Anserini

Anserini is an open-source information retrieval toolkit built on top of the Apache Lucene search library. It is designed to support reproducible research by providing end-to-end, ready-to-run retrieval pipelines and strong baseline capabilities for academic experiments in document and passage retrieval.

The toolkit offers indexing, searching, and evaluation components that enable researchers to build and test lexical retrieval systems. It supports standard retrieval models such as BM25 and query likelihood with Dirichlet smoothing, along with various preprocessing, tokenization, and indexing options. Anserini also includes utilities for preparing data, running experiments, and computing common IR evaluation metrics like MAP, nDCG, and precision at depth.

Anserini is part of a broader ecosystem that emphasizes interoperability and ease of use. It provides a Java-based core and a set of command-line tools, as well as bindings and examples that facilitate integration into research workflows. In addition, it forms the basis for Pyserini, a Python interface that enables researchers to access Anserini’s capabilities from Python and to combine lexical retrieval with neural reranking and other modern techniques.

The project is widely adopted in academia for benchmarking and reproducibility, and it has been applied to a range of standard information retrieval datasets, including established benchmarks and large-scale collections. Anserini’s design focuses on transparency, accessibility, and repeatable experimentation, making it a common starting point for IR research and teaching.