Incorporating LIWC in Neural Networks to Improve Human Trait and Behavior Analysis in Low Resource Scenarios

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4532–4539 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4532 Incorporating LIWC in Neural Networks to Improve Human Trait and Behavior Analysis in Low Resource Scenarios Isıl Do˘ga Yakut Kılıc¸ 1, Shimei Pan1 1University of Maryland, Baltimore County {yakut1, shimei}@umbc.edu Abstract Psycholinguistic knowledge resources have been widely used in constructing features for text-based human trait and behavior analysis. Recently, deep neural network (NN)-based text analysis methods have gained dominance due to their high prediction performance. However, NN-based methods may not perform well in low resource scenarios where the ground truth data is limited (e.g., only a few hundred labeled training instances are available). In this research, we investigate diverse methods to incorporate Linguistic Inquiry and Word Count (LIWC), a widely-used psycholinguistic lexicon, in NN models to improve human trait and behavior analysis in low resource scenarios. We evaluate the proposed methods in two tasks: predicting delay discounting and predicting drug use based on social media posts. The results demonstrate that our methods perform significantly better than baselines that use only LIWC or only NN-based feature learning methods. They also performed significantly better than published results on the same dataset. Keywords: Low-resource Learning Tasks, Domain Knowledge Incorporation, Deep Neural Networks 1. Introduction Psycholinguistics research has shown that language use provides clues into one’s emotional and cognitive states (Pennebaker et al., 2003; Pennebaker and King, 1999). Text-based human trait and behavior analysis aims at finding patterns in user-generated texts to help analyze the characteristics and behavior of the author such as their beliefs, attitudes, emotions, personality, and online/real world behaviors. Previously, psycholin- guistic lexicons such as Linguistic Inquiry and Word Count (LIWC) have been widely used in predicting hu- man traits and behavior from text (Riff et al., 2019; Robinson et al., 2013; Yarkoni, 2010). Recently, fu- eled by big data and big machines, deep neural network (NN)-based models have gained near dominance due to their high prediction performance (Majumder et al., 2017a; Ding et al., 2017). However, NN-based meth- ods may not work well when the training data is limited (e.g., with only a few hundred labeled training exam- ples). Such a low resource scenario is common in hu- man trait and behavior analysis because sophisticated psychometric evaluations are often required to obtain the ground truth (Zafarani and Liu, 2015). As a re- sult, it is expensive to acquire the grount truth at a large scale. When the ground truth data is limited, the power of deep learning can also be limited. Since high-level abstractions related to human emotions and cogni- tions may not be easily inferred from a small num- ber of examples, in such a low resource scenario, psy- cholinguistic knowledge may complement the infor- mation extracted automatically by NNs. In this re- search, we investigate diverse methods to incorporate Linguistic Inquiry and Word Count (LIWC), a widely used psycholinguistic lexicon, in neural network(NN)- Stage I: User representation learning Stage II: …

Content truncated to 500 words