Hybrid semantic-sequence clustering HSSC for AI-driven recruitment: pattern discovery in professional social networks

Professional social platforms such as LinkedIn provide large-scale candidate data that increasingly support AI-driven recruitment. However, the heterogeneity of professional profile data and the reliance on static keyword-based search make it difficult for recruiters to identify non-traditional, high-potential talent. This paper details the Hybrid Semantic-Sequence Clustering (HSSC) framework, a new approach using AI for the purpose of revealing career-patterns that are latent, by bringing together semantic understanding and temporal modeling of professional profiles. The framework integrates a sentence-BERT embedding of job descriptions and career summaries with long short-term memory (LSTM) embeddings of chronological job transitions. Following UMAP dimensionality reduction, six clustering algorithms were evaluated, from which Spectral Clustering was selected as the primary clustering method. HDBSCAN was used as a secondary diagnostic approach for outlier detection. We applied HSSC to extract meaning from a dataset of 10,000 anonymized LinkedIn profiles, and discovered eight clusters ranging from emerging tech freelancers to cross-industry digital strategists. From our analysis, we can conclude that the HSSC approach is capable of generating important and sometimes overlooked candidate clusters. In terms of transparency, explainability is provided by the usage of LDA topic modeling and validation by experts resulting in thematic perspectives of every cluster for recruiters. HSSC is able to strike a good balance between semantic interpretability, structural separation, and recruiter confidence in the chosen clustering method. The HSSC approach offers interpretable talent clusters along with practitioner evaluations indicating that the HSSC approach is valuable and trustworthy.