top of page

Insights into Extractive AI

1 January 2024
Anat Bielsky

Extractive AI is a specific category of artificial intelligence, designed to extract and summarize information from vast databases. In these AI models, the primary goal is to extract information from textual sources and present it concisely and efficiently.


Various methods are utilized by Extractive AI models, including:

  • Employing Natural Language Processing (NLP) for text comprehension

  • Identifying entities such as people, places, and events

  • Discovering keywords and key topics

  • Recognizing relationships between ideas within a text

  • Assessing the importance of sentences and selecting the most relevant ones for the abstract.


Examples of potential applications for extractive AI models include:

  • Text Summary: Automatically generating summaries for lengthy documents, such as articles, reports, or books. The abstract encapsulates the primary points and ideas derived from the original text.

  • Answering Questions: Extractive AI models can "read" extensive texts and locate answers to specific questions. This functionality proves valuable in Q&A-type systems.

  • Machine Translation: In the translation process, Extractive AI technologies employ translation assembly using existing words and linguistic structures. Additionally, they facilitate the conversion of speech into written text.

  • Text Mining: Extraction of specific information from extensive documents, including names of individuals, locations, unique terms, and more. This is particularly useful for constructing databases.

  • Call Analytics: Understanding text conversations in platforms like chats, WhatsApp, etc., for customer satisfaction analysis.

  • Medical Diagnosis: Analysis and generation of information based on medical reports.


While Extractive AI models leverage existing information and do not generate new content, they hold significant value in processing and summarizing content. However, these models have limitations, including:

  1. Difficulty in recognizing the most crucial sentences or phrases in a text, particularly in long and complex texts or those containing descriptive language and intricate ideas.

  2. Risk of producing syntactically incoherent or inaccurate abstractions.

  3. Potential bias or inaccuracy in the abstracts might be generated if the original text is biased or unbalanced.


Despite its advanced capabilities in summarizing texts, Extractive AI struggles with understanding complex, lengthy, or biased texts. Consequently, the produced abstracts may not always be accurate or clear enough.


In contrast to Extractive AI, Generative AI models, which serve a different purpose, are capable of creating entirely original content, including articles, poems, stories, or images. Unlike Extractive AI models, condensation in Generative AI models doesn't rely on complete ideas from existing texts, but rather on the generation of new content.


In conclusion, Extractive AI stands as a crucial technology with significant potential for efficiently extracting, summarizing, and organizing data and texts on a large scale. However, attention must be given to its limitations, such as the partial understanding of natural language, as of 2023, as described above.


Sources:

Where to Use Generative AI for Enterprise vs. Extractive AI - Aiimi

Combining Extractive and Generative AI for New Possibilities

Extractive and Abstractive Text Summarization NLP - INK


A hand holding a glowing light
bottom of page