Research On The Method Of Answer Sorting And Migration Learning In The Community Question Answering System

With the continuous development of Internet technology makes people access to knowledge,the way to solve the problem becomes more and more convenient.Traditional search engine companies,such as Yahoo,Google and so on for the growing number of Internet users to provide a more convenient way to obtain information,the user can search the dialog box enter the relevant keywords to quickly get the information they want.But with the popularity of the Internet and the Internet's own content is constantly enriched,people get the answer at the same time,also get the best answer to the convenience put forward higher requirements.Based on the community Q&A with personalized service to make up for the traditional search engine technology deficiencies and thus more and more attention by the various Internet companies.Community Q&A system is a new knowledge sharing model,through the user to submit questions and answers,the community has accumulated a large number of question answering pairs.When users submit new problems,how to sort,to provide users with accurate answers to the sequence,is an important part of community Q&A system.The traditional ranking algorithm mainly uses the supervised learning method to construct the sort model,which needs to train the model by a large number of artificial mark data.At present,scholars have put forward a lot of methods based on supervised sequential learning and applied well in real life,for example,sorted support vector machine,which is a typical representative of supervised learning based sorting algorithm.Through lots of annotation data,Input to the specified learning machine,and then automatically train to get a sort model.The method based on supervised sorting often needs considerable scale data to ensure the reliability of the training model,but in the actual environment due to the lack of label data.When the data is lacking,the reliability of the supervised sorting algorithm will be reduced accordingly.Training in a particular field sorting model,in the new field often can not get good results.And the Internet data update soon,before the data marked with the passage of time can not adapt to the current model of training.In order to solve the problem of insufficient marking in practice,the traditional sequential learning method is improved by using the idea of migration learning.Based on the feature selection.the migration learning ranking algorithm based on feature selection is proposed.The source domain and the target domain share the low-dimensional feature representation,and the user's multiple interest is the shared feature of the source domain and the target domain.By analyzing the characteristics of community Q&A system,we can observe that there are many tags based on user behavior.The feature-based migration learning method is used to integrate these user characteristics into the feature space.The feature-based migration learning sorting algorithm is optimized by selecting user tags and user behavior tags with specific values in the community.Such as the domain of the respondent's domain of expertise,a questioner may be good at multiple fields(such as tennis and badminton).The feature is represented by a Boolean type in the eigenvector,good at 1 is not good at 0.Then this feature in the badminton and tennis boolean types are 1,that is,this feature can be used as badminton and tennis are two different types of common features to use,thereby improving the sort of learning methods.The experimental results show that the algorithm can effectively improve the ranking of the answers.