A survey on automatic detection of hate speech in text
0 likes103 views
The document discusses the automatic detection of hate speech, outlining its significance, definitions, and previous research efforts. It presents a systematic literature review detailing methodologies, datasets, and performance metrics in hate speech detection. The study identifies challenges in the field, including the lack of expertise and evolving language, while highlighting opportunities for improvement through open-source platforms and comparative studies.
A survey on automatic detection of hate speech in text
- 1. 자연어처리 연구실 M2020064 조단비 Published: ACM Journals; ACM Computing Surveys, Vol.51, No.4, 2018
- 2. Content 1. Why study Hate Speech automatic detection? 2. What is Hate Speech? 3. What has been done so far in automatic Hate Speech detection? 4. Resources for Gate Speech classification 5. Research challenges and opportunities #Kookmin_University #Natural_Language_Processing_lab. 1
- 3. Introduction > Describe the motivation for conducting research - “how hate speech online has been evolving” - “who are the main targets of it” > Provide the detailed definitions > Analyze the previous survey with systematic literature review - focusing on descriptive statistics about Hate Speech detection - focusing on algorithms for Hate Speech detection #Kookmin_University #Natural_Language_Processing_lab. 2
- 4. 1. Why study Hate Speech automatic detection? - European Union Commission directives - Automatic techniques not available - Lack of data about hate speech - Hate speech removal - Quality of service #Kookmin_University #Natural_Language_Processing_lab. 3
- 5. 2. What is Hate Speech? > Definition from several sources > Our definition of Hate Speech : “jokes also must be marked as hate speech.” #Kookmin_University #Natural_Language_Processing_lab. 4 1 2 3 4
- 6. 2. What is Hate Speech? > Particular cases and examples of Hate speech - In Facebook, hate speech = a verbal attack + the target of the attack from “protected category” #Kookmin_University #Natural_Language_Processing_lab. 5
- 7. 2. What is Hate Speech? > Hate Speech and other related concepts - Hate: 증오 - Cyberbullying: 사이버 괴롭힘 - Discrimination: 차별 - Flaming: 모욕 - Abusive language: 욕설 - Profanity: 욕설 - Toxic language or comment: 악성 댓글 - Extremism: 극단주의 (폭력 조장) - Radicalization: 급진주의 #Kookmin_University #Natural_Language_Processing_lab. 6
- 8. 3. What has been done so far in automatic Hate Speech detection? 1) Systematic Literature Review > Method description > Document Collection and Annotation - A total of 127 documents (2016.09.01 ~ 2017.05.18) - “Law and Social Sciences”: 76 / “Computer Science and Engineering”: 51 - Low number of citations #Kookmin_University #Natural_Language_Processing_lab. 7
- 9. 3. What has been done so far in automatic Hate Speech detection? > Keywords in the Document - Related concepts (cyberbullying, cyber hate, sectarianism, …) - Machine learning (classification, sentiment analysis, filtering systems, …) - Social media (internet, social media, social network, …) #Kookmin_University #Natural_Language_Processing_lab. 8
- 10. 3. What has been done so far in automatic Hate Speech detection? > Social Networks & Number of Used Instances #Kookmin_University #Natural_Language_Processing_lab. 9
- 11. 3. What has been done so far in automatic Hate Speech detection? > General or Particular Hate Speech & Algorithms Used #Kookmin_University #Natural_Language_Processing_lab. 10
- 12. 3. What has been done so far in automatic Hate Speech detection? > Type of Approach in the Document #Kookmin_University #Natural_Language_Processing_lab. 11 9 1 17
- 13. 3. What has been done so far in automatic Hate Speech detection? 2) Documents focusing on descriptive statistics about Hate Speech detection - There are descriptive articles about Racism(인종차별), Sexism(성차별), Prejudice toward refugees(난민에 대한 편견), Homophobia(동성애 혐오증), and general hate speech(일반적인 증오심) 3) Documents focusing on algorithms for Hate Speech detection > Dataset used in the papers > Achieved performances - metrics: Precision, Recall, F-measure, accuracy, and AUC #Kookmin_University #Natural_Language_Processing_lab. 12
- 14. 3. What has been done so far in automatic Hate Speech detection? #Kookmin_University #Natural_Language_Processing_lab. 13
- 15. 3. What has been done so far in automatic Hate Speech detection? 4) Text mining approaches in automatic Hate Speech detection : feature extraction (1). general features used in text mining : dictionary, distance metric, Bag-of-words, N-grams, TF-IDF, Part-of-speech, … (2). The specific hate speech detection features #Kookmin_University #Natural_Language_Processing_lab. 14
- 16. 4. Resources for Hate Speech classification 1). Dataset & open source projects #Kookmin_University #Natural_Language_Processing_lab. 15
- 17. 4. Resources for Hate Speech classification https://paperswithcode.com/datasets? task=hate-speech-detection #Kookmin_University #Natural_Language_Processing_lab. 16
- 18. 4. Resources for Hate Speech classification https://github.com/kocohub/ korean-hate-speech #Kookmin_University #Natural_Language_Processing_lab. 17
- 19. 5. Research challenges and opportunities > challenges - Lack of expertise - Difficulty to track all racial and minority insults - Evolution of language among young population - Transition of hate speech such as sarcasm > Opportunity - Open source platforms or algorithms - Definition of a main dataset - Comparative studies - Multilingual research #Kookmin_University #Natural_Language_Processing_lab. 18