Lexical data augmentation for sentiment analysis

Abstract

Machine learning methods, especially deep learning models, have achieved impressive performance in various natural language processing tasks including sentiment analysis. However, deep learning models are more demanding for training data. Data augmentation techniques are widely used to generate new instances based on modifications to existing data or relying on external knowledge bases to address annotated data scarcity, which hinders the full potential of machine learning techniques. This paper presents our work using part-of-speech (POS) focused lexical substitution for data augmentation (PLSDA) to enhance the performance of machine learning algorithms in sentiment analysis. We exploit POS information to identify words to be replaced and investigate different augmentation strategies to find semantically related substitutions when generating new instances. The choice of POS tags as well as a variety of strategies such as semantic-based substitution methods and sampling methods are discussed in detail. Performance evaluation focuses on the comparison between PLSDA and two previous lexical substitution-based data augmentation methods, one of which is thesaurus-based, and the other is lexicon manipulation based. Our approach is tested on five English sentiment analysis benchmarks: SST-2, MR, IMDB, Twitter, and AirRecord. Hyperparameters such as the candidate similarity threshold and number of newly generated instances are optimized. Results show that six classifiers (SVM, LSTM, BiLSTM-AT, bidirectional encoder representations from transformers [BERT], XLNet, and RoBERTa) trained with PLSDA achieve accuracy improvement of more than 0.6% comparing to two previous lexical substitution methods averaged on five benchmarks. Introducing POS constraint and welldesigned augmentation strategies can improve the reliability of lexical data augmentation methods. Consequently, PLSDA significantly improves the performance of sentiment analysis algorithms.

Will sentiment analysis need subculture? A new data augmentation approach

“…It has undergone instruction fine-tuning on the basis of LoRA (Hu et al, 2021), and the constructed prompt's instruction is "Please determine whether the following content expresses a positive sentiment, and output 0 or 1 -->," with specific content included as input, corresponding labels as output. In contrast, the comparative methods encompass the commonly used techniques such as DICT (Zhang et al, 2015), EDA (Wei & Zou, 2019), and PLSDA (Xiang et al, 2021). The metrics is the accuracy (%).…”

“…The manual aggregation of domain‐specific training datasets can be a labor‐intensive and costly endeavor. This challenge underscores the critical need for an exploration into data augmentation strategies, which has secured a broad footprint in the realm of image processing, typically implemented through transformations (Alqudah et al, 2023; Niu et al, 2023; Shorten & Khoshgoftaar, 2019; Xiang et al, 2021).…”

“…Instead, unique tasks dictate the introduction of bespoke enhancement methodologies. This could span the spectrum from commonsense reasoning and automatic translation to text comprehension and generation, extending to entity extraction and sentiment analysis (Abonizio et al, 2021; Hsu et al, 2021; Liesting et al, 2021; Liu et al, 2021; Shen et al, 2020; Xiang et al, 2021; Yang et al, 2020). Regarding sentiment analysis, popular augmentation strategies encompass synonym substitution (Zhang et al, 2015) and a technique known as easy data augmentation (EDA) (Wei & Zou, 2019).…”

Aspect sentiment mining of short bullet screen comments from online TV series

“…The UDA approach offers a new way of effectively detecting and utilizing noise in unlabeled data (Xie et al, 2020). It unifies the learning process between labeled and unlabeled data through a specific form of contrastive learning and corresponding data augmentation (Xiang et al, 2021), and achieves superior results in image and regular text classification (Xie et al, 2020). The potential of UDA for short BSCs is worth investigating.…”

Data augmentation in natural language processing: a novel text generation approach for long and short text classifiers

“…Thus many methods have been tried out in research so far. Among them are methods for swapping [ 59 ], deleting [ 16 , 38 ], inducing spelling mistakes [ 6 , 10 ], paraphrasing [ 28 ], and replacing of synonyms [ 25 , 61 , 66 ], close embeddings [ 2 , 58 ] and words predicted by a language model [ 11 , 18 , 24 ] on word-level. On a broader level, methods which change the dependency tree [ 45 , 62 ], perform round-trip-translation [ 27 , 47 ], or interpolate the input instances [ 9 , 65 ] are used.…”