Atesa-bært: A Heterogeneous Ensemble Learning Model For Aspect-based Sentiment Analysis

The increasing volume of online reviews has made possible the development of sentiment analysis models for determining the opinion of customers regarding different products and services. Until now, sentiment analysis has proven to be an effective tool for determining the overall polarity of reviews. To improve the granularity at the aspect level for a better understanding of the service or product, the task of aspect-based sentiment analysis aims to first identify aspects and then determine the user’s opinion about them. The complexity of this task lies in the fact that the same review can present multiple aspects, each with its own polarity. Current solutions have poor performance on such data. We address this problem by proposing ATESA-BÆRT, a heterogeneous ensemble learning model for Aspect-Based Sentiment Analysis. Firstly, we divide our problem into two sub-tasks, i.e., Aspect Term Extraction and Aspect Term Sentiment Analysis. Secondly, we use the argmax multi-class classification on six transformers-based learners for each sub-task. Initial experiments on two datasets prove that ATESA-BÆRT outperforms current state-of-the-art solutions while solving the many aspects problem.

Introduction. Sentiment Analysis is currently an efficient and wildly used tool for extracting the overall user opinion from product and service reviews [7]. Aspect-Based Sentiment Analysis (ABSA) is an extension of Sentiment Analysis that introduces an additional granularity level represented by aspect identification and polarity prediction. The current literature proposes various models, from simple classifiers that solve specific tasks (aspect extraction, keywords extraction, or sentiment analysis) to more complex architectures involving transformers, pre-trained models, or chained neural networks. Even though the general sentence and document level sentiment analysis has been used in a many domains [22, 24, 25, 34], the ABSA task is not something that is implemented very often due to the huge challenge created by the large variety of possible aspects and categories.

Discussion / Conclusion. Sentiment Analysis is one of the main tools used to determine the opinion of used users from product and service reviews. Extracting the aspects from online reviews with corresponding polarities can result in high efficiency in understanding customer problems for different domains, ranging from simple shops and restaurants to online retailers and business-to-business scenarios. In this paper, we propose, design and implement a new heterogeneous ensemble architecture, ATESA-BÆRT, consisting of 12 different models that use both pre-trained and fine-tuned BERT and BART transformers, in order to solve the Aspect-Based Sentiment Analysis problem. We split the problem into two sub-problems, for a better understanding and to better understand and observe the behavior of the models: (1) Aspect Term Extraction (ATE), and (2) Aspect Term Sentiment Analysis (ATSA). We test ATESA-BÆRT on two datasets: textitSemEval16 Task 5 Restaurants (SE2016T5R) [27] and Multi Aspect Multi Sentiment (MAMS) [13]. The experimental results show that the proposed model obtains a high accuracy after training it for only 2 epochs using a train-test split of 80%-20%.