Extracting sections from document based on list of keywords - Python

Question

I am new to NLP and I would like to ask how can I extract sentences from the text based on keywords that I have using Python. I created a list of keywords which will be used to extract sentences from the document.

If this will be a simple tokenization problem in which you will loop the list through the tokens, how can I capture synonyms or related words?

For example:

Keyword: Internal business

Sentence: You can only use this software for your business only.


Keyword: Confidentiality

Sentence: Information will be kept as secure as possible.

I actually implemented text categorization using TF-IDF, but with small dataset and large number of keywords. I don't think this will work to. Thanks in advance.

Is it possible to apply pre-trained models like word2vec?

Is it also possible to create a custom model that will fit my concerns?

Answer 1

The ideal way to get the related sentences would be to try to get a sentence vector for the sentences you want to categorise and then compare the vectors of your predefined keywords with the obtained sentence vectors . You can get the sentence vectors by just averaging the word vectors of the words present in the sentences . Once the sentence vectors are obtained , you can use cosine similarity to compare the keyword vectors and the sentence vectors . The one with the max cosine similarity will give you the result .

Answer 2

One option is Word Mover’s Distance (WMD), an algorithm for finding the distance between pairs of strings. It is based on word embeddings (e.g., word2vec) which encode the semantic meaning of words into dense vectors.

The WMD distance measures the dissimilarity between two text documents as the minimum amount of distance that the embedded words of one document need to "travel" to reach the embedded words of another document.

For example:

Source: "From Word Embeddings To Document Distances" Paper

In your case, you would take your keyword(s) and compare it to each sentence. If the distance is below a threshold the keyword(s) are related to the sentence.