Using online information resources to build knowledge bases to provide knowledge answering services would help auto companies or third-party platforms to gain competitive advantages. Therefore, a construction plan of automobile maintenance expert system based on knowledge graph was proposed by integrated modelling. In terms of the entity recognition algorithm, the BM LSTM (Boyer-Moore Long Short-Term Memory) algorithm was proposed by integrating hidden Markov model, Conditional Random Field (CRF), Bi-directional Long Short-Term Memory (BiLSTM), BiLSTM-CRF and Lattice LSTM, which improved the accuracy index F1-score. In terms of the text quality evaluation algorithm, a secondary text quality evaluation system was designed. It evaluated the matching quality of the problem based on the word toolkit Synonyms and Levenshtein Distance algorithm. And it evaluated the quality of the answer text based on the TF-IDF (Term Frequency-Inverse Document Frequency) similarity algorithm and centered on completeness, accuracy, reliability, and argument strength. Finally, experiments are carried out on the proposed model and algorithm to prove its effectiveness.