| At present,users are facing a serious problem of information overload.Recommender system is playing an indispensable role in our daily lives as well as in the Internet industry for the problem of information overload.In practical applications,almost every piece of data has a time label,and the time span between data is a problem that cannot be ignored.Therefore,the temporal recommendation system has received widespread attention.Intuitively,the older the data is,the less time weight will be in recommendation,so the conventional research always use the forgetting curve to model the time factor.However,these tasks only take time as a common attribute rather than a dimension.Each interaction actually is not independent in the time dimension and their chronological order contains a lot of information,utilizing the symbolic sequence relationship among the interaction and the overall structure of the data network will use these contexts efficiently and benefit the measure precision of similarity search.As the emerging topic to solve the loss of time dimension information,sequential recommender systems has attracted increasing attention in recent years.Sequential recommender systems treat the user-item interactions as a dynamic sequence and take the sequential dependencies into account to capture the current and recent preference of a user for more accurate recommendation.However,some problems also perplex the sequential recommender system.First,in the sequential recommender systems,the short-term sequential pattern may cover the long-term sequential pattern.Secondly,both the users' preference and items' popularity are dynamic rather than static over time.In fact,a user's preference and taste may change over time.The popularity of items is deeply influenced by seasonal,holiday,weather and other factors.Such dynamics are of great significance for precisely profiling a user or an item for more accurate recommendations.Most of the existing sequential recommender systems can only capture simply the dynamic change in the short term.Finally,like other recommendation systems,sequence recommendation systems also face the challenge of data sparsity and cold start.As a promising solution to address these issues,cross-domain recommender systems have gained increasing attention in recent years.This kind of algorithm tries to utilize explicit or implicit feedbacks from multiple auxiliary domains to improve the recommendation performance in the target domain.Unfortunately,most existing migration methods cannot handle time informationIn this article,we combine similarity search to construct a new sequential recommendation model to solve the problems of long-term sequence patterns being covered by short-term sequence patterns and dynamic changes in user preferences and commodity popularity.Then,the sequence completion algorithm is used to construct a cross-domain sequential recommendation system.Our major contributions are summarized as follows:1.Recommendation with Temporal Dynamics Based on Sequence Similarity SearchWe propose a recommendation framework called SeqSim,which can solve the problem of timing and preference evolution in recommendation systems.In SeqSim,we re-model users or items and use sequences to represent them.Based on this,the calculation of similarity between users or items is transformed into the calculation of sequence similarity.In order to measure sequence similarity,a new similarity algorithm is proposed.At the same time,in order to improve the efficiency of similarity search,we designed a new time clustering algorithm to convert item sequences into clustering sequences.A new algorithm with preference curve is proposed for recommendation.We systematically compared the SeqSim method with other mainstream algorithms on MovieLens and Amazon datasets.The results confirm that our new method greatly improves the accuracy of recommendations2.Long Short-Term Memory with Sequence Completion for Cross-DomainSequential RecommendationIn order to solve the challenges of data sparsity and cold start,we defined the concept of sequence completion and proposed a sequence completion algorithm.Among them,we use the rating matrix between users and items,the intrinsic characteristics of users and items and the temporal characteristics of user behavior to establish a similarity measure of sequence completion.Then the algorithm was applied to three sequential recommendation models to prove the effectiveness of the algorithm.First,use the SeqSim model we built above;second,use the Long Short-Term Memory(LSTM)directly;Finally,we deeply integrate the sequence completion algorithm into the Long Short-Term Memory,improve Long Short-Term Memory network by building the connection between the output layer and the input layer of the next time step.Complete the sequence by adding this input logic unit.We have conducted a lot of experiments on the Amazon dataset,and the experimental results show that our three models are better than the current mainstream models in recommendation performance.In addition,we combined the experimental results under different evaluation standards to analyze the pros and cons of the three models in a more granular manner. |
