A decision-support model through online reviews: Consumer preference analysis and product ranking

Online reviews have become an important reference for consumers when choosing and purchasing products (or services); however, previous research based on these reviews has overlooked consumer preferences at the attribute level, as well as product ranking at the regional level. Here, we propose a decision-support model through online reviews involving 4 modules: (i) data collection and preprocessing, (ii) transformation from star ratings into probabilistic linguistic term sets (PLTSs), (iii) consumer preference analysis, and (iv) product ranking. After data collection and preprocessing, star ratings are represented as probabilistic linguistic term sets. Then, the most satisfactory product attribute and the least satisfactory product attribute are extracted from text comments using the well-known term-frequency-inverse document frequency (TF-IDF) algorithm. Thereafter, a data-driven best-worst method, in which preferences are given by objectively quantifying multiple product attributes based on star ratings and the weightage score obtained from the TF-IDF results, is established to analyze consumer preferences. Furthermore, product ranking is investigated from two levels: a region-based ranking method to discuss product ranking over different regions and a ranking position aggregation method to discuss product overall ranking based on a 0-1 integer programming model. We verify the decision-support model with a dataset of 10987 reviews of 20 automobile models from the Autohome platform. Some comparisons, theoretical and practical implications are provided.