Research And Implementation Of Art Toy Recommendation System Based On Fine-grained Feature Intersection

In recent years,the rapid development of internet technology has driven the real economy to tilt towards the virtual economy,leading to the rapid development of the e-commerce industry.Compared to shopping at physical stores,people can browse and trade on online shopping platforms anytime and anywhere,brings benefits in terms of space,time,economy,and other aspects.However,the number of users and commodities with exponential growth brings profit to the e-commerce platform,but also brings problems.For example,users have spent too long searching for products they are satisfied with,and seasonal new products launched by merchants cannot be presented to users who need such products in a timely manner.To solve the above problems,recommendation models have emerged,shorten the time for users to find products of interest,and improve user satisfaction.Integrating recommendation models into e-commerce platforms has become a necessary means for mainstream e-commerce platforms.This article focuses on popular trendy products and develops a trendy game recommendation system based on dual fine-grained feature interaction.According to the official report,by the end of 2023,the market size of Art Toy industry will reach57.4 billion yuan.However,as a new thing,the development of the trendy game industry in China is far behind the market demand.How to select the goods of interest for the trendy game enthusiasts from the vast pool of trendy game goods has become a major problem.Therefore,this thesis conducted in-depth research on the application and promotion of feature crossover in the field of trendy game recommendation,proposed a dual fine-grained feature crossover trendy game recommendation model,and developed a trendy game recommendation system to address the shopping needs of trendy game enthusiasts.Specifically,the research included the following two aspects:(1)Firstly,after analyzing the characteristics and user characteristics of trendy products,it was found that users who purchased trendy products had a strong temporal order,and their purchased intentions were closely related to their purchasing behavior.Moreover,the characteristics of the purchased products were closely related to current trends.As a niche audience product,the user features that the Tide Play system could collect are very limited.Therefore,in order to improve the accuracy of the Tide Play recommendation model,it was necessary to pay attention to the user features collected by the system.This article proposed a new feature crossover method-dual fine-grained feature crossover.The process of dual fine-grained feature intersection was based on the fusion of explicit and implicit feature intersections,and models user and product features.During the training process,the model continuously optimized the intermediate parameter matrix,which could dynamically learn the combination relationship between different dimensions of features,undoubtedly enhanced the model’s expression ability.However,if a large number of user and product features were directly intersected,it would result in a very large number of parameters that needed to be stored and calculated during the model training process.On the one hand,attention mechanism were be used to assign high weight values to features with higher importance before the input feature crossover process,while reduced the weight ratio for features with less important features.On the other hand,residual network layers were used to avoid model degradation caused by excessive network layers.This article could demonstrate through comparative experiments with mainstream recommendation models that the recommendation model based on dual fine-grained feature crossover had significantly improved evaluation indicators such as F1 score,accuracy,and recall.(2)The recommendation model based on the intersection of dual fine-grained features mentioned above was the core of this article,and a trendy game recommendation system had been developed.First,the user and the market was fully investigated to analyze the real user needs.Second,after determining the system needed analysis and outline design,each functional module of the system was analyzed in detail,and database design was carried out to finally complete the coding of the entire system.This system could not only achieve basic functions such as user login,browsed and bookmarked products,and viewed personal footprints,but also improved the overall recommendation effect of the system through the recommendation model implemented in this article.