169 |
| Online Access: | https://hdl.handle.net/11441/154169 https://doi.org/10.1016/j.ipm.2021.102645 |
| Access Level: | Open access |
| Keyword: | Electronic word-of-mouth Convolutional neural encoding Unique topics Customer preferences |
| Summary: | Previous studies have concluded that there are significant differences in travelers’ preferences depending on the trip type. The problem of extracting users’ preferences from a corpus of text can be solved by using traditional clustering algorithms, which work quite well when there is no predefined data structure. However, in this paper, we consider the problem of extracting users’ preferences when they belong to a finite number of classes represented by the trip type. In this paper, we propose an encoding method based on a Convolutional Neural Networks (CNNs), trained as a classifier for the classes that predefine data structure. The intuition behind convolutional neural encoding is its ability to maximize the distance between documents belonging to different classes in the new, derived feature space. Findings reveal that CNNs encoding has better discriminative properties than alternative encoding methods such as Latent Dirichlet Allocation or average word2vec encoding. Moreover, we demonstrate that CNNs encoding can be used to identify the unique topics associated with the predefined data structure determined, in this case, by the four trip types. |
|---|