Research And Design Of Sequence Recommedation System Based On Graph Neural Network And Sequence Pattern Analysis

In the era of explosive growth of Internet information,the recommendation system has successfully evolved into one of the basic tools of information services,which can help users make reasonable choices and decisions,improve the efficiency of data processing,and effectively alleviate the problem of information overload.Traditional recommendation algorithms require additional semantic information such as user profiles and historical scores to construct recommendation tasks to predict items of real interest to users,but in practice,due to privacy policies or restrictions on anonymous access by users,most of these profiles are not directly available,and the only valid information available is the click behavior record in the current session.As a result,the session-based sequence recommendation method was born and received widespread attention from academia and industry.At present,the existing session sequence recommendation algorithm tends to ignore the noise impact caused by the uncertainty of user behavior in the feedback data,and there is a problem of inability to accurately and effectively capture the complex dependencies of items in the session sequence.In view of the above problems,the main research work of this thesis is as follows:(1)A session-oriented demand perception graph neural network recommendation model(DAAGNNSR)is proposed.Firstly,the session data with time series is constructed as a graph,and the node embedding representation on the graph is learned by introducing the graph neural network;Secondly,the extracted project features are linearly aggregated into the user’s potential demand matrix using the demand aware aggregator to automatically weaken the noise interference,and the low-rank multihead attention network is used to interact with all the project features to generate the project characterization of the demand enhancement item by item;The sequential association between projects is further analyzed by combining independent position coding again,and the generated independent position embedding is linearly integrated with the project characterization;Finally,a list of ranking recommendations is generated by the prediction layer.(2)Aiming at the problem that it is difficult for the existing session sequence recommendation method to accurately extract user preferences from noisy sessions,a graph neural network session recommendation model(CLSR-GNN)combined with comparative learning is proposed.Specifically,the model first constructs all the session data into a graph,and obtains the local embedding representation of the node through the item information on the graph neural network aggregation diagram.Secondly,the attention mechanism containing the noise filter is used to display the representation of unimportant nodes filtered out,and the comparative learning technology is introduced to set the optimization strategy to guide the model denoising learning to obtain the global embedded representation.Finally,the gating mechanism is used to assign the weights adaptively for local and global characterization,and the weighted summation generates an effective session representation to achieve item ranking recommendations.(3)Based on the above research work,a corresponding product sequence recommendation system was established.The system uses the real platform dataset in the field of ecommerce to model and carry out the corresponding functional design and model finetuning tests,and jointly provides the corresponding recommendation service for the user with the session-oriented demand perception attention graph neural network recommendation algorithm and the graph neural network session recommendation algorithm combined with comparative learning,and finally trains it into a commercially available realworld recommendation model,thereby verifying the recommendation effect of the DAAGNNSR model and the CLSR-GNN model in the actual recommendation platform.