Aspect-Based Sentiment Analysis for Open-Ended HR Survey Responses

Lois Rink,Job Meijdam,David Graus

2024-02-07

Abstract:Understanding preferences, opinions, and sentiment of the workforce is paramount for effective employee lifecycle management. Open-ended survey responses serve as a valuable source of information. This paper proposes a machine learning approach for aspect-based sentiment analysis (ABSA) of Dutch open-ended responses in employee satisfaction surveys. Our approach aims to overcome the inherent noise and variability in these responses, enabling a comprehensive analysis of sentiments that can support employee lifecycle management. Through response clustering we identify six key aspects (salary, schedule, contact, communication, personal attention, agreements), which we validate by domain experts. We compile a dataset of 1,458 Dutch survey responses, revealing label imbalance in aspects and sentiments. We propose few-shot approaches for ABSA based on Dutch BERT models, and compare them against bag-of-words and zero-shot baselines. Our work significantly contributes to the field of ABSA by demonstrating the first successful application of Dutch pre-trained language models to aspect-based sentiment analysis in the domain of human resources (HR).

Computer Science

What problem does this paper attempt to address?

This paper mainly discusses how to use machine learning methods for aspect-based sentiment analysis (ABSA) in employee satisfaction surveys. The researchers face the challenge of understanding employees' opinions and emotions, which are crucial for effective human resource management. However, open-ended responses introduce noise and variability, making large-scale analysis challenging. They propose using a Dutch BERT model to identify and extract specific aspects and their sentiments in employee satisfaction surveys to support more comprehensive analysis.

The paper first identifies six key aspects through cluster analysis: compensation, schedule, communication, contact, personal attention, and agreements. These aspects were validated by domain experts. The researchers constructed a dataset containing 1,458 Dutch survey responses, revealing imbalanced labels and sentiment distributions for each aspect.

Next, the paper presents a few-shot classification approach based on the Dutch BERT model and compares it with bag-of-words and zero-shot baselines. The researchers focus on several sub-problems, including the performance of the BERT-based model in few-shot classification and the impact of data augmentation on performance improvement.

The study found that the Dutch BERT model outperforms traditional bag-of-words models and zero-shot baselines in handling ABSA in employee satisfaction surveys. Additionally, data augmentation can improve classification performance, especially when data is limited.

The main contributions of the paper include:
1. Creation of a publicly available dataset of Dutch employee satisfaction survey responses annotated for ABSA.
2. Development of an ABSA model for Dutch open-ended employee satisfaction survey responses, enabling automated sentiment and aspect extraction.
3. Demonstration of the successful application of a Dutch pretrained language model in ABSA in the human resources domain.

The paper also reviews related work, including ABSA research based on English and the application of the Dutch BERT model in various tasks. Lastly, the paper provides detailed descriptions of the methodology, including dataset, data preparation, aspect selection, annotation study, data augmentation, model implementation, and evaluation metrics. Experimental results show that the BERT model with few-shot samples significantly outperforms traditional methods in aspect sentiment extraction, providing a new tool for ABSA in the human resources domain.