Comprehensive analysis of aspect term extraction methods using various text embeddings

• A wide comparison with ablation analysis of Aspect Term Extraction methods. guideline to choose the best models for particular user needs. We analyzed 88 method combinations, model customizations (LSTM vs. BiLSTM, CRF no CRF) and pre-trained word embeddings. also evaluated results three contextual text representations (BERT, Flair, ELMo) using BiLSTM-CRF model. influence on performance extending vectorization step character-based The experimental SemEval datasets revealed that BiLSTM could be used as a very good predictor. Language model-based are not always obvious choice vector representation layer in NLP tasks. Recently, variety designs methods have blossomed context sentiment domain. However, there is still lack comprehensive studies Aspect-based Sentiment Analysis. want fill this gap propose various embeddings particularly focused simple architectures based long short-term memory (LSTM) optional conditional random field (CRF) enhancement different Moreover, we bi-directional (BiLSTM) predictor, even comparing sophisticated complex huge or language models. presented LSTM-based architecture word/character version