Fine-grained Prediction of Political Leaning on Social Media with Unsupervised Deep Learning

Authors: Tiziano Fagni, Stefano Cresci

JAIR 2022 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluated our technique in two challenging classification tasks and we compared it to baselines and other state-of-the-art approaches. Our technique obtains the best results among all unsupervised techniques, with micro F1 = 0.426 in the 8-class task and micro F1 = 0.772 in the 3-class task.
Researcher Affiliation Academia Tiziano Fagni EMAIL Stefano Cresci EMAIL Institute of Informatics and Telematics (IIT) National Research Council (CNR) via G. Moruzzi 1, 56124 Pisa, Italy
Pseudocode No The paper describes the methodology using high-level overview diagrams (Figure 1, Figure 3) and lists steps textually (e.g., Section 5, steps for clustering). It does not contain structured pseudocode or algorithm blocks with code-like formatting.
Open Source Code No The paper states that data is publicly available, but there is no explicit statement about open-source code for the methodology or a link to a code repository. 'Our data are publicly available for scientific purposes5. 5. https://doi.org/10.5281/zenodo.5793346'
Open Datasets Yes Our data are publicly available for scientific purposes5. 5. https://doi.org/10.5281/zenodo.5793346
Dataset Splits Yes Finally, we performed a stratified sampling to split our dataset into a training (90% 18,169 users), a validation (3% 604 users) and a test (7% 1,426 users) partition.
Hardware Specification No The paper does not provide any specific hardware details such as GPU/CPU models, memory, or processor types used for running the experiments.
Software Dependencies No The paper mentions 'sklearn Python software package' and 'gensim library', and 'UMAP with default parameters' but does not specify version numbers for these software components. '13. https://scikit-learn.org/stable/' '12. https://radimrehurek.com/gensim/'
Experiment Setup Yes In this work, we fixed k = 5 in Equation (3)... Th = 0.5 is a reasonable value... We leveraged UMAP with default parameters... we assume that we know the number of clusters we want to obtain at the end of clustering process (i.e., 8 clusters for the party prediction task and 3 clusters for pole prediction task)... Parties + clustering: ... step 2 with a feature reduction to 64 features, and step 4 using Gaussian Mixture with default parameters... Parties enriched + clustering: ... clustering process for the party prediction task is performed by applying only step 3 and step 4 using KMeans as the clustering algorithm. For the pole prediction task, we used instead step 1, step 2 with a feature reduction to 64 features, and step 4 using the KMeans algorithm.