The Application Of Node2vec Model In Recommendation System

With the rapid development of the Internet and the continuous improvement of the quality of life,consumers’ purchasing interests tend to be diversified,and obtaining a more accurate recommendation system has become an important magic weapon for ecommerce platforms to improve users’ click-through rate and order rate.The research of the system can reduce the cost of enterprises and improve the commercial profit and the value of the e-commerce platform,so the exploration of the system is of great significance for the future development of e-commerce.In addition,there are two problems in the application of traditional recommendation systems at present.On the one hand,the complexity of the recommendation model is high,the time consumption cost is high,and it is difficult to meet the requirements of real-time recommendation under the background of 10000 or even 100 million users.At the same time,the interactive information in the recommendation system is sparse,which is difficult to capture the effective information between different users and different commodities,and it is easy to lead to dimension disaster in high-dimensional space;On the other hand,the increasing number of new products and new registered users may cause the cold start problem of the system.Therefore,this paper conducts an in-depth exploration of the recommendation system for the above two problems,improves the original model on the basis of cutting-edge algorithms,and further improves the accuracy of the recommendation system.The main work includes the following three parts:First,Innovative application of node2 vec model to improve the recommendation results.In order to reduce the time consumption of the model and reduce the complexity of the model,this paper uses node2 vec shallow neural network model with the help of network structure data method.This model is often used to predict edge relationships between network structure data nodes,and this paper extends its application to recommender systems.This model has two major advantages,one is that the model is simple and the complexity is small.The node2 vec model only contains one hidden layer,which greatly reduces the complexity of the feedforward neural network model and the recurrent network model in the traditional recommendation system,reduces the time consumption of model training,and improves the recommendation efficiency of the recommendation system;Second,the model is flexible and the training is controllable.The node2 vec model is an improvement based on the Deep Walk basic graph embedding model.This model improves the random walk of the Deep Walk model in the network.By adding two walk parameters,the walk between nodes is transformed into a biased way.Walking makes the model more inclined to capture the structure and the homogeneity in the network structure,and the resulting recommender system is more flexible and adaptable.Second,adding side supplemental information to address cold start issues.On the one hand,each user only interacts with some merchants or products,the information contained is sparse,and the available content is less;On the other hand,There is no interaction between new products or merchants,so,at this time,it is necessary to add the side information of the products and merchants as supplements to complete the targeted recommendation for such new individuals.The specific implementation method in the model is to add the side information as a new dimension feature to the corresponding vector,which not only improves the problem of less interaction and lack of effective information,but also makes up for the defect that the newly added individual has no interaction.The cold start problem in the model is better solved.Finally,an example is verified to analyze the advantages of the model.In this paper,the node2 vec model and other models are applied to the actual data of Mei Tuan Takeaway for exploration,and the accuracy between different models is compared.It is found that the node2 vec model with side information added has the highest accuracy.By improving the performance of the recommendation system,it can better capture the differences between different merchants,and at the same time obtain similar characteristics between similar merchants,which has significant advantages and greatly compensates for the existence of the word2 vec model and the Deep Walk model in the recommendation system.which enhances the flexibility and practicability of the recommender system.On the whole,this paper verifies the effectiveness and practicality of node2 vec model in the recommendation system to a certain extent,and opens up new ideas for the application of network structure algorithm in the recommendation system in the future.Secondly,the conclusion that supplementing side information can indeed enhance the overall expression of the model provides a new method for the solution of cold start in the future.