Towards Semantic Integration of Opinions: Unified Opinion Concepts Ontology and Extraction Task

Gaurav Negi, Dhairya Dalal, Omnia Zayed, Paul Buitelaar · 2025

Read Towards Semantic Integration of Opinions: Unified Opinion Concepts Ontology and Extraction Task by Gaurav Negi, Dhairya Dalal, Omnia Zayed, Paul Buitelaar (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-18703.txt.

Introduction

Gaurav Negi, Dhairya Dalal, Omnia Zayed, and Paul Buitelaar give opinions a shared semantic structure in a 2025 language technology study. Opinion mining extracts who feels what about which target, but every data set encodes those pieces differently, so tools and corpora cannot be combined. Their ontology unifies the facets of opinion that earlier research studied separately, using symbolic descriptions to fix the meaning of each part.

A new extraction task follows: pull opinions from text in the unified form, with richer expression than older schemes allowed. Shared structure should let opinion data travel between domains and applications.

Catalog record

CREATORGaurav Negi, Dhairya Dalal, Omnia Zayed, Paul Buitelaar
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-18703

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Towards Semantic Integration of Opinions: Unified Opinion Concepts Ontology and Extraction Task Gaurav Negi, Dhairya Dalal, Omnia Zayed, Paul Buitelaar Abstract This paper introduces the Unified Opinion Concepts (UOC) ontology to integrate opinions within their semantic context. The UOC ontology bridges the gap between the semantic representation of opinion across different formulations. It is a unified conceptualisation based on the facets of opinions studied extensively in NLP and semantic structures described through symbolic descriptions. We further propose the Unified Opinion Concept Extraction (UOCE) task of extracting opinions from the text with enhanced expressivity. Additionally, we provide a manually extended and re-annotated evaluation dataset for this task and tailored evaluation metrics to assess the adherence of extracted opinions to UOC semantics. Finally, we establish baseline performance for the UOCE task using state-of-the-art generative models. 1 Introduction Opinion 1 1 1 We use the term opinion as a broad concept that covers sentiment and its associated information such as opinion target and the person who holds the opinion, and use the term sentiment to mean only the underlying positive, negative or neutral polarity implied by opinion. mining has seen a move from a traditional sentence- and document-level analysis Pang et al. 2002 to fine-grained approaches. Aspect-based Sentiment Analysis (ABSA) is a notable approach for fine-grained opinion mining, and it has been extensively studied in natural language processing (NLP) research. Pontiki et al. 2014 ; Pontiki et al. 2015 ; Pontiki et al. 2016 ; Maia et al. 2018a . The task focuses on identifying the aspects of the entities and their associated sentiments from a given text sequence. In the following sentence: “I had hoped for better battery life , as it had only about 2-1/2 hours doing heavy computations (8 threads using 100 % of the CPU) ." ABSA results are extracted as the following tuple: { battery life , Battery#Operational_Performance, negative}. The extracted tuple is in the form {aspect term/opinion target, entity#aspect category, sentiment polarity}. Opinion target (often called aspect term) is the word or phrase over which an opinion is expressed. The aspect category is an attribute of the opinion target, and sentiment polarity specifies whether the opinion is positive, negative, or neutral. This fine-grained analysis allows for a more detailed understanding of opinions and sentiments expressed in the text. Structured sentiment analysis Barnes et al. 2022 is another formulation of opinion mining, where the nodes are spans of sentiment holders, targets and expressions, and the arcs are the relations between them. Figure 1 illustrates this formulation. Figure 1: Structured Sentiment Analysis ABSA and structured sentiment analysis overlap significantly in extracting specific opinion facets 2 2 2 We use the term facet as used by Bing Liu to describe various subtasks and the building blocks of an opinion. . None of the formulations fully incorporate all opinion facets proposed by Liu 2017 , which reduces the expressiveness and granularity of the extracted opinions. The example above shows that the opinion is valid for specific individuals or groups engaged in " doing heavy computations ". The reason for opinion is also expressed, i.e. “it had only about 2-1/2 hours" . None of the existing opinion mining formulations enable these extractions. This work investigates semantic representations of opinions to enrich their expressiveness. Towards this end, we studied the specification of opinion for the Semantic Web as described by the Marl Ontology Westerski et al. 2011 . However, it has a limited cross-compatibility with the opinion formulations researched in NLP. We unify the opinion facets studied extensively in NLP with the semantic structures described in the Marl Ontology to develop a comprehensive Unified Opinion Concepts (UOC) framework. The UOC ontology consolidates and formalises these opinion components into an exhaustive set, enabling the semantic representation of opinions in a structured and unambiguous manner. UOC leverages the implicit hierarchies and relationships across diverse NLP frameworks based on the theoretical foundations of Liu and Zhang 2012 . Our contribution 3 3 3 Github Repository: https://github.com/gauneg/UnifiedOpinionConcepts_LDK_2025 can be summarized as follows: • We introduce the UOC ontology (Section 3 ) that conceptualises semantic representation of an opinion, improving on the existing opinion formulations in terms of expressivity and cross-compatibility. • We define Unified Opinion Concept Extraction (UOCE) as an NLP task (Section 4.1 ) grounded in the rich semantic representation of the UOC ontology (Section 4.4 ). • We extend annotations of an existing gold standard opinion mining dataset (Section 4.3 ) to create an evaluation dataset for UOCE. We propose tailored evaluation metrics (Section 4.2 ) for rigorous baseline assessment. 2 Related Work Opinion Mining in NLP. ABSA evolved from feature-based summarisation Hu and Liu 2004 ; Zhuang et al. 2006 ; Ding et al. 2008 and the foundational work on opinion mining by Liu and Zhang 2012 , which involves extracting and summarising opinions on features (attributes/keywords). The downstream tasks that spun out of the ABSA research space can be classified into the following categories based on the opinion facets they address: Opinion Aspect Co-extraction Qiu et al. 2011 ; Liu et al. 2013 ; Li et al. 2018 ; Wang et al. 2017 , Aspect Sentiment Triple Extraction (ASTE) Zhang et al. 2020 ; Xu et al. 2020 ; Wu et al. 2020 , Target-Aspect-Sentiment Detection (TASD) Ma et al. 2018 ; Wu et al. 2021 , Aspect-Category-Opinion-Sentiment (ACOS/ASQP) quadruple extraction Cai et al. 2021 ; Gou et al. 2023 ; Xiong et al. 2023 . Barnes et al. 2021a ; Barnes et al. 2021b perform opinion tuple extraction as dependency graph parsing, where the nodes are spans of sentiment holders, targets and expressions, and the arcs are the relations between them (see Figure. 1 ). We extend these existing opinion formulations by adding more elements to increase expressivity and formalise the relationships between opinion facets with an ontology. Ontological Methods. Ontologies provide an explicit machine-readable specification of shared conceptualization, and our inquiry into existing ontologies for opinion expression led us to the Marl Ontology 4 4 4 https://www.gsi.upm.es/ontologies/marl/ Westerski et al. 2011 . It is a standardised schema designed to annotate and describe subjective opinions expressed on the Semantic Web and in information systems Sánchez-Rada et al. 2016 ; Buitelaar et al. 2013 . However, the Marl ontology cannot describe fine-grained opinion mining currently being researched in NLP. Schouten and Frasincar 2018 propose a task ontology to facilitate sentiment classification of the given aspect terms; it does not contribute towards highlighting fine-grained opinion representation. Our work reformulates and extends the domain ontology of an opinion, improving the interfacing of opinion description across different disciplines. Neuro-Symbolic Methods. Sentiment Analysis with neuro-symbolic methods adds knowledge and symbolic constraints to assist the deep learning models. This knowledge can be in the form of structured linguistic characteristics with WordNet, SentiWordNet Kocon et al. 2022 , word-sense disambiguation Baran and Kocon 2022 ; Zhang et al. 2023 or using domain-specific knowledge He et al. 2023 . Neuro-symbolic work on opinion mining does not extend or introduce novel formulations of opinion-mining tasks. We utilise the nuances of opinion-mining literature to reformulate the opinion ontology to bridge the gap between the differences in the semantic conceptualisation of opinion expression. We align the concepts of Marl with the various opinion

Read the complete text