No abstract

No abstract

No abstract

No abstract

Introduction: An important stage in a software development life cycle is the support phase, when customers can contact the support service of the supplier company and request a solution to an issue encountered in the software. To solve the request, engineers often have to refer to the relevant documentation. In order to reduce the complexity of the maintenance phase, the search for the necessary documentation pages can be automated. Purpose: Development of an approach to semantic search through documentation using Doc2Vec machine learning algorithm in order to automate the solution of customer requests. Results: An approach is proposed to semantic search through text documentation files and wiki pages using Doc2Vec machine learning algorithm. The documentation pages with semantic similarities to the textual description of an unresolved customer request help the engineer to process the request more efficiently and rapidly. Based on the proposed approach, a software tool has been developed which provides the engineer with a report containing links to documentation pages semantically related to the unresolved request. During the configuration of this tool, the optimal parameters of the Doc2Vec algorithm were found, providing the necessary quality of the semantic search. The idea of the experiment was to apply the tool to unresolved requests and evaluate its effectiveness. The developed approach and software tool were successfully tested in an open source Apache Kafka project. In the course of the experiment, 100 requests from Jira bug tracking system were downloaded and analyzed. The experimental results show the advantage of using the tool in software product support. The average documentation analysis time has been reduced as compared to the traditional manual approach. Practical relevance: The research results were used to solve real customer requests. The developed approach and the software implemented on its basis can reduce the complexity of the maintenance phase.