Publicação
Measuring user influence in financial microblogs: experiments using stocktwits data
| Resumo: | In this paper, we study the effect of graph structure user in- fluence measures in financial social media. In particular, we explore rich and recent data, composed of 1.2 million Stock- Twits messages, from June 2010 to March 2013. These data allow the creation of social network graphs by considering direct active interactions (retweets, shares or replies). Using such graphs and a realistic rolling windows evaluation, we analyzed four user influence measures (indegree, between- ness, page rank and posts) under two criteria: Percentage of Quality Users (PQU), as manually labeled by StockTwits; and the daily sentiment correlation between top lists of in- fluential users and other users. The sentiment was based on a StockTwits labeled dataset and assessed in terms of three selections: overall sentiment (ALL) and filtered by two ma- jor technological companies (Apple – AAPL and Google – GOOG). Promising results were obtained, with several top lists pre- senting PQU values higher than 80% and correlations higher than 0.6. Overall, the best results were achieved by the page rank and posts measures. |
|---|---|
| Autores principais: | Cortez, Paulo |
| Outros Autores: | Oliveira, Nuno Miguel Rocha; Ferreira, João Carlos Peixoto |
| Assunto: | Sentiment analysis Microblogging data Social networks User influence Stock markets |
| Ano: | 2016 |
| País: | Portugal |
| Tipo de documento: | comunicação em conferência |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | In this paper, we study the effect of graph structure user in- fluence measures in financial social media. In particular, we explore rich and recent data, composed of 1.2 million Stock- Twits messages, from June 2010 to March 2013. These data allow the creation of social network graphs by considering direct active interactions (retweets, shares or replies). Using such graphs and a realistic rolling windows evaluation, we analyzed four user influence measures (indegree, between- ness, page rank and posts) under two criteria: Percentage of Quality Users (PQU), as manually labeled by StockTwits; and the daily sentiment correlation between top lists of in- fluential users and other users. The sentiment was based on a StockTwits labeled dataset and assessed in terms of three selections: overall sentiment (ALL) and filtered by two ma- jor technological companies (Apple – AAPL and Google – GOOG). Promising results were obtained, with several top lists pre- senting PQU values higher than 80% and correlations higher than 0.6. Overall, the best results were achieved by the page rank and posts measures. |
|---|