| With the advent of the Internet era and the rapid development of computer and big data technology,people in modern society have entered an era of information explosion.The number of goods is increasing exponentially,which has brought an extremely serious problem of information overload.As an effective information screening method,recommendation system can not only effectively reduce the time and energy for users to find satisfactory products,but also explore the products that users are potentially interested in.With the increasing amount of system data,the traditional linear recommendation algorithm is gradually unable to accurately recommend users due to the disadvantages of large training cost,weak feature expression ability and insufficient generalization ability.In recent years,with the development of artificial intelligence,deep learning,as an important tool of feature extraction,has been widely used.At the same time,this paper focuses on the design and implementation of the depth recommendation system.By combing the relevant literature at home and abroad,this paper integrates the research methods of word embedding,depth factor decomposition machine and attention mechanism,and constructs axdfm algorithm;Based on the proposed algorithm,an adaptive collaborative regularization method is designed to effectively reduce the influence of over fitting;Based on the proposed algorithm,a set of commodity recommendation system is built,which can recommend users in real time and offline.The main work of this paper is reflected in the following three points:1.Construction of depth factor decomposition machine for fusion attention capsule.In view of the current situation that the depth factor decomposition machine does not consider the different impact of users’ historical behavior on users’ interests when recommending,a depth factor decomposition machine algorithm integrating attention capsule is proposed by using the attention mechanism to give different cross features with different feature weights.The algorithm calculates the attention size between different features through the attention capsule,so that different features have different weights,so as to solve the noise problem caused by feature intersection.2.Adaptive cooperative regularization method.In view of the problem that the deep learning algorithm is easy to lead to the over fitting phenomenon due to the large amount of training data,too complex network structure,too many parameters and other factors in the training process,an adaptive cooperative regularization formula is added to effectively reduce the over fitting phenomenon of the algorithm without affecting the performance of the algorithm,A comprehensive experimental comparison with other mainstream regularization methods is carried out.3.Design and construction of recommendation system.Based on big data components such as Spring,Scala and spark,a complete commodity recommendation system that can effectively recommend users is developed.It is mainly composed of offline recommendation system,real-time recommendation system and business system.The system has the basic functions of viewing commodity information,scoring commodities,recommending potentially interesting commodities and associating similar commodities,it can recommend users accurately and quickly. |
