Multi-label Learning Algorithm And Its Application In Product Evaluation And Scoring

In traditional supervised learning framework,each instance only corresponds to one label.But in real life,an instance may be described by more than one label.For example,a document may belong to several pre-defined subjects,a picture may contain multiple semantics simultaneously and a gene may have multiple functions at the same time.The aim of multi-label learning in machine learning is to learn with the label ambiguity.Multi-label learning has attracted much attention at present.However,for some problems with more ambiguity,multi-label learning can't deal with it directly,and label distribution learning is proposed to solve it.Label distribution learning is an extension of multi-label learning.In theory,label distribution learning has more usage scenarios.The key to research the multi-label learning and the label distribution learning is to fully exploit and utilize the correlation between labels.Based on this,this paper mainly studies the correlation between multi-label learning and label distribution learning.This paper mainly covers from the following aspects:Firstly,multi-label learning algorithm based on association rules.If the relationships between different labels are ignored,the multi-label learning problem will be transformed into multiple single label learning problems.The information of the connections of labels will be lost and we may not obtain better classification results.To solve this problem,this paper puts forward the algorithm of association rules mining association between labels.The multi-label datasets get modified by the association rules.Considering the actual situation,the correlation between different labels only exists in some sub datasets.The multi-label datasets can be more reasonably modified to obtain a better classification result.The experiment results illustrate the effectiveness of proposed algorithm in dealing with multi-label learning problems.Secondly,label distribution learning algorithm based on label correlation.At present,the literature that dealing with the learning with ambiguity can be separated from two aspects.One is that there is a priori knowledge between different labels.The other is to build different models to calculate the correlation between labels.Unfortunately,most of the algorithms are applied to the multi-label learning framework and are less used in the label distribution learning.To solve this problem,this paper measures the correlation of two labels by measuring the distance between the corresponding columns.Finally,L-BFGS algorithm is used as the optimal algorithm.The experimental results show that the algorithm proposed in this paper have good performances.Thirdly,the application of commodity evaluation based on multi-label learning algorithm.Commodity evaluation is a widely researched topic under the background of recommender system.This kind of problem is that a customer scores a product and the system forecasts and recommends the highest scores to other customers.The multi-label learning algorithm can be applied to the commercial evaluation.Several commonly label distribution learning algorithm and multi-label learning algorithm are compared with it.The results show that this method has a higher classification accuracy in commodity evaluation.The result demonstrates the effectiveness of the proposed algorithm and extends the multi-label learning in practical application.