Research On Product Recommendation Algorithm Based On Information Fusion

With the continuous development of e-commerce,product recommendation has become one of the most concerned issues in e-commerce platforms.At present,the core of the product recommendation system is how to use massive user behavior data and product information to provide users with personalized and accurate product recommendations.However,recommendation systems often struggle to achieve accurate predictions when faced with challenges such as data sparsity and feature combinations.Information fusion is an important solution,which aims to integrate multiple data sources to improve the performance of recommendation systems.This paper aims to study a product recommendation algorithm based on information fusion,which improves the accuracy of the recommendation system by fusing different features.1.Aiming at the data sparsity problem in the recommendation system,this paper proposes a recommendation model based on historical interaction behavior information and attribute auxiliary information.The model utilizes dual autoencoder to perform dimension reduction on sparse rating data,and captures the potential relationship between users and items in the same low dimension after encoding.In order to improve the efficiency of convolution neural network in extracting commodity text features,this paper weights the information before and after the text to highlight the key position information.For high-dimensional and multi-type situations in user and product attributes,this paper uses deep neural networks to achieve dimension reduction and mine deep connections between different features.Finally,this paper adopts multi-layer perceptron to realize deep learning of feature interaction,so as to improve the accuracy of rating prediction.2.Aiming at the feature combination problem in the recommendation system,this paper proposes a click-through rate prediction model based on multi-head selfattention.The model adopts grouped crossover and convolution feature compression methods to learn high-order display feature interaction information.In addition,the dropout technique optimized by the binary mask is used to reduce the complexity of the model.In the feature intersection module,this paper adopts a multi-head selfattention network to capture the feature interaction of a feature in different subspace to form meaningful high-order features.At the same time,this paper introduces the residual network into the multi-head self-attention network,which can preserve the input feature information.In summary,this paper proposes two commodity recommendation algorithms based on information fusion.Experimental results show that both algorithms can significantly improve the accuracy and diversity of product recommendations,and provide better recommendation services for e-commerce platforms.The future research direction is to further explore information fusion methods to improve the efficiency of recommendation algorithms.