[Paper Review] Natural Language Processing for Information Extraction

Sonit Singh|arXiv (Cornell University)|Jul 6, 2018

Topic ModelingComputer Science55 references48 citations

TL;DR

The paper surveys Information Extraction (IE) in NLP, detailing tasks like NER, NEL, CR, temporal/event extraction, and RE, along with state-of-the-art methods, tools, and KB reasoning approaches.

ABSTRACT

With rise of digital age, there is an explosion of information in the form of news, articles, social media, and so on. Much of this data lies in unstructured form and manually managing and effectively making use of it is tedious, boring and labor intensive. This explosion of information and need for more sophisticated and efficient information handling tools gives rise to Information Extraction(IE) and Information Retrieval(IR) technology. Information Extraction systems takes natural language text as input and produces structured information specified by certain criteria, that is relevant to a particular application. Various sub-tasks of IE such as Named Entity Recognition, Coreference Resolution, Named Entity Linking, Relation Extraction, Knowledge Base reasoning forms the building blocks of various high end Natural Language Processing (NLP) tasks such as Machine Translation, Question-Answering System, Natural Language Understanding, Text Summarization and Digital Assistants like Siri, Cortana and Google Now. This paper introduces Information Extraction technology, its various sub-tasks, highlights state-of-the-art research in various IE subtasks, current challenges and future research directions.

Motivation & Objective

• Introduce Information Extraction and its role in processing unstructured text.
• Review the sub-tasks of IE (NER, NEL, CR, Temporal/EVENT extraction, RE, KB reasoning) and their dependencies.
• Discuss state-of-the-art methods across IE tasks, including pattern-based, gazetteer-based, ML-based, and deep learning approaches.
• Highlight IE tools, shared tasks, and knowledge-base construction and reasoning as enabling technologies.

Proposed method

• Describe the Information Extraction pipeline and essential pre-processing steps (tokenization, stemming, POS tagging, parsing).
• Classify IE approaches into pattern matching, gazetteer-based, and machine learning-based methods, including CRFs, SVM, MaxEnt, and distant supervision.
• Present state-of-the-art methods for IE subtasks (NER, NEL, CR, temporal/event extraction, RE) and discuss joint modeling trends.
• Enumerate public, commercial, and specialized IE tools and their roles in real-world IE deployments.
• Explain Knowledge Base construction, reasoning, and link prediction as applications of IE in KBs like Freebase, DBpedia, YAGO, and Google Knowledge Graph.

Experimental results

Research questions

• RQ1 What are the main IE sub-tasks and how are they interrelated in NLP pipelines?
• RQ2 What are the prevailing state-of-the-art approaches for each IE sub-task (NER, NEL, CR, Temporal, RE) and their performance trends?
• RQ3 How do distant supervision and multi-instance learning mitigate labeled data challenges in relation extraction?
• RQ4 How can IE outputs be integrated with and enhanced by Knowledge Bases and reasoning for downstream tasks like QA and search?
• RQ5 What tools and shared tasks have driven progress in IE across domains and languages?

Key findings

• IE comprises multiple interrelated sub-tasks, including NER, NEL, CR, Temporal Information Extraction, RE, and KB reasoning.
• ML and deep learning methods have become dominant in many IE sub-tasks, often surpassing traditional rule- or pattern-based systems.
• Distant supervision and multi-instance learning address labeled data scarcity in RE but introduce noise and require careful modeling (e.g., MIML-RE).
• A wide ecosystem of IE tools exists (public, commercial, and specialized) enabling practical deployment and evaluation.
• Joint modeling across sub-tasks (e.g., CR-NER-NEL) and cross-domain/open multilingual approaches show promise for improving IE performance.