Trend Detection Using Time Based Sequential Recommendations

Abstract

One of the most efficient and common application of data mining and machine learning are recommendation Systems. General recommender and sequential recommender are two applied modelling paradigms for recommendation tasks. General recommender focuses on modelling the general user preferences, ignoring the temporal aspects in user behaviors, whereas sequential recommender focuses on temporal patterns existing in user behavior. Sequential recommendation algorithms are typically designed to learn temporal patterns from user behaviors present in the form of interaction logs. One of the important problems which can be solved using sequential recommendations is trend detection. Trend detection is an important task of detecting trends for individual as well as community. The patterns generated from sequential recommendations are used to predict the users next action within an ongoing session which will be the trends for user itself or to detect short-term trends for community. Deep Learning techniques can be useful for sequential recommendation tasks. Recurrent Neural Networks are successful deep learning models for processing sequences. They are models which can be extended flexibly and also these models can integrate different types of data having temporal patterns. Due to these features, they are perfectly suitable for generating recommendations which are sequential. In this paper, we propose a model based on the extension of recurrent neural network that is Gated Recurrent Neural Network which is used to produce personalized next item recommendation in the sequential settings based on temporal properties present in user behavior and then used these next item recommendation to generate trends for particular user as well as community trends. We also perform experiments on real world dataset to analyze the model on the basis of several evaluation metrics.