Intelligent Testing: A Scenario‑Driven Approach Across the Five Testing Phases

In the previous article "Exploring Baidu Intelligent Testing's Three Stages," the three stages of intelligent testing were introduced; this piece proposes a scenario‑driven, phased approach to systematically scale intelligent testing across all stages.

Testing activities can be divided into five steps: test input, test execution, test analysis, test localization, and test evaluation. Each step has distinct goals, so applying intelligence uniformly to the whole process can cause confusion and hinder implementation.

Scenario‑driven approach: Treat a specific test activity within a service as a pilot scenario, then expand from point to line, plane, and volume, as illustrated in the accompanying diagram.

01. Intelligent Exploration in the Test Input Phase The goal of test input is to identify comprehensive, accurate test actions, data sets, and realistic environments to maximize coverage and code reach. Traditional input relies heavily on experience and historical cases. AI‑enabled techniques such as automated anomaly case generation, massive query mining, page‑traversal recommendation, and fuzzing of function or API parameters improve recall and filter out irrelevant or duplicate features, thereby enhancing coverage while maintaining execution efficiency.

02. Intelligent Exploration in the Test Execution Phase Test execution aims to run the selected test set with minimal cost while preserving defect‑detection capability. Conventional practice executes all test cases without optimization, leading to redundancy and high resource consumption. Intelligent testing selects, deduplicates, balances, and schedules test cases, employing static and dynamic assessments, mutation testing, and flaky detection to prune, skip, reorder, or combine tests, dramatically improving execution speed and reducing cost.

03. Intelligent Exploration in the Test Analysis Phase Analysis interprets execution results to determine whether defects exist, essentially a quality‑assessment problem. Traditional analysis depends on expert‑defined thresholds and metrics, which can be brittle. Intelligent analysis leverages historical execution data, statistical smoothing, and machine‑learning models to automatically define reasonable ranges for data size, row count, business metrics, and visual UI checks, as well as applying DTW curve fitting for memory‑leak detection.

04. Intelligent Exploration in the Test Localization Phase Localization quickly identifies the root cause of test failures to enable rapid remediation. Conventional debugging involves manual inspection of tools, environments, and code. The proposed intelligent approach uses decision‑tree models to automatically rebuild or self‑heal tool failures, and employs change‑wall and knowledge‑graph techniques in monitoring to pinpoint affected components, assisting developers and operations in swift damage control.

05. Intelligent Exploration in the Test Evaluation Phase Evaluation assesses overall system risk based on test execution and system changes. By building a quality‑risk model from full‑process test data—selecting risk factors, collecting features, and training machine‑learning models—risk can be quantified objectively. The model provides real‑time risk scores for new projects, and continuous feedback from deployed projects refines the model, creating a sustainable risk‑assessment loop.

The article concludes with a recruitment notice for Baidu MEG Quality Efficiency Platform, seeking test developers, Java/C++/mobile engineers, and machine‑learning/data‑mining/NLP specialists.