A Shared-Private Representation Model with Coarse-to-Fine Extraction for Target Sentiment Analysis

Abstract

Target sentiment analysis aims to detect opinion targets along with recognizing their sentiment polarities from a sentence. Some models with span-based labeling have achieved promising results in this task. However, the relation between the target extraction task and the target classification task has not been well exploited. Besides, the span-based target extraction algorithm has a poor performance on target phrases due to the maximum target length setting or length penalty factor. To address these problems, we propose a novel framework of Shared-Private Representation Model (SPRM) with a coarse-to-fine extraction algorithm. For jointly learning target extraction and classification, we design a Shared-Private Network, which encodes not only shared information for both tasks but also private information for each task. To avoid missing correct target phrases, we also propose a heuristic coarse-to-fine extraction algorithm that first gets the approximate interval of the targets by matching the nearest predicted start and end indexes and then extracts the targets by adopting an extending strategy. Experimental results show that our model achieves stateof-the-art performance.

Hierarchical Interactive Network for joint aspect extraction and sentiment classification

“…In previous models Lin and Yang, 2020;Lv et al, 2021), the feature representations are extracted independently except for using shared input. In other words, two tasks have no associations with each other, which is not in line with human ituition.…”

“…Laptop Res Tweets Zhou (Zhou et al, 2019) 59.76 71.98 51.44 Hu-pipeline ) 68.06 74.92 57.69 Hu-joint (Hu et al, 2019 64.59 72.47 54.55 Hu-collapsed (Hu et al, 2019) 48.66 57.85 48.11 SPRM (Lin andYang, 2020) and SC. Whereas in sequential and parallel encoding, sentiment features have no direct impact on the information of aspect features.…”

“…β is a balanced parameter. During the test phase, HI-ASA outputs two parts, one is the scores of boundaries, based on which we leverage the heuristic algorithm (Lin and Yang, 2020) to extract all the aspects; the other one is corresponding sentiment polarities of extracted aspects.…”

“…In recent years, the span-based models Zhou et al, 2019;Lin and Yang, 2020;Lv et al, 2021) are increasingly becoming an alternative for ASA because of their inherent advantages (e.g., they can avoid sentiment inconsistency and huge search space problems in tagging-based models (Luo et al, 2019;, where the aspects are extracted by directly predicting the boundary distributions, and the sentiment polarities are classified based on the aspect-level words. For example, in the sentence "The screen size is satisfactory but the phone battery capacity is limited.…”

An Iterative Multi-Knowledge Transfer Network for Aspect-Based Sentiment Analysis

“…;Xu et al, 2020a, etc). By contrast, the integrated or joint methods (Wang et al, 2016a;Mitchell et al, 2013;Li and Lu, 2017;Schmitt et al, 2018;Lin and Yang, 2020;Liang et al, 2021;Chen and Qian, 2020) can model the interactive correlations and thus achieve promising results. Different from above studies, we focus on exploiting the inter-task correlations among the three aspect-level subtasks and thus incrementally boost one another.…”