216 / 2024-04-23 14:10:37

Extracting and categorizing customer requirements from informative online reviews using BERT and Importance-Performance Analysis

Informative reviews; LDA; Word2vec; BERT; Fine-grained sentiment analysis; Customer satisfaction; Importance-performance analysis (IPA); IPA-GAP1; IPCA;

The rapid advancement of internet technology has driven the growth of e-commerce, leading to a surge in online shopping and the sharing of customer experiences through online reviews. Online reviews provide valuable insights into customer requirements(CRs), customer preferences, and product improvement strategies. However, these reviews are often unstructured and the majority of them lack informative content for CRs, posing challenges for companies to explore and analyze. Addressing this need, this paper presented a novel framework designed to extract and categorize product attributes based on attribute-based sentiment orientation and customer satisfaction. To achieve this, it utilized advanced algorithms-Bidirectional Encoder Representations from Transformers(BERT) and its synergies BERT-CNN and BERT-RNN to filter out uninformative reviews, consequently eliminating noise and enhancing the overall efficiency of the methodology. The integration of Latent Dirichlet Allocation (LDA) and Word2Vec is utilized to identify product features by considering the semantic relationships within reviews and filtering out noise features related to extraneous factors such as logistics, customer service, and marketing strategies. Furthermore, it leveraged three advanced deep-learning models for both coarse and fine-grained sentiment analysis of reviews and attributes. Subsequently, attribute importance and performance are evaluated using random forest and frequency-based rules, followed by the application of Importance-Performance Analysis (IPA) and IPA-GAP1 for attribute categorization. Additionally, Importance Performance Competitor Analysis (IPCA) is conducted to assess attributes across different products. The proposed methodology is validated through an empirical study using phone reviews and an extended study involving iPad reviews from Jingdong.com. Comparative analysis with existing studies reinforces the effectiveness of the methodology.