Time-aware and Semantic Resume Analysis for Talent Similarity Calculation

Talent similarity calculation is an important and useful research problem for precise talent search in online recruitment. However, the semantic information and time-aware factor in the talent resume make it challenging to obtain effective talent similarity. In this paper, we proposed a time-aware and semantic resume analysis based talent similarity calculation method, termed TSRA. TSRA mainly learns the similarity between two key parts in the talent resume, i.e., skills and work experiences. The pre-trained language model is firstly employed to generate word embeddings to catch the semantic information in the resume. A greedy matching approach is adopted for skills similarity calculation to align the skill sets with different sizes. For work experience similarity, TSRA transforms the work experiences into a sequence with a time series component. Then a dynamic programming algorithm is utilized to integrate the job title embedding, duration factor, and freshness factor. Finally, talent similarity is regarded as the weighted sum of these two types of similarities. The comprehensive experimental study shows that TSRA significantly outperforms state-of-the-art methods.