Research On Personalized Recommendation Algorithm Based On User History Sequence

In the current sequence recommendation algorithm,extracting the relevant information of user history sequence through deep learning method has become the mainstream.However,at present,most recommendation models based on deep learning methods have the problem of insufficient utilization of sequence auxiliary information.In addition,how to construct the user’s long-term interest representation from the user sequence is also a key problem.Therefore,based on the above two problems,this paper proposes two new sequence recommendation models,and then uses the above proposed model to design a small recommendation system based on user history sequence.The main work is as follows:(1)Aiming at the problems that the sequence recommendation model based on self-attention mechanism ignores all kinds of auxiliary information,resulting in the model can’t use them to capture multi-level sequence relationship patterns,a recommendation algorithm integrating time context and feature level information for recommendation algorithm(ITFR)is proposed.Firstly,the item representation and each attribute representation are connected and input into an attention network.After attention weighting,an attribute-based item representation is obtained.Then ITFR applies the self-attention block of perceived time interval and the self-attention block based on item attribute to capture the relationship mode between item and interaction sequence time interval and the implicit relationship between item and attribute,respectively.Finally,the output representations of the two self-attention blocks are connected and input to the full connection layer as a joint output representation for the recommendation of the next item.The experimental results show that the method of enhancing sequence representation with auxiliary information can improve the recommendation performance.(2)Users’ interests are usually divided into long-term interests and short-term preferences.In view of the problems that the existing recommendation models ignore users’ long-term interests,which makes the model unable to make recommendations by using the representation of users’ long-term interests,a long-term and short-term sequence recommendation model based on similar user blocks(LSRSU)is proposed.Firstly,a user similar block structure is designed to learn the representation of dynamic items,and an embedded representation related to users based on items is obtained by using this structure.Each item based and user related embedded representation is spliced into an item related user representation matrix.After compression,the representation is finally taken as the long-term interest representation of users.The second part is to use the self-attention block to obtain the user’s short-term preference representation.The third part is to calculate the weight of long-term interest representation and short-term preference representation by calculating the similarity between candidate items and similar user related items and the similarity between candidate items and user’s recent sequence interactive goods.The end user generates a comprehensive representation of long-term and short-term interests.The experimental results show that the lsrsu model has different degrees of performance improvement compared with the benchmark model in the two public data sets.(3)Based on the algorithm proposed in this paper,a personalized film recommendation system based on user history sequence is designed by using Movie Lens-1M dataset.Firstly,the latest user historical behavior sequence is input into the ITFR model to generate a comprehensive representation.Secondly,the similarity between the integrated representation and the movie embedded representation in the movie library is calculated by inner product.According to the similarity score,the first N films are screened from high to low and entered into the recommended list of recall layer.After that,the candidate movie information in the recall layer recommendation list is input into the Deep-FM model for further sorting.The candidate movies are displayed from large to small according to the click through rate.Finally,the sorted movie recommendation list is pushed to the client.The final results show that the small-scale film recommendation system truly uses the user’s historical customer delivery sequence information and improves the recommendation performance,which has a certain practical significance.