The rapid, increasing adoption of businesses to deliver their services in Software as a Service (SaaS) products in the marketplace presents selection challenges to users. Recently, major cloud service providers such as Amazon Web Services and Microsoft Azure have introduced well-architected frameworks that assess SaaS products in different pillars (also referred to herein as features). Furthermore, customers leave feedback on these features after using SaaS products. However, they do not provide feedback on all the features of a product, which renders the reviews unusable to prospective users needing to assess a product’s quality before committing to it. Our study addresses this drawback by imputing or inferring the intensity of the customer’s feedback on features that they do not mention in their reviews. Specifically, we propose threshold-based nearest neighbors (T-NN) as an extension of the conventional k-nearest neighbor approach to determine the missing sentiment intensity score of a feature from the values of its other features. We evaluate the proposed approach in two different systems and compare our results with seven other data imputation techniques. The results show that the proposed T-NN approach performs better than the other imputation approaches on the SaaS sentiment dataset.

企业在市场上以软件即服务 (SaaS) 产品形式提供服务的快速、日益普及给用户带来了选择挑战。最近，亚马逊网络服务和微软 Azure 等主要云服务提供商推出了架构良好的框架，用于评估不同支柱（此处也称为功能）的 SaaS 产品。此外，客户在使用 SaaS 产品后会留下对这些功能的反馈。但是，它们不会提供有关产品所有功能的反馈，这使得需要在承诺使用产品之前评估产品质量的潜在用户无法使用评论。我们的研究通过估算或推断客户对他们在评论中没有提到的功能的反馈强度来解决这个缺点。具体来说，k-最近邻方法，用于根据其他特征的值确定特征的缺失情感强度分数。我们在两个不同的系统中评估了所提出的方法，并将我们的结果与其他七种数据插补技术进行了比较。结果表明，所提出的 T-NN 方法在 SaaS 情感数据集上的性能优于其他插补方法。