Research On Group Interest Modeling And Recommendation Algorithm For Agricultural Community

Agricultural informatization plays an important role in the process of agricultural modernization in our country.In recent years,the explosive growth of agricultural information resources has led to the serious problem of "information overload",which prevents agricultural users from accessing resources timely,accurately and effectively.Therefore,the current information technology is used to realize personalized recommendation service of information resources and provide information resources for agricultural practitioners to solve the problem of "information overload".By mining the community structure in the complex agricultural network and collecting the historical information documents of the community users,this research constructed the community user group interest model to predict the user interest preference,and combined with the personalized characteristics of users to provide information recommendation services,improving the agricultural information recommendation services effectively.The main work of this paper is as follows:(1)Research on improvement of deep sparse automatic encoder agricultural community discovery.Aiming at the problem that the traditional collaborative filtering algorithm has a large deviation in obtaining the nearest neighbor of the target user in the complex agricultural network,the complex agricultural network is divided into communities to obtain the community structure with close user relationship in this research.Firstly,the calculation method of node similarity matrix in complex agricultural network graph was improved.Then,the sparse penalty function in the automatic encoder was improved,and the deep sparse automatic encoder was constructed to extract features from the similarity matrix of agricultural network graph.Finally,k-means algorithm was used to cluster the extracted feature matrix,so as to obtain the community structure in the complex agricultural network,which was taken as the basis for subsequent research.Experiments showed that,compared with DBCS algorithm,Deepwalk algorithm and CoDDA algorithm,the improved algorithm in this paper improved the accuracy by 4.3%,15.6% and 5.5% respectively,and maintained about 70% on average in the simulation data set,with good stability.(2)Research on building interest model of agricultural community user groups.In view of the sparsity of data annotation in the construction of complex agricultural network user interest model.Firstly,tag recommendation and new tag extraction method were combined to label the unlabeled document set.In addition,based on Hownet algorithm to calculate the semantic similarity of community user aggregation tags,eliminating the semantic ambiguity of tags.Then used the clustering algorithm to aggregate the different types of interest of community user groups,and finally used the spatial vector method to express the group interest model.Experiments showed that compared with LDA algorithm,TextRank algorithm and LDA+TextRank algorithm,the improved text tag labeling algorithm in this paper improved the accuracy by 9.9%,6.8% and 2.3% respectively.(3)Research on mixed recommendation algorithm for agricultural community.In view of the cold start of the agricultural information recommendation and the poor quality of the recommendation.Firstly,a method based on community group interest hot tag information recommendation was proposed to complete the initial recommendation to new users of the system,alleviating the cold start problem of users.Then,based on the target user's community,a collaborative filtering recommendation algorithm based on community user relationship mining was proposed.Finally,Combined with the information recommendation algorithm based on community group hot tag and the collaborative filtering recommendation algorithm based on community user relationship mining to recommend information for users.Experiment shows,The collaborative filtering recommendation algorithm based on community user relationship mining has the best recommendation performance when the nearest neighbor is 40,The accuracy rate is 76.8% and the recall rate is 25.9%.Compared with CT algorithm,CS algorithm and UserCF algorithm,The accuracy rate increased by 16.0%,14.2%,and 10.4%,and the recall rate increased by 4.6%,5.1%,and 9.5%,respectively.When the two algorithms were combined,the diversity of recommendation was improved by 13.3% while the accuracy was guaranteed.