This research is dedicated to exploring a personalised product recommendation model for automated Q&A bots based on deep learning. Traditional product recommendation systems often rely on historical user behaviour and lack sensitivity to dynamic changes in user interests. To overcome this problem, we propose a deep learning model that incorporates natural language processing and recommender systems to capture users’ personalised needs more accurately. Our model processes the textual comments made by users and extracts deep user features \({F}_{U}\) using a bidirectional gated recurrent unit (Bi-GRU). Simultaneously, we provide a dynamic collaborative filtering approach that allows the model to more accurately represent the variations in users’ product evaluations throughout time period. The model couples textual and rating information through a shared layer, which further improves the recommendation accuracy of the model. Experimental results show that our model achieves significant performance gains on both real-world datasets, with test losses below 45 and 23, respectively. Compared with other models for rule-based reasoning, interest-based recommendation, and machine learning to generate lists, the proposed model has significant advantages in dynamic modelling and personalised recommendation. By learning hidden attention vectors in a targeted manner, we successfully improve the sensitivity of the model to changes in user interests, providing an innovative solution in the field of personalised product recommendation. In summary, our study provides an effective and innovative model for personalised product recommendation based on deep learning, which provides a useful reference for the development of the recommender system field. Future research directions can further extend the applicability of the model, consider more user behavioural features, and delve into the interpretability and explainability of the model.