Design And Implementation Of Dual Learning Based Top-n Recommendation System

The vigorous development of Web 2.0 technology has brought complex and diverse online information to the Internet.At present,users are facing a serious problem of information overload.Recommendation system has been developed to be the most commonly used means to solve this problem,which can help users obtain useful information of their own interest in mass informa-tion.The existing recommendation systems have the following problems:1)The effectiveness of the recommendation system depends on the richness of user behavior data in the system,and there is a deviation in the common rec-ommendation algorithms,when the user does not interact with the item,treat it as a negative sample.2)At present,the recommendation system focuses on the item recommendation task,ignores the marketing task of finding potential interested users for explosive items,and does not consider the relationship be-tween the recommendation task and the marketing task.In view of the problems in the above recommendation system,a top-N recommendation system based on dual learning is proposed.The main work of this paper is as follows:1.A top-n recommendation system based on dual learning is proposed.The system has the following advantages:1)Considering the needs of both client and business,users need the system to provide a good personalized expe-rience,and businesses need the system to provide an effective marketing expe-rience.Introduce "multi-objective task" into the system,and achieve the above requirements simultaneously through a unified algorithm.2)In recommenda-tion engine,through the modeling of user's behavior,the user's interest is mined to bring better personalized experience to users.In this paper,the functional modules and databases of the system are designed in detail from the perspec-tive of user requirements and functional requirements,and two algorithms are proposed to solve the problems in the recommendation system.Finally,the overall implementation of the system is completed,and the specific functions are tested in detail.2.Aiming at the problem that both client and business needs need to be considered in the algorithm,a multi-objective ranking algorithm based on dual learning is proposed.The algorithm aims to train the algorithm on unlabeled data sets by establishing the dual relationship between product recommenda-tion and marketing scenarios,and reduce the dependence of recommendation problems on labeled data sets.At the same time,two different scenarios can be considered unified,and data closed-loop between multi-objective tasks is estab-lished.Experiments on multiple datasets show that the proposed algorithm can improve the recommendation effect in multi-target scenarios from unlabeled datasets.3.Aiming at the need of optimizing user's personalized experience,a user behavior sequence extraction algorithm based on convolutional neural network is proposed.The algorithm embeds user's behavior into a fixed dimension vec-tor through convolutional neural network to realize the description of user's in-terest.Furthermore,a Pairwise-based deep factorization network is proposed.By introducing paiiwise loss into the deep factorization model,the effect of the model on ranking problem is improved.The validity of the model is verified by experiments on several data sets.