Context Window Extractor

easyPROQuestion AnsweringNLP 3: Advanced

Extract relevant context windows from a document for a given query.

Given a document (list of sentences) and a query, find sentences that contain at least one query word and return them with their indices.

Input: extract_context(query, sentences)

query: query string (space-separated words, lowercase)
sentences: list of N sentence strings (lowercase)

Output: A list of tuples (index, sentence) for sentences containing at least one query word. Matching is case-insensitive (both already lowercase).

Example

Input

machine learning
4
deep learning is a subset of machine learning
natural language processing uses rules
machine translation converts text between languages
learning algorithms improve with data

Output

[(0, 'deep learning is a subset of machine learning'), (2, 'machine translation converts text between languages'), (3, 'learning algorithms improve with data')]

Reasoning

Inputs: query is machine learning, so the query words are $Q = \{\text{machine}, \text{learning}\}$ . sentences holds 4 strings, numbered 0 to 3 by position.
Each sentence is broken into its whole words $S_i$ , and it is kept exactly when $Q \cap S_i \neq \varnothing$ , i.e. it shares at least one word with the query.
Sentence 0 contains both "learning" (twice) and "machine": kept, but only once. Sentence 1 ("natural language processing uses rules") shares nothing: dropped.
Sentence 2 contains "machine": kept. Notice that only exact whole words count: a sentence like "machines are relearning fast" would be dropped, since neither "machines" nor "relearning" equals a query word.
Sentence 3 contains "learning": kept.
Output: a list of (index, sentence) pairs in original order, so indices 0, 2, 3 appear with their full, unchanged text. If no sentence shares a word with the query, the result is an empty list [].

Constraints

Match whole words only (split on whitespace)
Both query and sentences are lowercase
Return matches in order of appearance
A sentence matches if it contains ANY query word

Learn this in

Retrieval-Augmented GenerationLLM Study Plan Question Answering & Information RetrievalNLP Study Plan

More from NLP 3: Advanced

Temperature Samplereasy ROUGE-1 Score Calculatoreasy Context Window Findereasy Keyword Answer Extractoreasy Levenshtein Distanceeasy

Concept primerPREVIEW

Background Knowledge

The "Context Window Extractor" problem falls under the category of Natural Language Processing (NLP), specifically within the topic of Question Answering. This involves analyzing and understanding human language to extract relevant information. A key concept here is tokenization, which is the process of breaking down text into individual words or tokens. In this problem, both the query and the document are already tokenized into lowercase words, simplifying the comparison process.

Understanding how to compare the query words with the sentences in the document is crucial. This comparison is based on string matching, where we look for occurrences of query words within each sentence. Since the matching is case-insensitive and both the query and sentences are in lowercase, we can directly compare the words without worrying about case differences. The concept of context window is also important, as it refers to the portion of the text (in this case, sentences) that is relevant to the query.

In NLP, information retrieval techniques are often used to find relevant documents or parts of documents based on a query. The "Context Window Extractor" problem is a simplified version of such techniques, focusing on extracting relevant sentences from a document. This requires understanding how to efficiently search through text data and how to determine the relevance of a sentence based on the presence of query words.

Algorithm/Approach

The general approach to solving this type of problem involves iterating through each sentence in the document and checking if any of the query words are present in the sentence. This can be achieved through a simple linear search algorithm, where each sentence is scanned for the occurrence of query words. The algorithm should keep track of the index of each sentence that contains at least one query word and return this information along with the sentence itself.

PRO

Continue the full explanation

You're reading the free preview. Unlock the complete walkthrough, the code editor, test runner and reference solution with Premium.

PRO

Editor locked

The code editor is locked for Pro problems. It is only available for free problems. Please upgrade to gain access to the code editor for all problems.

solution.py

Python 3.13

Test results