Skip to main content

Various BM25 algorithms for document ranking

Project description

Rank-BM25: A two line search engine

Build Status PyPI version

A collection of algorithms for querying a set of documents and returning the ones most relevant to the query. The most common use case for these algorithms is, as you might have guessed, to create search engines.

So far the algorithms that have been implemented are:

  • <input type="checkbox" checked="" disabled="" /> Okapi BM25
  • <input type="checkbox" checked="" disabled="" /> BM25L
  • <input type="checkbox" checked="" disabled="" /> BM25+
  • <input type="checkbox" disabled="" /> BM25-Adpt
  • <input type="checkbox" disabled="" /> BM25T

These algorithms were taken from this paper, which gives a nice overview of each method, and also benchmarks them against each other. A nice inclusion is that they compare different kinds of preprocessing like stemming vs no-stemming, stopword removal or not, etc. Great read if you're new to the topic.

Installation

The easiest way to install this package is through pip, using

pip install rank_bm25

If you want to be sure you're getting the newest version, you can install it directly from github with

pip install git+ssh://git@github.com/dorianbrown/rank_bm25.git

Usage

For this example we'll be using the BM25Okapi algorithm, but the others are used in pretty much the same way.

Initalizing

First thing to do is create an instance of the BM25 class, which reads in a corpus of text and does some indexing on it:

from rank_bm25 import BM25Okapi

corpus = [
    "Hello there good man!",
    "It is quite windy in London",
    "How is the weather today?"
]

tokenized_corpus = [doc.split(" ") for doc in corpus]

bm25 = BM25Okapi(tokenized_corpus)
# <rank_bm25.BM25Okapi at 0x1047881d0>

Note that this package doesn't do any text preprocessing. If you want to do things like lowercasing, stopword removal, stemming, etc, you need to do it yourself.

The only requirements is that the class receives a list of lists of strings, which are the document tokens.

Ranking of documents

Now that we've created our document indexes, we can give it queries and see which documents are the most relevant:

query = "windy London"
tokenized_query = query.split(" ")

doc_scores = bm25.get_scores(tokenized_query)
# array([0.        , 0.93729472, 0.        ])

Good to note that we also need to tokenize our query, and apply the same preprocessing steps we did to the documents in order to have an apples-to-apples comparison

Instead of getting the document scores, you can also just retrieve the best documents with

bm25.get_top_n(tokenized_query, corpus, n=1)
# ['It is quite windy in London']

And that's pretty much it!

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for rank-bm25, version 0.2.1
Filename, size File type Python version Upload date Hashes
Filename, size rank_bm25-0.2.1-py3-none-any.whl (8.5 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size rank_bm25-0.2.1.tar.gz (4.6 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page