This article proposes an intelligent customer service knowledge graph construction method based on BERT model and graph inference to address the problems of inconsistent information and low processing efficiency caused by multi-source heterogeneous data in customer service systems. The aim is to improve the customer service system’s ability to process various heterogeneous data such as text, images, and logs, and achieve efficient knowledge integration and intelligent response. Firstly, the BERT (Bidirectional Encoder Representation from Transformers) model is used to semantically encode customer service text data such as user inquiries, feedback, complaints, consultation records, and conversation logs through a bidirectional Transformer architecture, and entity extraction is performed using a Conditional Random Field (CRF) model. Subsequently, rule-based matching methods are used to preliminarily correlate heterogeneous data from different structures and formats. Graph embedding and relationship classification models are used to further explore the deep connections between the data, completing the fusion of data and the construction of a complete knowledge graph. Finally, with the help of knowledge graph reasoning technology, the relationships and contextual information between entities are fully utilized to achieve accurate matching between user questions and the knowledge base, generating efficient automatic responses. The experimental results show that this method not only has high knowledge acquisition efficiency and excellent automation service capabilities in dealing with complex data environments in customer service systems, but also can effectively handle multi-source heterogeneous data (including data of different structures and types), achieve efficient fusion and correlation of multiple formats of data such as text, images, and logs (when text and log data are fully integrated, the data fusion accuracy reaches 90%, and the knowledge acquisition efficiency is also improved by 15%). This study is based on the technology of multi-source heterogeneous data fusion and knowledge graph reasoning, which can significantly improve the data processing efficiency and response accuracy of intelligent customer service systems, and has wide application value.