Optimisation of e-commerce personalised recommendation algorithms driven by neural network models

With the rise of major e-commerce, how to make more customer groups choose to buy items in their own websites is the goal that major e-commerce platforms have been relying on. Therefore, a set of personalised recommendation system that can intelligently explore customers’ needs comes into being. In this paper, a graph neural network model is used to sort out the multi-path fusion neighbourhood relationship among three objects: user, product and query. The utility matrix is established and the collaborative filtering algorithm is used to derive the user’s preference situation for commodities. Subtractive clustering is combined with fuzzy C-means to obtain the clustering centre of gravity and cluster e-commerce users. Graph neural network is introduced to ensure that the data sparsity of the user dataset is within a reasonable range. The practical application effect of the model is evaluated through simulation experiments and empirical analysis, respectively. In this paper, according to the age of the users, the users are clustered and analysed, and three clustering centres of gravity are obtained, which are (3.16, 32.73), (45.35, 40.25), and (14.03, 52.89), so the users are classified into three clusters, and the analysis of simulation experiments is carried out. The training effect of this paper’s model is fitted, and the adjusted R² = 0.8292, which shows that the accuracy of personalised recommendation is high. Meanwhile, comparing with other algorithms, this paper’s method reaches a recommendation satisfaction level of 100% when the number of learning times is 60, which is significantly better than other algorithms.