Recom Mender System Based On User Context

The research methods of the recommender system is to build model about users interestand to recommend items to the potential user base which may be interested in it by miningthrough historical behavior records on users, and recommendation list of items plays animportant supportive role on the user while making decision about items-selecting. Therefore,to a certain extent, it decreases the impact of overloading-information on the user’s access toinformation. What traditional recommended systems mainly research is that analysis andmodel historical behavior records of users, and recommend the potential user base forindividualized goods, regardless of the factors of user and the context, such as time, mood,location and other information. But in recent years, with the rapid development of informationtechnology, the research and application of the recommendation system has made greatsuccess. Many researchers started to pay attention to and research the influence of contextinformation on the recommended system performance. These studies are mainly aboutrecommendation system which is based on time, location, emotion and social net-work, andso on. However, a large number of studies have only investigated the impact of the contextualfactors on the recommendation system. In fact, many of the contextual factors on therecommended system are interrelated. Therefore, in the field of recommender systems whichare based on contextual information, there are still a number of important issues that needstudy, addressing.Through its in-depth studies of existing user behavior data, this article includes theclassification of users’ emotion, the time-varying change of users’ mood over time, as well asthe impact of users’ emotion and time on the users’ interest preference This articles mainresearch are as follows:1. Discusses time-varying effects of the emotional characteristics of users andintroduces the concept of user sentiment shift. Users’ emotions can show somecontinuity over time, but the duration is different, because different users will havedifferent performance. At the same time, takes over the classification of users’ moodand divide the users’ emotions into positive emotions, negative emotions and theothers. When users are in different emotional states, reactions on items.2. Studies similarity between items based on a particular emotion, by calculating thesimilarity of emotions, it tries to correct the inaccuracy caused by the similaritybetween items by using the traditional method for solving the emotions caused byerrors, it turns out that accuracy of similarity is more accurate than ever before. On this basis, according to a particular emotional state, we are able to recommend to theuser the items vary, users’ emotions can affect the severity of ratings of items, andusers’ emotions can affect the choice of rating items which he might like. Later,taking item similarity based on particular emotions as a parameter, we add it touser-article in the matrix to form a three-dimensional matrix, so as to predict theusers’ score by decomposing singular values of a matrix, it is JMF_MS.3. Discusses the influence of time factor on recommender systems, users’ previousbehavior towards items has a greater impact on their current predicating behaviorthan their recent behavior because users’ preferences for interest will also vary withtime, and the less the time discrepancy is, the more similar they will be. Based on theinfluence of time factor on recommendation systems, we put forward improvedalgorithm based on a particular emotion of matrix singular value decompositionmethod. This improved algorithm JMF_MS_MT takes the time factor into accountthe time factor. Through a time decay function, the algorithm analysis predictivescoring based on a particular emotion. Because users’ emotions and preferences forinterest are the same and they have shown continuity over time, that is, the effect oftime on them are almost unanimously, so it is reasonable to add time decay functionfor correcting, the results also prove the correctness of this inference.4. Testes the algorithm JMF_MS and its improved algorithm JMF_MS_MT, as well astwo algorithms does not specify a particular emotional characteristics of algorithmsand matrix singular value decomposition on the same test data set and analyzes theresults. Experimental results shows that JMF_MS_MT than JMF_MS increase in therecommended accuracy as well as the recommended list sorting accuracy to a certainextent, this also shows that reasonable to consider a variety of contextual factors willbring more return of the high recommended performance.5. Studies a recommendation system composition and a recommendation engine overallarchitecture, on this basis, and combines the first few chapters of the work content,we design a prototype of the recommendation system based on user emotional factorsand time factors, and implement the recommendation system based on open sourcethe MyMedia recommendation system framework.