Personality Trait Identification Using the Russian Feature Extraction Toolkit

Proceedings of Recent Advances in Natural Language Processing, pages 583–592 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_066 583 Personality Trait Identification Using the Russian Feature Extraction Toolkit James R. Hull*, Valerie Novak*, C. Anton Rytting*, Paul Rodrigues†,*, Victor M. Frank*, and Matthew Swahn† *University of Maryland, College Park, MD, USA †Accenture, Washington, DC, USA {jhull1, vnovak, crytting, prr, vmfrank}@umd.edu, matthew.swahn@accenture.com Abstract Feature engineering is an important step in classical NLP pipelines, but machine learn- ing engineers may not be aware of the sig- nals to look for when processing foreign lan- guage text. The Russian Feature Extrac- tion Toolkit (RFET) is a collection of fea- ture extraction libraries bundled for ease of use by engineers who do not speak Rus- sian. RFET’s current feature set includes features applicable to social media genres of text and to computational social science tasks. We demonstrate the effectiveness of the tool by using it in a personality trait identification task. We compare the perfor- mance of Support Vector Machines (SVMs) trained with and without the features pro- vided by RFET; we also compare it to a SVM with neural embedding features gen- erated by Sentence-BERT. 1 Introduction Data scientists and computer scientists working on natural language processing (NLP) prob- lems are occasionally confronted with tasks in languages they do not know well. Technical linguistic issues that may be well known to researchers and language practitioners familiar with the language can be quite important to effective language engineering, and can take a lengthy career to master. The Russian Feature Extraction Toolkit (RFET) unites a suite of tools for Russian lan- guage processing that would allow a program- mer unfamiliar with the language the ability to get started quickly in common text classifica- tion tasks such as sociolinguistic factor classi- fication, sentiment classification, and emotion analysis. Additionally, this toolkit provides those advanced in Russian NLP convenient ac- cessibility of features and feature combinations to quickly iterate through multiple experiment scenarios. Deep learning approaches are showing great promise, and feature extraction is less com- monly utilized in these approaches. Still, there are existing systems in production which harness classical machine learning techniques. RFET, and tools like it for other languages, could improve performance of these systems without a complete redesign. Further, there are many languages in which deep learning ap- proaches still lack pre-trained models or even a sufficient quantity of text examples to create them. The methodology presented in this pa- per could be used to design Feature Extraction Toolkits for these languages, in these cases. This toolkit does not replace feature ex- traction functions in libraries like NLTK (Bird, 2006) or scikit-learn (Pedregosa et al., 2011). These systems can produce language- independent statistical measurements on text that could be used as features. It also does not seek to replace toolkits for basic NLP pipeline elements, such Natasha,1 though there is some overlap between the tools. This toolkit focuses on uniting Russian language resources that pro- vide linguistically-informed features that are distinctive of the Russian language or otherwise not …

Content truncated to 500 words