Publicação
Accounting for formative and reflective topics in product review data for better consumer insights
| Resumo: | Observations of product and service reviews suggest that textual product reviews may contain statements about the overall experience (“We had a great time”) or, similarly, about whether to recommend a particular product. The authors argue that such statements encapsulate an overall assessment and hence are not independently informative about, but rather reflect, overall ratings. The authors propose a model that allows for the distinction between topics that contribute to and topics that merely reflect an overall evaluation and apply the model to a dataset consisting of luxury hotel reviews. The findings show that, compared with a standard supervised latent Dirichlet allocation, the proposed model better fits the data and improves customer insights by resulting in more semantically coherent topics that point at specific aspects with positive and negative relationships to customers’ evaluation of their experience. |
|---|---|
| Autores principais: | Bueschken, Joachim |
| Outros Autores: | Otter, Thomas; Allenby, Greg M. |
| Assunto: | Bayesian modeling Rating analysis Statistical text analysis Structural modeling Supervised LDA |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | artigo |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Observations of product and service reviews suggest that textual product reviews may contain statements about the overall experience (“We had a great time”) or, similarly, about whether to recommend a particular product. The authors argue that such statements encapsulate an overall assessment and hence are not independently informative about, but rather reflect, overall ratings. The authors propose a model that allows for the distinction between topics that contribute to and topics that merely reflect an overall evaluation and apply the model to a dataset consisting of luxury hotel reviews. The findings show that, compared with a standard supervised latent Dirichlet allocation, the proposed model better fits the data and improves customer insights by resulting in more semantically coherent topics that point at specific aspects with positive and negative relationships to customers’ evaluation of their experience. |
|---|