An independent study trying to understand the relationship between scholarly articles and scholarly reference indicators.
In today’s age of social media, scholarly articles are shared on multiple social networks and are often presented with the opportunity to disseminate into these networks through the means of modern communication. These scholarly articles have the ability to impact the discourse on major social platforms with ease and the concept of limelight is quickly showered upon scholarly articles that have significant social media impact. Although it is easy to gain notoriety and obtain social media impact on a short term basis, the nature of these social networks makes it difficult for these scholarly articles to stay relevant and remain impactful over a longer period. In this paper we observed whether it is possible to predict long term social media impact of scholarly articles published years ago. We used different variants of recurrent neural network architectures to predict the long term social media impact. We followed both supervised and unsupervised approaches for training the models and features for the experiments were extracted partially from the scholarly articles and the remaining coming from altmetrics.
TBA
TBA
- Can we build features from scholarly texts to identify how does altmetrics for a scientific text change over time ?
- Can we build features from scholarly texts to identify how does social media impact for a scientific text change over time ?
- Can we identify what features are important for contributing the change in altmetrics and social media indicators over time ?
The second part of the research would also involve building machine learning models that use features from the scholarly texts for regression and neural network models in predicting social media metrics at different periods of time. The focus would be on the following:
- Building regression models that use features from scholarly papers to predict the social media count for a given scholarly paper at different time intervals.
- Evaluating the models by measuring mean squared error, r-squared value in order to identify which model performs the best when predicting the response.
- Using p-values to identify which coefficients are statistically significant in predicting different social media counts.