[Objective] We aimed to realize contextual topic identification of expert’s historical Q&A texts, in order to improve the accuracy of expert recommendation in CQA. [Methods] By combining the Labeled-LDA model and the BERT model, we made full use of the tag information to vectorize expert’s historical Q&A texts. Through dimension reduction and topic clustering, we achieved contextual topic identification and obtained the probability distribution of expert's topic interests. According to the results of topic interest excavation, we construct the topic sensitive PageRank algorithm (TSPR), and added the user quality weight to calculate the domain authority. Based on this, we proposed the TIDARank algorithm for expert recommendation in CQA. [Results] Based on the Stack Exchange public data set, the BERT-LLDA model outperformed TF-IDF, BERT, and BERT-LDA models on the silhouette coefficient (0.5756) and topic coherence (0.4766). The ACC@20 and MRR@20 of TIDARank reached 0.5807 and 0.2430 respectively, improved by 14.53% and 8.14% compared with the best-performing Bi-LSTM+TSPR baseline algorithm. [Limitations] We did not consider user activity in link analysis. [Conclusions] Based on the BERT-LLDA model, we achieved better topic clustering results for question-answering texts and improved the performances of expert recommendation in CQA.