SocialBERT -- Transformers for Online SocialNetwork Language Modelling

TL;DR

SocialBERT is a novel transformer-based model that incorporates social network information about authors to improve language understanding tasks, achieving up to 7.5% better probabilistic modeling for individual authors.

Contribution

This work introduces SocialBERT, the first model to embed social network data into BERT for enhanced social-aware language modeling.

Findings

Embedding social network info improves model accuracy by up to 7.5%.

SocialBERT generalizes well to unseen social groups.

Model and code are publicly available for practical use.

Abstract

The ubiquity of the contemporary language understanding tasks gives relevance to the development of generalized, yet highly efficient models that utilize all knowledge, provided by the data source. In this work, we present SocialBERT - the first model that uses knowledge about the author's position in the network during text analysis. We investigate possible models for learning social network information and successfully inject it into the baseline BERT model. The evaluation shows that embedding this information maintains a good generalization, with an increase in the quality of the probabilistic model for the given author up to 7.5%. The proposed model has been trained on the majority of groups for the chosen social network, and still able to work with previously unknown groups. The obtained model, as well as the code of our experiments, is available for download and use in applied tasks.