Installation

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

pip install fastbm25

fastbm25

The fast bm25 algorithm for text match optimized by reverted index. So the complexity will be no more than O(N(log N)).

（利用倒排索引加速的bm25文本匹配算法，从一堆数据中寻找最相似的文本）

usage

find top k similar sentences from corpus; Note you should tokenize text and use stop words in advance

from fastbm25 import fastbm25

corpus = [
    "How are you !",
    "Hello Jack! Nice to meet you!",
    "I am from China, I like math."
]
tokenized_corpus = [doc.lower().split(" ") for doc in corpus]
model = fastbm25(tokenized_corpus)
query = "where are you from".lower().split()
result = model.top_k_sentence(query,k=1)
print(result)

The result is list of tuple like

[('I am from China, I like math.', 2, -0.06000000000000001)]

For some language like Chinese that doesn't need tokenization. you can use this example

from fastbm25 import fastbm25

corpus = [
    "张三考上了清华",
    "李四考上了北大",
    "我烤上了地瓜.",
    "我们都有光明的未来."
]
model = fastbm25(corpus)
query = "我考上了大学"
result = model.top_k_sentence(query,k=1)
print(result)

[('李四考上了北大', 1, 1.21)]

find document pair similarity between document a and document b

Note that a and b don't need to be included in the reference corpus;

from fastbm25 import fastbm25
corpus = [
    "How are you !",
    "Hello Jack! Nice to meet you!",
    "I am from China, I like math."
]
tokenized_corpus = [doc.lower().split(" ") for doc in corpus]
model = fastbm25(tokenized_corpus)
document_a = "where are you from".lower().split()
document_b = "where are you".lower().split()

result = model.similarity_bm25(document_a,document_b)
print(result)

1.944187075527278