Design And Implementation Of Internet Technical Article Recommendation System Based On Hybrid Recommendation Model

With the development of science and technology,the numbers of Internet users is increasing,and the generation of information is growing exponentially,and the recommendation system has also ushered in major challenges.The development of science and technology has led to the prosperity of the software industry,and more and more people have joined the Internet industry.The iterative speed of Internet technology is very fast,and relevant staff need to learn all the time.With the increase of the technical personnel and the rapid development of technology,there are more and more technical article platforms in the market,and a large number of articles flood in,Therefore,the functional implementation recommended in this article can help the platform to improve the user experience.So that users can quickly find the content they are interested in and better retain users.At present,there are many mainstream recommendation models,which are mainly divided into single model and hybrid model.In the process of industrial practice,the data scale of the article is large,and a single model can not effectively mine the deep information of the article and can not meet the needs of users.The hybrid model is to mix multiple single models according to the actual application scenarios.The use of mixed recommendation can effectively improve the accuracy of recommendation,but at present,in the aspect of text mixed recommendation,most of them simply extract keywords according to the word frequency in the article and calculate the weight of keywords as the feature weight of the article.but can not accurately extract the weight of representative keywords in the article,at the same time,the time complexity of the hybrid model is larger than that of the single model,and the running speed is slower.In real use scenarios,when using a hybrid model for recommendation,the problem of slow running needs to be solved.In view of the above shortcomings,this paper studies and implements the technical article hybrid recommendation system based on the industrial big data open source framework of Spark.The whole system can meet the development and use of industrial data level.While recommending the article,it improves the calculation of the feature weight of the article,and can achieve better recommendation results.The main work of this paper is as follows:1.The weighted calculation scheme of feature weight in this paper is given.The calculation of the feature weight of the article mainly includes the keyword weight,the article word vector and the article ID,.In the process of calculating the keyword weight,on the basis of the traditional TF-IDF,the adjacent relation network between words is constructed by Text Rank,and the weight is calculated by combining the TF-IDF weight value and the Text Rank weight value.Through experiments,it is proved that after the improvement of feature weight,the accuracy of the article recommendation,recall rate,F1-Score and other evaluation indicators have been improved to a certain extent.2.Build a hybrid recommendation model based on Spark industry grade.The hybrid recommendation model is divided into two stages: recall and sorting.Collaborative filtering based on ALS model and recommendation based on Word2 Vec content model are used in the recall phase.In the sorting stage,the deep learning model is not used,but the machine learning scheduling model based on factor decomposer(FM)is adopted.based on good feature engineering,even machine learning can have a good recommendation effect.3.Implement and test the article hybrid recommendation system.Realize the technical article recommendation system composed of user login and registration,crawler module,page display and big data recommendation module,in which the back-end uses the Spring Boot+Spring Data Jpa framework to achieve business functions,and uses mysql to store business data.In the implementation of big data recommendation module,hive and hbase,the most popular big data technologies in the market,are used for data storage,in which hive is mainly used for offline data storage to synchronize the information of business database mysql.Hbase stores information such as the recommendation list of articles and user portraits.The final calculation results are written to Hbase,through the process of Spark preprocessing,analysis and modeling of the data in hive and transmitted to the display page through Hbase.Complete the overall function of the system.At the same time,the main functional interfaces of the system are tested.In order to alleviate the problems of cold start and data sparsity,the system designs effective functions on the historical data of users who are not logged in and do not operate when they log in,as well as the historical data that users have operations.it further alleviates the problems of cold start and data sparsity.