# Unsupervised Technical Domain Terms Extraction using Term Extractor

**Authors:** Suman Dowlagar, Radhika Mamidi  
**arXiv:** [2101.09015](https://arxiv.org/abs/2101.09015)  
**Date:** 2021-01-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2101.09015v1/

## Key points

- Terminology extraction is a subtask of information extraction that involves automatically extracting relevant words or phrases from a given corpus.
- The paper focuses on an unsupervised automated domain term extraction method called TermTraction for ICON 2020 shared task 2.
- Automatic Term Extraction (ATE) aims to extract terms such as words, phrases, or multi-word expressions from a corpus and is used in various natural language processing tasks.
- Unsupervised algorithms for domain term extraction do not rely on labeled training data or pre-defined rules or dictionaries. Instead, they utilize statistical information from the text.
- The algorithm involves several steps: simple rules using techniques like chunking or POS tagging, naive counting, preprocessing, candidate generation and scoring, and final set selection.
- The paper mentions the use of TF-IDF measure for term weighting in current approaches to domain term extraction.
- The proposed method aims to contribute to the field of terminology extraction by participating in the ICON 2020 shared task 2: TermTraction.

## Summary

Terminology extraction, also known as term extraction, is a subtask of information extraction that involves automatically extracting relevant words or phrases from a given corpus. This paper focuses on the unsupervised automated domain term extraction method for ICON 2020 shared task 2: TermTraction. The method considers chunking, preprocessing and ranking domain-specific terms using relevance and cohesion functions. The aim of Automatic Term Extraction (ATE) is to extract terms such as words, phrases or multi-word expressions from a corpus. ATE is widely used in various natural language processing tasks like machine translation, summarization, document clustering and information retrieval. Unsupervised algorithms for domain term extraction do not rely on labeled training data or pre-defined rules or dictionaries. Instead they utilize statistical information from the text. These algorithms typically involve several steps in their pipeline: simple rules using techniques like chunking or POS tagging to extract noun phrases for multi-word extraction; naive counting counting how many times each word occurs in the corpus; preprocessing removing punctuation and common words (stop words) from the text; candidate generation and scoring utilizing statistical measures and ranking algorithms to generate a set of potential domain terms; final set selection arranging the ranked terms based on scores and selecting the top N keywords as output. The paper also mentions the use of TF-IDF measure for term weighting in current approaches to domain term extraction. Overall this study presents an unsupervised automated approach for extracting technical domain terms using relevant techniques such as chunking, preprocessing and ranking based on relevance and cohesion functions. The proposed method aims to contribute to the field of terminology extraction by participating in the ICON 2020 shared task 2: TermTraction.