A Syntax-Injected Approach for Faster and More Accurate Sentiment Analysis

Sentiment Analysis (SA) is a crucial aspect of Natural Language Processing (NLP), focusing on identifying and interpreting subjective assessments in textual content. Syntactic parsing is useful in SA as it improves accuracy and provides explainability; however, it often becomes a computational bottleneck due to slow parsing algorithms. This article proposes a solution to this bottleneck by using a Sequence Labeling Syntactic Parser (SELSP) to integrate syntactic information into SA via a rule-based sentiment analysis pipeline. By reformulating dependency parsing as a sequence labeling task, we significantly improve the efficiency of syntax-based SA. SELSP is trained and evaluated on a ternary polarity classification task, demonstrating greater speed and accuracy compared to conventional parsers like Stanza and heuristic approaches such as Valence Aware Dictionary and sEntiment Reasoner (VADER). The combination of speed and accuracy makes SELSP especially attractive for sentiment analysis applications in both academic and industrial contexts. Moreover, we compare SELSP with Transformer-based models trained on a 5-label classification task. In addition, we evaluate multiple sentiment dictionaries with SELSP to determine which yields the best performance in polarity prediction. The results show that dictionaries accounting for polarity judgment variation outperform those that ignore it. Furthermore, we show that SELSP outperforms Transformer-based models in terms of speed for polarity prediction.

💡 Research Summary

The paper addresses a well‑known bottleneck in syntax‑driven sentiment analysis: conventional dependency parsers such as Stanza provide high‑quality Universal Dependencies (UD) trees but are computationally expensive, limiting their use in large‑scale or real‑time applications. To overcome this, the authors propose a Sequence‑Labeling Syntactic Parser (SELSP) that reformulates dependency parsing as a token‑wise labeling problem. SELSP predicts, for each token, its head index and dependency relation using a BiLSTM‑CRF or Transformer‑based encoder, achieving linear‑time inference.

SELSP is trained on UD‑annotated corpora for English and Spanish. While its unlabeled and labeled attachment scores (UAS/LAS) are modestly lower (≈1–2 % absolute) than Stanza’s, the authors demonstrate that this small loss does not affect downstream polarity classification because sentiment analysis requires only “good enough” syntactic information. The core sentiment analysis engine follows a rule‑based architecture: (1) a sentiment lexicon (e.g., SO‑CAL, VADER, SentiWordNet) supplies sentiment‑bearing words; (2) the parser‑generated dependency tree is traversed; (3) a set of compositional rules computes the effect of negation, intensification, conjunction, and other polarity‑shifting constructs. Rules are applied bottom‑up, allowing the system to correctly handle long‑distance dependencies that simple fixed‑window heuristics miss.

The authors evaluate SELSP against three baselines: (a) Stanza (full UD parser), (b) VADER (lexicon‑based, no syntax), and (c) a RoBERTa‑based transformer fine‑tuned for a five‑label sentiment task (strongly positive, weakly positive, neutral, weakly negative, strongly negative). Experiments are conducted on parallel English and Spanish datasets covering ternary polarity (positive, neutral, negative). Results show that SELSP processes sentences roughly 3.5× faster than Stanza while achieving 0.4–1.2 % higher accuracy on the polarity classification task. Compared with VADER, SELSP offers comparable speed but markedly better accuracy on sentences with complex syntactic scopes (e.g., “not in any respect good”). Against the RoBERTa model, SELSP is 4–5× faster, and while RoBERTa attains a slightly higher macro‑F1 on the five‑label task, its inference cost makes it less suitable for high‑throughput scenarios.

A further contribution is the systematic analysis of sentiment lexicons. The authors test several dictionaries, including those that account for polarity‑variation (e.g., VADER‑Extended) and those that provide only static polarity scores. They find that variation‑aware dictionaries consistently outperform static ones, and that combining multiple lexicons yields the best performance, suggesting that lexical diversity helps capture nuanced sentiment expressions.

In summary, the paper makes three key contributions: (1) introducing the first sequence‑labeling‑based parser for fast, syntax‑aware sentiment analysis; (2) demonstrating that this parser, when coupled with transparent rule‑based polarity computation, outperforms both traditional parsers and state‑of‑the‑art transformer classifiers in terms of speed while maintaining or improving accuracy; and (3) providing an empirical study on the interaction between sentiment lexicons and syntactic rules, offering practical guidance for building efficient, explainable sentiment analysis pipelines. The work is especially relevant for industry deployments where latency, resource consumption, and interpretability are critical.