Question Recommendation For User-Interactive Question Answering System

Over the past few decades,people have witnessed the rapid development of the Internet, and the coming of the new information age.However,the amount of information on the Internet is growing and being updated at an unprecedented speed.While we are in the information age,we are also in an age of information explosion.People cannot afford the amount of information provided on the Internet nor make effective usage of it.People are interfered with so much redundant information that they cannot analyze the needed information correctly nor make the right choice from the information provided.Because of this,many researchers have begun to study how to organize the millions of websites effectively,and retrieve useful information from these sites to serve people.As a result,information retrieval technology is being paid more and more attention to.Now,the most important and successful information retrieval system is the search engine.However,the search engine also has it shortcomings that even become the new source of information overload:first,the number of search results is usually quite large;second,the results are sorted linearly.Among the existing technologies,question answering and personalized recommendation are considered the possible ways to improve the quality of search results. QA systems can be classified into automatic QA systems and user-interactive QA systems. Automatic QA system depends on the semantic matching methods to get the answer. User-interactive QA system is a collaborative network,where all users are efficiently organized to collaborate on the virtual social network,and answer the questions posted by others.User-interactive QA systems have the following advantages over the automatic ones: firstly,instead of using the isolated keywords users can ask questions in natural language,and get an accurate answer of what he needs;secondly,it allows other users to answer the questions so that complex answers which need reasoning or summarizing can be provided. Personalized services,a.k.a,recommendation systems,aims to collect,organize and classify resources from all kinds of sources,and recommend relevant information to users according to the settings of users in order to meet the needs of users.Generally speaking,the personalized service differentiates the traditional passive mode service in that it takes full advantage of a variety of resources,to meet the needs of individual users as its purpose.So it's an interesting research problem on how to integrate the two technologies together.This paper does an in-depth research on how to apply the recommendation technologies to the user-interactive QA systems and investigates on how to recommend questions to users in a more reasonable and effective way considering the specialty of the QA systems.So that the users can accurately get the interested questions and be more willing to answer others' questions and answer them correctly.The main research areas in this paper are as follows:First of all,we proposed a Singular Value Decomposition algorithm combined with time weight for collaborative filtering.The algorithm accepts the matrix of users' ratings on questions as input,and leverages the gradient descent algorithm to find its singular value decomposition,which will be used directly to predict a user's rating on a question.The algorithm takes all of the known ratings as a whole into consideration,instead of considers only similar users.Meanwhile,each rating score is assigned with a weight based on the rating time,and thus the interest drift with time is taken into consideration.The algorithm effectively overcomes the sparsity problem of ratings and significantly improves the quality of the recommendation.Secondly,we proposed a balanced question recommendation method for user-interactive QA systems.The method is used to recommend new questions to the users.In this algorithm, the individual users are analyzed to get their authority levels and interest levels of different domains.The load balancing mechanism is introduced so that the most important questions are recommended first and the questions are distributed as evenly as possible to different users.This algorithm effectively improves the user's participation rate,as well as the answering rate and the accuracy of the answers.Based on the above methods,we implemented the prototype of the recommendation module in the QA system.The result of the experiment shows that the two proposed methods can improve the efficiency of the QA system.