Reviews serve as critical sources of information for potential buyers and valuable feedback for reviewed entities. Recognizing their pivotal role, companies increasingly appreciate the efficacy of helpful reviews as a marketing tool. Extensive interdisciplinary research, ranging from philosophy to artificial intelligence, has investigated factors influencing the prediction of helpful reviews.
This study focuses on the creation of a valuable resource for new users by effectively analyzing product reviews and highlighting the strengths and weaknesses associated with different aspects of a product. To achieve this, we employ a multi-step approach. Firstly, we adopt a review segregation technique, dividing each review into discrete chunks, typically clauses, wherein each chunk represents a specific aspect of the product. Next, we apply a modified version of the Textrank algorithm to assign a numerical rank to each chunk, reflecting its relative importance within the overall review. We further employ topic modeling using BERT-based transformers to group chunks into aspects. By merging the chunks back into reviews, we ascertain the aspects covered in each review. Subsequently, we analyze the attacking relationships between reviews pertaining to specific aspects of the product. Leveraging argument mining approaches, we identify more reliable reviews based on the attack network of reviews, enabling us to discern the strengths and weaknesses of the product across different aspects through these reliable reviews.
To evaluate the proposed approach, we have developed a series of GUI software tools within the scientific workflow platform, Orange3, and applied them to the Amazon Product Review dataset. The implementation in Orange3 offers intuitive graphical interfaces, tunable components, and visualization tools, enhancing user understanding of the underlying mechanisms and the significance of the output. The proposed framework contributes to enhancing the understanding of product interactions in reviews, aiding both consumers and businesses in making informed decisions.
Reviews serve as critical sources of information for potential buyers and valuable feedback for reviewed entities. Recognizing their pivotal role, companies increasingly appreciate the efficacy of helpful reviews as a marketing tool. Extensive interdisciplinary research, ranging from philosophy to artificial intelligence, has investigated factors influencing the prediction of helpful reviews.
This study focuses on the creation of a valuable resource for new users by effectively analyzing product reviews and highlighting the strengths and weaknesses associated with different aspects of a product. To achieve this, we employ a multi-step approach. Firstly, we adopt a review segregation technique, dividing each review into discrete chunks, typically clauses, wherein each chunk represents a specific aspect of the product. Next, we apply a modified version of the Textrank algorithm to assign a numerical rank to each chunk, reflecting its relative importance within the overall review. We further employ topic modeling using BERT-based transformers to group chunks into aspects. By merging the chunks back into reviews, we ascertain the aspects covered in each review. Subsequently, we analyze the attacking relationships between reviews pertaining to specific aspects of the product. Leveraging argument mining approaches, we identify more reliable reviews based on the attack network of reviews, enabling us to discern the strengths and weaknesses of the product across different aspects through these reliable reviews.
To evaluate the proposed approach, we have developed a series of GUI software tools within the scientific workflow platform, Orange3, and applied them to the Amazon Product Review dataset. The implementation in Orange3 offers intuitive graphical interfaces, tunable components, and visualization tools, enhancing user understanding of the underlying mechanisms and the significance of the output. The proposed framework contributes to enhancing the understanding of product interactions in reviews, aiding both consumers and businesses in making informed decisions.