---
title: "Efficient Parsing Techniques for Automated Entity Extraction in Software Requirements"
id: "plos-one-7-parsing-based-approach-for-entity-extraction-from-natural-language-software"
canonical_url: "https://medichelpline.com/clinical-feed/plos-one-7-parsing-based-approach-for-entity-extraction-from-natural-language-software"
content_type: "clinical_feed_article"
specialty: "Gastroenterology"
source_name: "PLOS ONE (Medicine)"
source_url: "https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0361105"
published_at: "2026-10-21T14:00:00.000Z"
evidence_level: "Journal Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# Efficient Parsing Techniques for Automated Entity Extraction in Software Requirements
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/plos-one-7-parsing-based-approach-for-entity-extraction-from-natural-language-software
- **Specialty:** [Gastroenterology](https://medichelpline.com/clinical-feed/gastroenterology.md)
- **Primary Source:** PLOS ONE (Medicine)
- **Source URL:** [Original Journal Publication](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0361105)
- **Published At:** 2026-10-21T14:00:00.000Z
- **Evidence Rating:** Journal Feed
## Executive GIST (TL;DR)
- This study addresses ambiguities in software requirements that occur due to natural language usage. - Manual extraction of **requirement entities** is typically labor-intensive and error-prone. - A novel **hybrid approach** combining constituency and dependency parsing is proposed for entity extraction. - The dataset is sourced from the healthcare domain and manually annotated to ensure accuracy. - Results indicate the hybrid method significantly outperforms individual parsers in both recall and F1-score while achieving precision comparable to the constituency parser. - The proposed technique enhances comprehension of requirements without reliance on extensive training datasets or machine learning models.
## Clinical Analysis & Structured Key Points
Parsing-based approach for entity extraction from natural language software requirements | PLOS One Browse Subject Areas ? Click through the PLOS taxonomy to find articles in your field. For more information about PLOS Subject Areas, click here . Article Authors Metrics Comments Media Coverage Reader Comments Figures Figures Abstract Software requirements are written in natural language, so they are inherently ambiguous. This ambiguity results in the misinterpretation of the requirements and leads to an incorrect implementation of the system that does not meet the user’s needs. So, it is important for all the stakeholders to comprehend the requirements. The requirement statements revolve around the requirement entities. Identifying requirement entities is essential to understand requirements, and they are used in identifying coreferent entities, creating glossaries, producing summaries of documents, feature extraction, requirements classification, and domain understanding. Manual entity extraction is a labour-intensive task. Recently, transformer-based models have played an important role in improving entity extraction results, but they require extensive pre-training of models for domain-specific tasks. To overcome these challenges, the proposed automated approach is used to extract entities from software requirements without the need for manual efforts and a large dataset for pre-training. In this study, two widely used parsing techniques, constituency parsing and dependency parsing, are applied to extract noun phrase-based entities. In addition, a hybrid approach comprising the combined extractions of both parsers is developed. The dataset comprises the software requirements from the healthcare domain and is manually annotated to establish a gold standard for evaluation. Experimental results indicate that the hybrid approach achieves the best overall extraction performance among the evaluated methods. While the constituency parser achieves the highest precision, it exhibits comparatively lower recall. In contrast, the hybrid approach achieves the highest recall and F1-score while maintaining precision comparable to that of the constituency parser. These results demonstrate that the hybrid method provides a more effective solution for automatically extracting entities from software requirements by leveraging syntactic structures of sentences rather than relying on domain-specific model training. Citation: Amjad S, Saleem A, Khan AS, Anwar MW (2026) Parsing-based approach for entity extraction from natural language software requirements. PLoS One 21(10): e0361105. https://doi.org/10.1371/journal.pone.0361105 Editor: Mario Graff-Guerrero, Infotec: Centro de Investigacion e Innovacion en Tecnologias de la Informacion y Comunicacion, MEXICO Received: March 19, 2026; Accepted: October 6, 2026; Published: October 21, 2026 Copyright: © 2026 Amjad et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: ***A/PROS at Accept: Please request public link to data at accept and update DAS*** The minimal dataset underlying the results presented in this study has been deposited in a restricted-access Zenodo repository. The repository can be accessed during peer review using the following review link: https://zenodo.org/records/20840646?preview=1&token=eyJhbGciOiJIUzUxMiJ9.eyJpZCI6IjMzYTU0NGE2LTMzOWMtNDY3YS1iY2IxLWJlOTQyZDg1ZDY0OSIsImRhdGEiOnt9LCJyYW5kb20iOiJlNDczYTNlYmRkZDQwYWU4YWM2NmI3YzBjNGRkMWJlZiJ9.1PWF0S1qfCFnVcpPEauhMUPZ8_horzZic52Yx1tdHTdfmfZpk65fln_j9aA4Zw8NRt2Ag_r0up045X5bLlfN_Q The dataset has been assigned the reserved DOI https://doi.org/10.5281/zenodo.20840646 and will be made publicly available upon acceptance of the manuscript. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Introduction Software requirements cover the scope of the project. Requirements indicate what is to be developed, serving as a guideline for stakeholders and the development team. Software requirements include both the functional and non-functional requirements of the project [ 1 ]. Requirements are important for all stakeholders, including technical and business participants. The subsequent software development phases are dependent on the requirements, whether it is the development or operation phase [ 2 ]. The requirement engineers write requirements in natural language, which is understandable for all the stakeholders [ 1 - 3 ]. Natural language has some inherent properties that lead to quality issues, including inconsistency, incompleteness, and ambiguousness [ 2 , 4 ]. The needs and requirements refer to the things that are called entities. The terms may belong to different levels, which include business unit, product, system, or system element, and enterprise [ 5 ]. Requirement statements have explicit specification of concepts, which are the function objects. The function objects represent the requirements entities in the statements [ 6 ]. The process of extracting the core information of the requirement document is called information extraction or entity recognition [ 7 ]. In the existing studies, different terms are used to represent the same or similar concept. For the same concept, the terms used are Entity Extraction [ 1 , 6 - 8 ], Entity Terms Extraction, and Domain-Specific Entity Terms Extraction [ 9 ]. For a similar concept, the terms used are: Requirements Terms Extraction, Terminology extraction [ 10 ], Glossary Term Extraction [ 11 , 12 ], and Requirements-Related Term Extraction [ 13 ]. These requirement-specific entities are the noun phrases [ 6 - 9 ], whereas glossary terms consist of an entity list and an action list. The entity list consists of concrete nouns, while the action list consists of process nouns and concrete verbs [ 14 ]. Some linguistic approaches have extracted only noun phrases as requirement glossary terms [ 15 ]. The requirement entities play a key role in the understanding of the requirement concepts. Once the entities are extracted from requirement statements, they are used in information extraction [ 6 - 8 ], feature modelling [ 10 ], feature extraction [ 16 ], size estimation of the project using function point analysis [ 17 ], and identifying domain elements for creating various UML diagrams, particularly class diagrams [ 18 ]. Requirement entities also play a significant role in improving the quality of software requirements in quality assessment systems [ 1 ]. Requirement terms are used to extract domain knowledge for understanding the system domain by the analyst [ 13 ]. Glossary term extraction is used in the creation of glossaries [ 9 , 11 , 12 ]. These terms are further used in the identification of term aliases, in which different terms refer to the same concept or entity. These aliases need to be identified to avoid misinterpretation of requirements [ 4 , 9 , 12 ]. These terms should have a common definition in the glossaries to avoid any misunderstandings of technical terms among stakeholders in the project domain [ 15 ]. Manual extraction of entities from requirement statements is a laborious and time-consuming task. It also leads to missing and duplicate entity extraction [ 6 , 10 ]. Heuristics are applied to extract entities from the requirements and find their corresponding match to identify co-referent entities and anaphoric ambiguities. Since the approaches are based on human judgement, they are not suitable for domain-specific tasks [ 19 ]. Parsers for Parts of Speech (POS) tagging are trained on generic corpora like public news corpora. There is a need to create a domain-specific dictionary that needs to be integrated into the parser to analyse the words correctly for a domain-specific task [ 17 ]. Parser-based approaches are also used in the feature extraction tasks [ 10 ]. To extract the requirements-related knowledge based on the terms, a document repository is required [ 13 ]. Rule-based approaches are used to identify term aliases [ 9 , 12 ], extract glossary terms and generate glossaries [ 11 , 15 , 20 ], extract requirements-related knowledge [ 13 ], and identify entity terms for domain elements [ 18 ]. POS tagging-based techniques have relatively less precision and recall. A few studies have used Machine Learning (ML) and Deep Learning (DL) based approaches for the Named Entity Recognition (NER) [ 1 , 7 ], entity extraction [ 6 , 8 ], and glossary term extraction [ 21 ] tasks. The models used are pre-trained on general corpora. As the requirements are domain-specific, the models give the best results for precision and recall evaluation measures when trained on a large volume of domain-specific corpus. Generating a domain-specific corpus requires extensive annotation of large volumes of data. There are chances of inconsistencies and discrepancies in the labelling of data by humans [ 1 , 6 , 7 ]. Different software needs are expressed using various sentence structures, so an automated approach is required to correctly identify the structure of the sentences [ 18 ]. The goal of this research is to automatically extract entities from software requirement statements based on sentence structure. An automated syntax-driven approach is proposed to extract software requirement entities by integrating constituency parsing and dependency parsing techniques. The proposed approach does not require additional training on domain-specific data, as it relies on pretrained parsers on general linguistic corpora. It also does not need the extensive configurations of machine learning models and the setting of different parameters. Since the entity extraction approach is based on grammatical structure and does not rely on extensive manual labelling of requirement entities, it is potentially transferable across different domains. An analysis of constituency parsing, dependency parsing, and a hybrid approach is performed to identify the most effective strategy for the entity extraction task. The following are the contributions of this research: Prepare a dataset comprising requirement statements from the healthcare domain to demonstrate the effectiveness of the parsing techniques. A list of manually annotated entities serving as the gold standard. Using encoder and decoder-based constituency and transition-based dependency parsers to extract noun phrases termed here as requirement entities from software requirements. These parsers are implemented using spaCy and adapted to deal with special characters like hyphens and apostrophes. Development and use of a hybrid approach by integrating the results of constituency and dependency parsers after removing duplications to improve the coverage of noun-phrase spans. A comparison of the results of the constituency parser, the dependency parser, and the hybrid approach. The output of the approach is the list of requirement entities from each requirement statement, and from that list, unique entities are also extracted. The evaluation of approaches indicates that the proposed approach advances the state of the art in requirements entity extraction. The remaining research paper has the following sections. Section II presents the approaches available in the literature for entity or terminology extractions. Section III consists of the background of the two parsing techniques, namely constituency parsing and dependency parsing. Section IV states the methodology, including dataset preprocessing, manual entity annotation, entity extraction using parsers, and a hybrid approach, and the evaluation metrics used to evaluate the results. In Section V results are discussed and presented. Section VI is about the external and internal threats to the validity. Section VII concludes the research paper and presents the future opportunities. Related work In this section, approaches used in the literature for extracting entities, terminologies, and glossary terms from software requirements are discussed. These approaches include rule-based or statistical approaches and learning-based approaches. Rule-based or statistical approaches Entity extraction is used to identify term aliases from requirements specifications, which may cause problems in understanding the requirements. POS tagging is applied on the requirements, and entities are extracted based on noun tags. Syntactically and semantically related term pairs are identified by using a combination of the approaches, including latent semantic similarity, tag-based similarity, transitive similarity, and co-location similarity. As a result, a cluster of entity-term aliases is produced [ 9 ]. The approach presented in [ 9 ] is later extended, and a more efficient solution is proposed, which results in an extended list of alias terminologies. The Latent Semantic Analysis (LSA) model is improved by adding semantic compression. The tag-based similarity approach is improved by adding tag-equivalence-based compression. WordNet-based analysis is used to eliminate false positive results. The process of cluster generation of aliases is also improved [ 12 ]. Glossaries are helpful for understanding the requirement terms and removing ambiguities and vagueness of the concepts. Clustering of these glossary terms groups related terms. The approach presented in [ 20 ] differs from existing work, as it does not provide a flat list of terms; instead, it makes clusters by relevance. Moreover, the approach is evaluated based on two industrial case studies from different domains. The terms are extracted through a noun chunking pipeline along with some heuristics. For clustering of the terms, the Expectation Maximization (EM) algorithm is used. An automated, mathematically based procedure, the Bayesian Information Criterion (BIC), is used for choosing the number of clusters. Later, the research is extended in [ 15 ] by computing the syntactic and semantic relevance between terms using state-of-the-art methods. The clusters of these relevant terms are produced by using three clustering algorithms: K-means, Agglomerative Hierarchical, and EM (performs best). The experiment is conducted on three case studies, and the perception of subject matter experts is also included. The evaluation indicates that for requirements documents, this approach gives 20 percent more accuracy in terms of recall when compared with generic term extraction tools, while the precision is comparable. An automated hybrid approach to extract the most relevant and domain-specific glossary terms is proposed in [ 11 ], which is composed of linguistic terms extraction and statistical filtering. These technical glossary terms are the noun phrases. Linguistic extraction is based on the Natural Language Toolkit (NLTK), and it is used in the process of tokenization, POS tagging, chunking, and lemmatization. Relevance is computed by calculating the number of requirements in which the candidate term is present, and specificity is computed by doing a comparison of the number of occurrences with a non-domain-specific text corpus. The evaluation measures indicate that careful application of linguistic and statistical approaches improves recall [ 11 ]. Requirements-related knowledge is important for understanding the requirements of a specific domain, particularly when an analyst does not have relevant domain knowledge. It helps classify the requirements into functional and non-functional categories. During requirement elicitation, the real-time snippet is matched with the documented repository. The exact matched terms are extracted. The relevance of terms with respect to context is measured by calculating lexical association using generative models (based on Weighted Finite State Transducers (WFSTs) and statistical Language Models (LMs)) and discriminative models (based on Kernel methods and Support Vector Machines (SVM)). They are used to extract and classify the most related snippet [ 13 ]. To understand the problem domain, terminologies are extracted from natural language requirements using natural language processing (NLP) techniques and ML algorithms. Requirements terms are extracted based on noun phrase identification by applying a constituency parser using spaCy. An algorithm using spaCy and two heuristic rules is used to extract the requirement terms. These terminologies are ranked with respect to relevance by using a TextRank algorithm based on unsupervised graphs, and features are extracted for feature modelling [ 10 ]. Entity term identification is also important during the extraction of domain elements for generating class diagrams from software requirements. A language model is constructed to identify the type of structure of the sentence based on the verb patterns proposed by Hornby. Type dependency is generated by the Stanford Parser API, which is used to identify the semantic relationships of words in a requirement statement to identify the sentence structure. POS tagging is done for each sentence to extract entity terms based on noun phrases, and heuristics are applied to identify entity classes and attributes. Different types of relationships (including association, generalization, and aggregation) are identified based on rules, and a class diagram is generated. The evaluation metrics for the class diagram used in the experiments are correctness, completeness, and redundancy [ 18 ]. Learning-based approaches Glossary terms are not only extracted from software requirements but also from source code and the documentation of the project. The concept extraction process consists of the extraction of a seed term, the extraction of an extended term, and the merging of an alias. Initial seed terms are extracted from the source code and sentences using heuristics. A semi-supervised learning method, Long Short-Term Memory-Conditional Random Fields (LSTM-CRF), is used iteratively to extract more candidate terms from the sentences. Aliases of the same concept are detected from the candidate terms. The lemmatization of the candidate terms is performed, and then morphological synonyms are identified and merged. Similarly, abbreviations are also detected and merged [ 21 ]. To address the challenges of requirements entity extraction, including domain specificity and extensive effort in labelling, a semi-automated approach called Requirement Entity Extraction (RENE) is proposed. It is a transfer learning approach, and its main components are the LSTM language model and the LSTM-CRF model. The LSTM language model is trained on a general corpus and fine-tuned on unlabelled requirement sentences. The LSTM and embedding layers from the LSTM language model are transferred to the LSTM-CRF model. Each requirement entity is assigned a word sequence using labelled requirements, and a BIO (Beginning, Inside, Outside) tag is assigned to each word in the word sequence. The CRF layer is trained using sequence tagging of words. The output is the automated extraction of requirement entities in BIO format. This approach requires labelled requirements for sequence tagging, which is a labour-intensive task [ 6 ]. Machine learning approaches are used for information extraction tasks. The spaCy provides an NLP pipeline that includes neural network-based components, and BERT is a transformers-based model that uses bidirectional encoder architecture. These models are pre-trained on a generic dataset and are used for the information extraction tasks. The dataset is BIO tagged to extract entities, including name, location, facility, and relation types. The sentences are extracted from paragraphs containing these tags. Relationship types are extracted using dependency parsing in spaCy. For domain-specific tasks, the models are trained on a domain-specific dataset to extract the domain-specific entities and to
## Related Clinical Research

- [Artificial intelligence applications in liver transplantation: scoping review of 196 studies](https://medichelpline.com/clinical-feed/plos-one-15-the-applications-of-artificial-intelligence-in-liver-transplantation-a.md)
- [Stool chemistry plus metagenomics identifies distinct adult gut health profiles](https://medichelpline.com/clinical-feed/biorxiv-6-integrating-stool-chemistry-measurements-with-metagenomics-in-adults-reveals.md)
- [IBDex frozen atlas converts bulk transcriptomics into reusable IBD reference](https://medichelpline.com/clinical-feed/medrxiv-17-ibdex-a-frozen-patient-scale-atlas-for-bulk-transcriptomics-in-inflammatory.md)
- [ERG344 ACOD1 inhibition reduces colitis and tumor growth via a shared myeloid mechanism](https://medichelpline.com/clinical-feed/biorxiv-19-small-molecule-acod1-inhibition-attenuates-colitis-and-tumor-growth-through-a.md)
- [Fatty liver in lean patients and five other research briefs](https://medichelpline.com/clinical-feed/bmj-2-leaning-into-fatty-liver-findings-and-other-research.md)

## Navigation
- [← Back to Gastroenterology Feed](https://medichelpline.com/clinical-feed/gastroenterology.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.