Learning Product Attributes from User-Generated Content for Dynamic Promotion Strategies
MetadataShow full item record
One widely adopted product attribute classification in the literature is the “Search” versus “Experience” dichotomy. Because the costs involved in searching and experiencing products vary across consumers and over a product’s life time, it is important for marketers to understand consumers’ evaluation of these attributes in order to formulate scalable and dynamic promotion strategies. This thesis attempts to address this challenge by proposing a text analytics framework for understanding consumers’ evaluation of product attributes to support agile promotion strategies. In the past, researchers have attempted to classify entire product categories as search or experience via questionnaires or using quantitative approaches by analyzing review star ratings. This thesis uses objective consumer reviews and text mining techniques to extract product features that can define search or experience attributes. A hybrid of unsupervised and supervised learning techniques was used to generate labelled training data from eight different product categories of Amazon and train classification models to determine the likely position of a product within the search-experience product classification spectrum. Extensive experiments using best-case and worst-case scenario were used to improve the accuracy levels of decision-tree based classification models and demonstrate the scalability of the text analytics framework. The proposed approach also incorporated a mechanism to aggregate the scores that the model gives to each individual review in order to determine the likely position at a product level. It is also shown that a product’s position in the search-experience spectrum may change during its review cycle, indicating that marketers need to investigate reviews for any periods of interest to develop effective promotion strategies in a more agile fashion. From a theoretical view, the text mining approach significantly adds to the existing body of knowledge in the classification of product attributes for supporting promotions. In addition to detecting dominant signals for search and experience positions, marketers can uncover a great deal of contents to formulate more specific advertising messages.