TP-GCN

Forum-java is a log dataset that we collected in an open source java-based web forum system {https://github.com/Qbian61/forum-java.}. It is a Java-based forum platform developed by a technology company and widely used for social media and programming technique sharing it contains abundant and diverse functions, like posting articles, creating FAQs, etc., which can satisfy most of the requirements of users. In order to generate negative samples, we imitate the real bug situation and inject some faults from real industrial cases into the normal system, namely call change, condition change, argument change, and chain change.

This dataset contains tens of millions of log data collected by the web servers during system operation, and every log records a series of dynamic system behaviors generated during the operation of the forum system. This dataset contains 172,443 dynamic session networks based on trace IDs, which contain 56,080 abnormal negative samples, with an abnormal proportion of 32.5%. Each node in the network represents a log event and contains features about call status, duration, and exceptions. Each edge denotes that the event occurs after the previous event.