【 title 】A Reinforced Active Learning Approach for Optimal Sampling in Aspect Term Extraction for Sentiment Analysis

【 The author team 】Manju Venugopalan, Deepa Gupta

【 Date of publication 】2022.7.21

【 Thesis link 】https://www.sciencedirect.com/sdfe/reader/pii/S0957417422013793/pdf

【 Recommended reasons 】 Aspect level emotion analysis is a detailed task in emotion analysis , It identifies product features from a self righteous text , And map emotions to each feature . Supervised ML The algorithm reports relatively high performance in aspect level emotion analysis , But the cost is a lot of qualitative marker data . The data marking of such detailed tasks also requires domain expertise . therefore , The mechanism of extracting the smallest subset of information representing almost the whole data will be a breakthrough to reduce the annotation cost to a great extent . The method proposed in this paper is a sampling strategy based on active learning , Used for aspect term extraction , This is a subtask of aspect level affective analysis , Used to identify product features . The sampling strategy is automated through reinforcement learning , Extract the best samples from the whole unlabeled training data , This optimizes data annotation by reducing the time and effort associated with the marking process . In the era of data driven , This work is very important . The model has been used in SemEval(2014-2016) Data sets of laptops and restaurants were tested . Experimental proof , The size of training data can be significantly reduced on different data sets .