Information filtering systems serve as robust tools in the ongoing difficulties associated with overwhelming volumes of data. With constant generation and accumulation of reviews in online communities, the ability to distill and provide valuable insights to assist customers in their search for relevant information is of considerable significance. This study devised an effective review filtering system for a popular online physical experience review site.
This study entailed an investigation of a hybrid approach for a review filtering system augmented with various text mining-based operational variables to extract the linguistic signals of online reviews. Moreover, we devised three ensemble models based on multiple machine learning and deep learning algorithms to build a high-performance review filtering system.
The main findings confirm the effectiveness of using the derived operational variables when reviewing filtering systems. We found that the reviewer’s tendency and history macros, as well as the readability and sentiment of the reviews, contribute significantly to the filtering performance. Furthermore, the proposed three ensemble frameworks demonstrated good efficiency with an average accuracy of 89.39%.
This study provides a methodological blueprint for operationalizing variables in online reviews, covering both structured and unstructured datasets. Incorporating different variables enhances the efficiency of the algorithm and provides a more comprehensive understanding of user-generated content. Furthermore, the study affords a strategic perspective and integrated guidelines for developers seeking to create advanced review filtering systems.
