An XAI-Driven Linkedin Profile Optimization System Using Sentence Embeddings and Semantic Clustering
DR. S. Gnanapriya1, Aisha S2
1Associate professor, Department of Computer Applications, Nehru College of Management, Coimbatore, Tamil Nadu, India.
ncmdrsgnanapriya@gmail.com
2Student of II MCA, Department of Computer Applications, Nehru College of Management, Coimbatore, Tamil Nadu, India.
aishashamoon99@gmail.com
Abstract
This paper presents LinkedAI Pro, a computational framework designed to advance the field of career analytics through the application of high-dimensional vector space modeling and Transformer-based Natural Language Processing (NLP). Current recruitment technologies often suffer from "keyword cold-start" problems, where rigid matching algorithms overlook qualified candidates due to linguistic variance. To mitigate this, we propose a bi-encoder architecture utilizing the all-MiniLM-L6-v2 model to transform unstructured professional narratives into 384-dimensional dense vectors. By shifting the analytical focus from lexical frequency to semantic intent, the system enables a more robust calculation of document alignment through PyTorch-accelerated cosine similarity measures.
The core of the proposed system is an Explainable AI (XAI) attribution engine and an unsupervised learning module for thematic gap detection. We utilize K-Means clustering to partition target job descriptions into latent domain pillars, identifying experience voids where candidate embeddings fall below a similarity coefficient of $0.45$. This transparency is augmented by a sentence-level attribution layer that isolates the specific professional evidence used to justify an AI-driven match. This dual-layered approach ensures that the model's outputs are not only predictive but also prescriptive, providing actionable insights into industry transferability and semantic SEO optimization for digital professional personas.
Experimental validation of the framework includes the introduction of "Career Velocity," a longitudinal metric that quantifies professional transformation by measuring the semantic drift between historical and contemporary career datasets. When integrated with lexical diversity scoring and impact-oriented regex parsing, the system generates a weighted probabilistic model for interview success: $P_{match} = (SEO imes 0.45) + (Lex imes 0.25) + (Impact imes 0.30)$. Our findings suggest that this multi-metric approach significantly enhances the granularity of applicant tracking, providing a scalable engineering solution for cross-domain skills mapping and document branding alignment in the modern labor market.
Keywords:- Career analytics, Cosine similarity, Explainable AI (XAI), K-Means clustering, Natural language processing (NLP), Sentence transformers, Vector space modeling.
