Requirement Mining in Software Product Forums
James Tizard · 2019
The majority of software projects fail, around 71% according to recent research. A shortage of user feedback and missed requirements are cited as primary reasons for failure. There are several prominent online platforms where software users post product feedback, including: app stores, Twitter, issue trackers and product forums. I have identified the study of product forums as a gap in the current requirement mining literature, and have selected them as the focus of this research. Product forums are widely used in the software industry, supporting online discussions between a products users and owners. While their primary function is to help customers use the product, forums are also a rich source of untapped user generated requirements. However, the manual effort to extract these requirements is prohibitively time consuming due to their large volume and inconsistent quality. Analysis tools to assist in requirement mining have been applied successfully to online platforms previously, but as of yet, not in the forum domain, where current techniques may be insufficient. My preliminary research has found that forums contain feedback useful for software maintenance and evolution, including several categories of feedback not identified in the current literature. I have developed forum specific classifiers to help categorise the different feedback in forum posts. I demonstrate that these classifiers significantly outperform a leading app store tool when applied to forums. In this report I present my preliminary findings, then outline my research plan with the final goal of producing an industry evaluated, forum analysis tool.