JARAD: An Approach for Java API Mention Recognition and Disambiguation in Stack Overflow

Invoking APIs is a common way to improve the efficiency of software development. Developers often discuss various problems encountered or share the experience of using the API in communities, like Stack Overflow and GitHub. To avoid the duplicate discussion of issues and support downstream tasks such as API recommendation and API Mining, it is necessary to recognize APIs mentioned in these communities and link them to the fully qualified name. This work is often referred to as the task of API mention recognition and disambiguation in informal texts, which is the main focus of our paper. We start from Java posts in Stack Overflow and analyze the proportion of the posts that involve discussion on API (API Post for short), with short names or fully qualified names, and the characteristics of API Post. We also extract the APIs associated with more than 30,000 posts in Stack Overflow, and automatically establish $$$$ pairs to construct the dataset JAPD. Finally, we propose a novel approach JARAD to infer the associated APIs in a post. In our approach, we first use BiLSTM and CRF to fuse context information in text and code snippets to obtain a set of associated API candidates. The candidate API is then scored by the frequency of the API type appearing in the post to infer API's fully qualified name. Our evaluation experiments demonstrate that JARAD achieves 71.58%, 76.84% and 74.12% on Precision, Recall and F1 respectively.