E-petition popularity: Do linguistic and semantic factors matter? 논문 정리

www.sciencedirect.com/science/article/abs/pii/S0740624X16301253

E-petition popularity: Do linguistic and semantic factors matter?

 

Reference

- Twitter studies: impact of textual patterns on retweets

- Communication studies of persuasion

 

Variables (0,1로 태깅된 것 제외하고 모두 log scale로 사용)

1. dependent variable

- total signature counts

 

2. control variables

- # of signatures gathered during first 24 hours

- # of petitions started on same day

 

3. predictor variables

3.1. linguistic style variables

- extremity

   - 특정 단어(e.g. much more) 나타나면 1, 아니면 0

- urgency

   - 특정 단어(e.g. immediately) 나타나면 1, 아니면 0

- informativeness

   - # of unique words가 평균 이상이면 1, 아니면 0

- repetition

   - # of total words / # of unique words가 평균이 이상이면 1, 아니면 0

- request

   - 특정 단어(e.g. spread, sign) 나타나면 1, 아니면 0

- sentiment

   - stanford sentiment analyzer 기준으로 neutral 아니면 1, 아니면 0

- internet activity

   - 인터넷 링크 첨부하면 1, 아니면 0

 

3.2. semantic variables

- NER variables

   - Stanford CoreNLP NER tagger:# of names of persons, locations, organizations

   - evaluation: F-measure

- topic variables

   - MALLET topic model package: 15 topics

   - evaluation: cross validation

 

 

Regression analyses

- hierarchical ordinary least squares(OLS)

- to determine the explanatory power of variables

 

 

 

 

Model 4: adjusted R-squared 32%

 

linguistic style variables

- Model 2: 모두 통계적으로 유의함

   - sentiment, urgency: positively correlated

   - internet activity, extremity, repetition: negatively correlated

- Model 3: sentiment만 통계적 유의성 잃음

- Model 4: extremity 빼고 모두 통계적 유의성 잃음

   - suggest linguistic variables are interrelated

 

semantic variables

- NER 중 person만 통계적으로 유의함

   - person: negatively correlated

- Topic

   - popular topics: positively correlated

   - unpopular topics: negatively correlated