Research On E-commerce Recommendation Algorithm Base On Data Augmentation Attention Mechanism

Due to its huge number of users and goods,the field of e-commerce has become one of the "heavy disaster areas" for information overload.Therefore,excellent ecommerce recommendation algorithms have become the supporting technology and necessary weapons for many large-scale e-commerce platforms.Improving the performance of the recommendation algorithm in the e-commerce field can not only prevent users from getting lost in the huge forest of commodities,but also greatly increase the turnover and revenue of merchants and platforms.In traditional ecommerce recommendation algorithms,most of the available features have no obvious spatial or temporal relationship,which makes it difficult to apply methods such as recurrent neural networks in this field.The core idea of session-based recommendation algorithm is to divide the user's historical behavior sequence into the form of conversation,so that the methods of natural language processing and sequence data processing can be introduced into the recommendation algorithm.We found that in the original session data in real ecommerce scenarios,there are similar phenomena such as insufficient number of sessions,short sessions,and uneven session distribution;Most existing session-based recommendation algorithms do not strictly distinguish between different types of data features,resulting in a large number of basic attribute features that have not been fully utilized.At the same time,models based on recurrent neural networks often only focus on the current session while ignoring the timing information implicit between sessions.To salve the above problems,this paper improves the existing session-based recommendation algorithm at the model improvement level and the data enhancement level,and proposes a session recommendation algorithm based on the attention mechanism and four data augmentation methods.method.The main research work of this paper is as follows:1.Apply the data augmentation methods in the field of image processing and natural language processing to the data augmentation of e-commerce systems,and propose four session-based recommendation algorithm data augmentation methods:EDA,WS,seq2seq,and DeepWalk,which solve the problems of insufficient session data,short sessions,and uneven distribution.Based on the original method,these methods have been improved for the e-commerce environment,making them more suitable for user session data enhancement.Experimental results show that these methods can effectively improve the quality of the original session data,thereby improving the performance and scalability of the session-based recommendation algorithm at the data level.2.In order to solve the problems of insufficient utilization of basic attribute features and limited expression of conversational interest in the original sessionbased recommendation algorithm,we proposes an Attentive session-based Deep Factorization Machine(ASDeepFM),which is composed of a DeepFM part and an attentive session part.This article analyzes the data features in the e-commerce recommendation scenario,and divides the available features into the basic attribute features of users and products,and the historical behavior sequence features of users.We use the DeepFM part to process the basic attribute characteristics without time and space relationship,and propose an attention session module that specifically deals with user behavior sessions based on the method of multi-head attention mechanism.This module includes session segmentation,intra-session attention extraction,and intersession attention.It can make full use of sequence information in historical behavior to mine user interest and achieve better recommendation results.Finally,we connected the DeepFM part and the attentive session part in series,and proposed an Attentive session-based Deep Factorization Machine(ASDeepFM).Experimental results show that the ASDeepFM model proposed in this paper has achieved significant performance improvements on both public data sets.