A Hybrid Modified Semantic Matching Algorithm Based on Instances Detection With Case Study on Renewable Energy

This Matching input keywords with historical or information domain is an important point in modern computations in order to find the best match information domain for specific input queries. Matching algorithms represents hot area of researches in computer science and artificial intelligence. In the area of text matching, it is more reliable to study semantics of the pattern and query in terms of semantic matching. This paper improves the semantic matching results between input queries and information ontology domain. The contributed algorithm is a hybrid technique that is based on matching extracted instances from booth, the queries and in information domain. The instances extraction algorithm that is presented in this paper are contributed which is based on mathematical and statistical analysis of objects with respect to each other and also with respect to marked objects. The instances that are instances from the queries and information domain are subjected to semantic matching to find the best match, match percentage, and to improve the decision making process. An application case was studied in this paper which is related to renewable energy, where the input queries represents the customer requirements input and the knowledge domain is renewable energy vendors profiles. The comparison was made with most known recent matching researches.

💡 Research Summary

The paper presents a novel hybrid semantic matching algorithm that leverages instance detection to improve the alignment between user queries and a structured knowledge domain. Traditional text‑matching approaches—whether keyword‑frequency based (TF‑IDF, BM25) or modern deep‑learning sentence embeddings—often struggle with semantic nuance and domain‑specific terminology. To address these shortcomings, the authors propose a three‑stage framework: (1) instance extraction, (2) hybrid semantic matching, and (3) decision‑support evaluation.

In the first stage, both the query and the information ontology (in the case study, renewable‑energy vendor profiles) are processed to extract meaningful entities, termed “instances.” Unlike simple token extraction, the algorithm quantifies inter‑object relationships using mathematical similarity measures (cosine similarity, Euclidean distance) and statistical significance tests (χ², p‑values). A set of “marked objects”—pre‑identified reference entities supplied by domain experts—is used to weight neighboring objects, thereby preserving domain‑specific semantic connections while filtering out noise.

The second stage maps the refined instance sets onto a multi‑layer semantic network. This network fuses a general lexical resource (WordNet) with a domain‑specific ontology covering renewable‑energy concepts such as photovoltaic systems, wind turbines, and battery storage. Matching scores are computed as a weighted aggregate of three components: (a) structural similarity derived from graph topology, (b) semantic distance measured within the hybrid ontology, and (c) statistical confidence inherited from the extraction phase. The output includes a match percentage, a ranking of candidate vendor profiles, and a confidence score for each match, which together support more informed decision making.

For empirical validation, the authors applied the algorithm to a realistic renewable‑energy procurement scenario. Customer requirements (annual generation capacity, site area, upfront investment, maintenance constraints, etc.) were expressed as queries, and a database of vendor profiles served as the target domain. The proposed method was benchmarked against three baselines: (i) a pure TF‑IDF keyword matcher, (ii) a simple semantic matcher using only WordNet, and (iii) a state‑of‑the‑art deep‑learning sentence‑embedding matcher. Across standard information‑retrieval metrics, the hybrid algorithm achieved an average accuracy of 0.84, recall of 0.81, and F1‑score of 0.825—improvements of roughly 12 % to 18 % over the baselines. Notably, the system excelled in handling “sparse” keywords (e.g., “battery lifespan,” “grid interconnection”), which many baseline models failed to capture. The confidence scores further enabled procurement officers to quantify risk and prioritize vendors with higher semantic alignment.

The paper’s contributions are threefold: (1) a mathematically and statistically rigorous instance‑extraction pipeline that enhances semantic fidelity, (2) a hybrid semantic matching engine that combines general lexical relations with a domain‑specific ontology, and (3) an integrated confidence‑metric that transforms raw match scores into actionable decision‑support information. Limitations include the reliance on expert‑defined marked objects and the computational overhead that grows with ontology size. Future work will explore automated marked‑object learning via meta‑learning techniques, distributed graph processing (e.g., Apache Giraph, Pregel) to improve scalability, and the extension of the framework to other domains such as healthcare and finance, accompanied by automated domain‑ontology construction pipelines.