Ranks Relevant Files in the Top 10 with 70% Accuracy
On average, CodeSearchFinder-AI successfully ranks one or more relevant code files in the top 10 with 70% accuracy.
Our AI-powered approach analyzes a given service request and corresponding source files by computing two key scores:
- Lexical Similarity Score – Measures textual similarity between the service request and code files.
- Probabilistic Score – Based on the Vector Space Model, determining relevance through statistical analysis.
Each score is derived from four different search types, each using a distinct set of indexed terms from both the service request and the code file.
To ensure the most accurate ranking, we:
- Generate eight different ranking combinations by applying these scoring methods.
- Rank all files in descending order for each combination.
- Select the best rank for each file across all eight rankings.
Additionally, CodeSearchFinder-AI evaluates each service request and code file individually by analyzing key elements such as:
- Summary and stack trace
- Stemming and linguistic variations
- Code comments and file names
When relevant and available, these factors are leveraged to further optimize ranking accuracy.