Research For News Recommendation Based On Fusion Of Interests And Emotion

With the rapid spread and development of the Internet,more and more users are choosing to access information online.The explosion of information force users to spend more time on identifying the information they want,or even difficult to find the information they want.In this situation,how to determine the user's interest and preference accurately,and recommend the required information to users,including to improve efficiency of recommendation and users' satisfaction,which have focused in the field of recommendation research.Most of the current recommendation systems based on traditional recommendation algorithms such as content-based and collaborative filtering.The meta data for user interest model is generally obtained from user's historical data indirectly.The description of user interests is generally only divided into two types: "interest" and "no interest",the accuracy of information recommendation is also affected by the large granularity of analysis.In order to solve the above problems,after analyzing the traditional recommendation algorithms of content-based and collaborative filtering,this paper proposes a measurement of reading interest based on emotion,in which,the consideration of emotional space is fused with recommendation systems.In details,the eye tracking data is collected and analyzed by the model of generalized regression neural network,and then,the measurement of users' emotions are used to describe users' interests.The main work of this paper is shown as follows:(1)A survey was conducted on the major technologies and related user models in current news recommendation.By analyzing the results of survey,we propose a solution based on fusion of interests and emotion.(2)Based on the idea of fusion of emotion and interest,we propose an architecture of detecting interests based on emotional space.This system consists of interest prediction model based on emotion space is proposed.The system consists of collecting eye movement information,quantifying emotional space,fusing emotion and interest,and detecting interest.The module of detecting interest is the core of proposal,which is based on the eye movement tracking and emotion quantization.(3)We propose an approach of describing user interest by emotional space.Through analyzing typical emotion models,we build a two-dimensional PA model,in which,the three-dimensional PAD emotion model is selected as a base,and D dimension is removed after analyzing the effects of each dimension with reading interest.Furthermore,we formulate mapping rules from emotion to interest and classifies reading interests based on emotional space.Through analyzing relations between eye movement and emotion,20 types eye movement data indicators such as average value,maximum value and frequency form are selected as the impact factor.And then,the generalized regression neural network is used to construct the user interest model.(4)An interest detection model is built,and evaluated by analyzing experiment results.We design two experiments to evaluate the model of quantifying emotional space and detecting user interest respectively.The users' data of eye-tracking and emotional space collected through experiments are analyzed,and used to quantify users' emotion in PA emotional space proposed by us.Based on the quantifying results,we compare the predicated value with its actual value by experiments,the results show that our proposal has good applicability.(5)A news recommendation system fused the emotional space and interest is built by combining a content-based recommendation algorithm.The results show that the importance of user interest classification based on emotional space to recommendation system,which provides more accurate services than traditional recommendation systems.The recommendation results show the effectiveness of the fusion recommendation system based on fusion of interests and emotion is effective in accurate recommendation.