使用 Homebrew 安装 trec_eval

项目链接

• 主页
• 仓库

命令和别名

• trec_eval

项目历史

The Text Retrieval Conference began in 1992 as a NIST and U.S. Department of Defense effort to provide infrastructure for large-scale evaluation of text retrieval systems. Within that culture, trec_eval became the shared evaluator for ad hoc retrieval experiments, pairing submitted run files with standard judged results.

The official README describes trec_eval as the standard tool used by the TREC community for evaluating an ad hoc retrieval run. Its file layout reflects that role: parsers for qrels and TREC result formats, measure implementations, tests, and a single small C command-line program that researchers can build with make.

The code has evolved with the evaluation methodology. The 2008/2009 version 9 work was a large rewrite that separated individual measure calculations, made it easier to add measures and input formats, and added measures such as nDCG and preference-evaluation support. The 2026 version 10 changelog records a breaking scoring fix, added measures, stdin support for results files, Windows/GCC fixes, and continued maintenance.

采用历史

TREC itself grew from an annual research evaluation into a long-running benchmark ecosystem: NIST describes decades of datasets, measurement processes, and tools, and notes that the evaluation software is available so organizations can evaluate their own retrieval systems.

For package users, trec_eval matters because it is not just another metrics script; it is the reference implementation many IR papers, TREC tracks, and regression tests expect. Packaging it in Homebrew gives macOS researchers and search engineers the same executable name and behavior used in benchmark instructions.

使用方式

The typical workflow is to run `trec_eval` with official qrels and submitted results, optionally adding `-q` for per-query output, `-c` and `-M1000` for official-style handling, or `-m` to print a selected measure.

In package-manager culture, it is often installed for reproducible evaluation notebooks, search-ranking experiments, and old TREC run comparisons where the exact evaluator is part of the experiment.

为什么软件包爱好者会关心

trec_eval is a compact example of a domain-standard scientific CLI: tiny build surface, old C code, public benchmark authority, and enough historical behavior that package availability helps preserve reproducible research.

时间线

• 1992: TREC starts as a NIST/DOD large-scale text retrieval evaluation effort.
• 2006: trec_eval adds infAP and fixes bpref-related relevance-level behavior.
• 2008: version 9.0 alpha rewrites the evaluator around separate measure calculations and extensible formats.
• 2009: version 9.0 is released after the rewrite.
• 2020: GitHub workflow/CI and portability fixes appear in version 9.0.8.
• 2026: version 10.0 records scoring fixes, additional measures, stdin support, and compiler/platform fixes.

Related projects

• TREC qrels and run files are the main data formats that trec_eval consumes.
• The broader TREC ecosystem includes many NIST tracks and datasets for search, question answering, multimedia, and related retrieval evaluations.

来源

• https://github.com/usnistgov/trec_eval
• https://raw.githubusercontent.com/usnistgov/trec_eval/main/CHANGELOG
• https://raw.githubusercontent.com/usnistgov/trec_eval/main/README
• https://trec.nist.gov/
• https://trec.nist.gov/overview.html