NLP overview

＊사용된 모든 영문 image의 출처는 cs231n 강의 자료입니다.＊

 

<NLP overview>

    1. NLP Applications

    2. Academic Disciplines related to NLP

    3. Trends of NLP

---

NLP Applications

 

 

NLP(Natural Language Processing)

• 많은 양의 natural language datafmf computer program process에 적용시키는 방법
• NLP goal: Language의 contextual nuances를 포함하여 document의 contents를 computer가 'understanding'하게 하는 것

 

 

NLP  Applications

• Text Classification: Spam Detection (INBOX - SPAM FOLDER), Sentiment Analysis ···
• Question Answering: Search Engine (Question에 대한 이해) ···
• Machine Translation (Source language → Target language)
• Chatbot
• Personal Assistant (Sound → Natural Language → Command)
• Text Summarization (Input document → Abstractive summarization / Extractive Summarization)

---

Academic Disciplines related to NLP

 

 

Natural Language processing

• Major conferences: ACL, EMNLP, NAACL

- Low-level parsing: Tokenization, stemming

- Word and phrase level: Named Entity Recognition, Part-Of-Speech, Semantic relation extraction

- Sentence level: Sentiment analysis, Machine trainslation

- Multi-sentence and paragraph level: Entailment prediction, Question answering, Dialog systems (chatbot), Summarization

 

Text mining

• Major conferences: KDD, The WebConf, WSDM, CIKM, ICWSM

- Text와 document data에서 유용한 insight 추출

- Document clustering (e.g. topic modeling)

- Computational Social Science: sns에서 사람들의 사회, 과학적인 insight 도출

 

Information retrieval

• Major conferences: SIGIR, WSDM, CIKM, RecSys

- Recommendation system

---

Trends of NLP

 

 

Trends of NLP

• Text data는 sequence of words로
• 각 word는 embedding vector로 표현 (e.g. Word2Vec, Glove)
• Seqeuntial data를 RNN, LSTM, GRU 등의 RNN-based model로 분석
• Attention과 Transformer 모델 등장으로 RNN은 self-attention으로 대체
• 최근에는 Self-Supervised model (No label)을 사용한 BERT 등의 model이 large dataset을 fine-tuned하여 다양한 downstream task에서 사용