Time-varying Item Feature Conditional Variational Autoencoder for Collaborative filtering

Recommender system;Conditional variational autoencoder;LSTM;670.42

데이터마이닝 ;

University:서울대학교 대학원

关键词: Recommender system;    Conditional variational autoencoder;    LSTM;    670.42;   

Others  :  http://s-space.snu.ac.kr/bitstream/10371/161028/1/000000157701.pdf

美国|英语

来源: Seoul National University Open Repository

【 摘 要 】

We can assume that some factors that affect the user;;s decision to select products are several factors such as the time-invariant user;;s unique taste and the external trends or fashion that varies with time. Both mentioned factors should be considered in order to create a precise recommendation system, but the current recommendation system has the problem of making recommendations based only on the user;;s history without taking into account the timing of creating a recommendation.Therefore, considering the timing of the recommendation made, this paper proposes a recommendation system that reflects inter-items trends of time-based bin. We focus on creating a model that could effectively combine the content-based recommender system with the context-based recommender system. Specifically, we use Conditional Variational Autoencoder to add a time dynamic item features to user-item implicit feedback data. In this case, distributed representation of items in the specific period is used as a condition that is added to input and latent variable of VAE respectively. The distributed representation per periods can be extracted using LSTM.By putting a condition into VAE, a hybrid recommendation system can be created to reflect the item trend. The model proposed in this paper differs from existing research in that it reflects the changing characteristics inherent in the product and utilizes it in the recommendation. In addition, we can get the additional effect of solving sparsity problem by using item feature to mitigate sparsity problem. The Movielens data (ml-1m) data and Amazon women;;s clothing dataset are used for the evaluation of the proposed model. The effectiveness of this model is verified by designing experimental methods to evaluate the recommended systems that reflect the time point of recommendation.