Research On Chemical Product Recommendation Based On Attention Mechanism

With the rapid development of the chemical industry in recent years,the production and variety of chemicals has increased.However,the distribution channels for chemicals have not been optimised.This has led to a failure to recommend chemicals to those who need them,and this has also led to a decline in sales of chemicals.This is an obstacle to the development of the chemicals industry.While the chemical industry is developing at a rapid pace,the technology in artificial intelligence is also developing at a rapid pace.Recommendation systems in the field of artificial intelligence have been used in major e-commerce websites.Therefore,in order to solve the problem of chemical sales,this paper tries to use the method of recommendation systems to recommend chemicals to users and thus solve the dilemma of chemical sales.In the field of recommendation system research,most of the application scenarios are for major online shopping malls,such as Taobao and Jingdong.In this paper,we propose to design a new recommendation system model and apply it to the chemical recommendation scenario.In the model,we use an attention mechanism to solve the problem of click-through rate of chemical advertisements and thus increase the sales of chemical products.In this paper,a recommendation system model using graph neural networks is used to recommend chemicals to users.The research in this paper is as follows:(1)We propose a Deep Spatio-Temporal Attention Network for Click-Trough Rate Prediction(DSTAN)to effectively use temporal and spatial information for CTR prediction.The temporal information is the chemical ads that users have clicked and not clicked during a certain period of time,so that we can understand users’ preferences through temporal information;we consider that chemical ads on the same page as the target ad may affect users’ interest,so we add spatial information to the auxiliary information,i.e.contextual ads that appear on the same page as the target ad.In order to make the model as little affected by the noise in the dataset as possible,we added a self-attention mechanism to the model,which is used to reduce the noise in the dataset;in order to distinguish the influence of different auxiliary information on the target ad,we added an interactive attention mechanism to the model,which is used to distinguish the contribution between different auxiliary information.Finally,the heterogeneous data are fused into a coherent framework to calculate the probability of clicking on the target ad.To verify the effectiveness of the model,we put the DSTAN model into the chemicals dataset,Taobao dataset for offline experiments and found that DSTAN outperformed several of the most common methods in CTR prediction.This suggests that DSTAN is an effective CTR prediction method that can help chemical companies better understand their target audience and improve the success rate of chemical recommendations.Graph neural networks have been increasingly used in recommendation systems with good results.Therefore,in this paper,we try to use the existing graph neural network model(Knowledge Graph Attention Network for Recommendation,KGAT)to solve the chemical recommendation problem.(2)Graph neural networks are increasingly used in recommender systems with good results.Therefore,this paper attempts to apply an existing graph neural network model(Knowledge Graph Attention Network for Recommendation,KGAT)to the chemical recommendation problem.the KGAT model explicitly models the higherorder connections in KG in an end-to-end manner and uses higher-order relationships in the hybrid structure to connect two items with one or more associated attributes,which is an important factor for recommendation success.KGAT uses recursive propagation of embeddings,starting with a node’s neighbours(which can be users,items or attributes)to refine the embedding of nodes and using attention mechanisms to distinguish the importance of neighbours.KGAT is conceptually superior to existing KG-based recommendation methods that either exploit higher-order relationships by extracting paths or model them implicitly through regularisation.experimental results of the KGAT model on a chemical dataset show that good performance is achieved by applying the graph neural network model to chemical scenarios.It shows that the graph neural network is able to mine the user-chemical relationship and also shows that the graph neural network has good performance.