Research On Personalized Recommendation Algorithm Based On Mahout's Interest Distribution Hybrid Model

The rapid development of Internet technology is changing the whole world,all walks of life are changing rapidly with the mode of the Internet plus.The development of science and technology has led to an explosion of information,and it has become increasingly difficult for people to acquire the knowledge they want in the face of massive amounts of data.To solve the problem of information overload,many scholars propose to use the recommendation system,which can provide personalized services for people and dig up resources that people want.The personalized recommendation service will give a list of recommendations more in line with the user's interests based on the user's preference behavior.For each personalized recommendation algorithm,there are advantages and disadvantages.In this era of multi-informatization,one recommendation algorithm can not solve all problems.The successful application of the personalized recommendation algorithm in the industry promotes scholars' theoretical research on the algorithm.The main problems faced by the algorithm are data sparsity,cold start and so on.The sparseness problem is also the main factor affecting the accuracy of the recommendation algorithm.For this problem,experts and scholars also improve the algorithm through various methods to improve the recommendation quality of the algorithm.The main research purpose of this paper is to improve the accuracy of the personalized recommendation algorithm for predicting the score and the quality of the recommended list.Aiming at the problem of data sparsity in the current recommendation field,an in-depth analysis of the user-based collaborative filtering and implicit semantic model is conducted,and the advantages and disadvantages of the two are compared.Combine the Gini coefficient in economics to measure the interest distribution pattern of the user and propose a Distributed mixed recommendation model.The model firstly preprocesses the user's behavior data and the classification information of the project,and then divides the user group according to the Gini coefficient of the user's interest distribution to narrow down the search for nearest neighbors,and uses the data sparse insensitivity of the implicit semantic model to pre-populate.The user scores data items,thereby searching for more accurate neighbor users,improving the prediction score of the personalized recommendation algorithm,and improving the recommendation quality of the algorithm.In the end,this paper designs a detailed comparison experiment for the proposed hybrid model.The experimental platform adopts the Mahout algorithm framework and the java language to implement the mixed model.It compares with the traditional single recommendation algorithm,respectively,in the recommended prediction score and recommendation.The quality of the list was compared between the two recommended stages and the experimental results demonstrated the effectiveness of the hybrid model.