Intelligent Assessment Method for Enterprise Employee Performance Based on Multi-Source Behavioral Data

Abstract

The increasing complexity of digital enterprises and intelligent industrial systems has led to the large-scale generation of heterogeneous behavioral data collected through interconnected sensing and communication infrastructures operating in complex electromagnetic information environments. In such scenarios, traditional enterprise performance evaluation remains constrained by fragmented data sources and subjective assessment, making it difficult to capture employees’ actual contributions. This study proposes an intelligent performance assessment framework that integrates multi-source behavioral data by consolidating enterprise information system logs, collaboration communication records, and project management system data. A behavioral feature system is constructed across three dimensions, including task execution, knowledge contribution, and collaborative interaction. Furthermore, a hybrid model combining temporal convolutional networks with an attention mechanism is developed to adaptively learn behavioral patterns across different job categories, while a state influence coefficient is introduced to compensate for assessment deviations caused by short-term fluctuations. Validation was performed using six-month longitudinal data from R&D and sales departments of an internet enterprise, including 168 R&D engineers, 49 test engineers, and 156 sales specialists with 3.897 million behavioral records and an average of 21,500 valid records collected daily. Experimental results demonstrate that the proposed method improves assessment accuracy by 29.8% compared with a linear regression model using identical behavioral features while achieving a Gini coefficient of 0.19. By incorporating temporal decay characteristics of behavioral information and accounting for heterogeneous work patterns across specialized positions, the proposed framework enhances the reliability of intelligent decision-making and provides a scalable methodology for multi-source information fusion in data-driven management systems operating under complex communication and sensing environments.