Personalized Recommendation Methods Based On User Interest Mining

In recent years,the content of the Internet has greatly been enriched due to the rapid expansion of Internet applications,yet a serious information overload problem also comes with it.It is difficult for users to accurately find their desired information from the massive information on the Internet.However,personalized recommendation systems which can extract valuable information from a large amount of complex data have been widely used in recent years.It can solve the problem of information overload and has become one of the indispensable technical application in the Internet age.The design of personalized recommendation system faces many challenges in the application process,such as user interest modeling,timeliness modeling,cold-start user,data sparsity,interpretability and so on.In order to cope with these challenges,researchers have proposed a lot of solutions.However,the modeling algorithms of user interest merely with a technical perspective cannot truly reflect the general developing process of user interest.Personalized recommendation system predicts whether a user likes a certain product based on the user interests,and the quality of the model of user interests will directly affect the performance of the recommendation system.Interest is an intrinsic feature of human beings.Studies in social psychology have shown that human interests have specific law of development.In this paper,we model the dynamic developing process of user interest from the perspective of its law of development.We mainly solve the problems of user interest modeling,interest transfer prediction and cold-start user recommendation in the personalized recommendation system.To summarize,the main work and major contributions of this paper are as follows:1.Temporal Recommendation via Modeling Dynamic Interests with Inverted-U-CurvesInteraction log of a user' s interaction with system contains the user's interests.Existing studies have generally considered that the more timely the temporal information generates,the higher effectiveness will be gotten in rating prediction.Existing user interest models usually use monotonically decreasing function to calculate timeliness of information.However,our preliminary investigation shows that the interest pattern of a user is personalized and the evolution of a user interest is a nonmonotonic process which can be divided into two stages:rising stage and declining stage.In this paper,we propose a new recommendation framework called SimIUC.Under SimIUC,we can effectively identify multiple user interests from his or her temporal interaction log,and adapt the inverted-U-curves to model user interest patterns to reflect the dynamic evolution process of user interests.Then,we can predict the interest evolution trend of a user in the future and make recommendation.We systematically compare the proposed SimIUC approach with other algorithms on the dataset of Movielens and Netflix.The results confirm that our new method substantially improves the precision of recommendation.2.Sequential Recommendation for Modeling Interest-Transferring via High-Order Markov ModelOur preliminary investigation shows that user interests have two characteris-tics.First,user interest changes significantly in the process of user' s interaction with system.Second,the duration of a particular interest and the change fre-quency are personalized.To model interest-transferring and make recommender system be more " human-minded",we propose a new recommendation frame-work called HOMMIT.Under HOMMIT,we can effectively identify user interests from sequential behavioral data and adapt the high-order Markov chain method to model the dynamic transition process of user interests.Then,we can predict the transition trends of user interest and make personalized sequential recom-mendation.We systematically compare the proposed HOMMIT approach with other algorithms on real-world datasets.The results confirm that our new method significantly improves the effectiveness of recommendation.3.Cross-Domain Recommendation for Mapping Sentiment Review PatternCross-domain recommendation systems are gradually becoming more attrac-tive as a practical approach to improve the quality of recommendations and to alleviate cold-start problem,especially in small and sparse datasets.These al-gorithms mine knowledge of users and items in a source domain to improve the quality of the recommendations in a target domain.Most existing works about cross domain recommendation tend to aggregate knowledge from different do-mains from the perspective of explicitly specified common information or trans-ferring latent features.However,the aggregated knowledge is merely based on ratings,tags,or the text information like reviews,and ignores the sentiments implicated in the reviews.In this paper,we propose a new cross-domain recom-mendation framework called SRPM.Under SRPM,we can effectively identify the sentiment orientation of user reviews and adapt topic modeling approach to de-duce the sentiment review pattern from user reviews.We systematically compare the proposed SRPM approach with other algorithms on the Amazon dataset.The results confirm that our new method substantially improves the performance of cross-domain recommendation.4.Sentiment-A ware Review Feature Mapping Approach for Cross-Domain Rec-ommendationKnowledge transferring from a data-rich product domain is an effective way to address data sparsity and cold-start issues.Accurately extracting and transfer-ring user latent features from the source product domain to the target product domain is the key to improving the performance of the cross-domain recommen-dation algorithm.In most existing works of cross-domain recommendation,the aggregated knowledge is merely based on ratings,tags,or the text information like reviews,and does not take advantage of the sentiments implicated in the reviews efficiently.In this paper,we consider sentiment information in the cross-domain recommendation problem and transfer positive and negative sentiment-aware re-view feature from source domain to target domain respectively.We propose a new cross-domain recommendation framework called SARFM.Under SARFM,we can effectively identify the semantic orientation of user reviews and extract the sentiment-aware review feature.Through transferring SARF of users,the SARFM method gets a superior performance in cross-domain recommendation.