Research On Sequential Recommendation Algorithm Based On Deep Learning

In recent years,with the rapid development of the Internet and mobile terminals,major Internet companies have adopted various recommendation algorithms to recommend information or items in order to seize limited user resources and promote their own products and services.Sequential recommendation algorithm,as a sub-field of recommendation algorithm field,is especially widely used in e-commerce shopping platforms.The main task of the sequential recommendation algorithm is to obtain the current sequence preference through the interaction sequence between the user and the platform,then predict the next items that the user may interact with,and provide the user with a suitable recommendation list.Currently,researchers are beginning to apply Gated Graph Neural Networks(GGNN)to sequential recommendation,but the existing research still has the following two problems.First,the items position information in the sequence,as the major feature of sequence recommendation,cannot be used well.The lack of use of item location information will result in the inability to obtain a better feature representation of sequence preference.Second,there is no effective use of the full sequence information.As an information collection composed of massive sequence information,the full sequence information contains deep-level connections between items,and has the characteristics of multiple content and mixed information.Based on the e-commerce shopping platform,this thesis proposes an Item-Fused Boosted Gated Graph Neural Network(IF-BGGNN)for sequential recommendation tasks.The main research of this thesis is as follows:(1)Aiming at the problem of insufficient utilization of position information within the sequence,a Boosted Gated Graph Neural Network model(BGGNN)is proposed.BGGNN is mainly used to capture the preference expression of sequence information.Its main idea is based on BGGNN model.In the process of converting the input sequence into a graph,it increases the self-circulation of item node information,and adopts a strategy of biased information flow.The attention mechanism obtains the long-term and short-term preferences of the sequence,integrates the internal position information of the sequence at the same time,and finally splices the sequence preference feature representation.(2)Aiming at the problem of insufficient utilization of full-sequence information,an Item Feature Extraction Model(IFEM)is proposed.IFEM is mainly used to capture the interrelationships between items from the global sequence and discover the general characteristics of the items.Its main idea is to process all the sequence information,extract the full sequence item relationship set of the items,learn the relationship between the items through the attention mechanism,and finally obtain the general characteristics of the item.(3)Propose an Item-Fused Boosted Gated Graph Neural Network model(IF-BGGNN)that integrates the general feature representation of the item into the sequence preference feature representation,and obtains a mixed sequence preference feature representation,which is used as a recommendation basis.