An online recommendation system based on collaborative filtering and a long-short memory network

Abstract

An online recommendation system based on collaborative filtering and a long-short memory network comprises a real-time module, a near-line module and an offline module; the real-time module receives auser request in real time and outputs the user request to an HDFS cluster distributed file system built by a plurality of terminals after preprocessing, and meanwhile real-time recommendation and online evaluation are conducted according to the request; The near-line module obtains user request data from the real-time module, distributes the requests to the distributed cache through the message queue in real time, carries out offline training on the recommendation model by using characteristics in the user requests, and carries out incremental updating on the data in the load cache of the recommendation model which is subjected to offline training; The offline module loads characteristics from the characteristic pool; and training in the training set by using the recommendation model, realizing collaborative filtering of the trained recommendation model through an autoencoder to mine the long-term preference of the user, thereby performing recommendation model estimation in the test set by combining with the long and short memory network, and finally performing offline evaluation on the effect of the recommendation model to obtain a final recommendation result.

Description

technical field [0001] The invention relates to a technology in the field of information processing, in particular to an online recommendation system based on collaborative filtering and long-short memory network. Background technique [0002] In recent years, due to the improvement of computing power and the increase of big data storage devices, deep learning has attracted great attention. At present, deep learning has made great breakthroughs and achievements in the fields of speech recognition, image processing, and natural language processing, and there have been related studies applying deep learning to the field of recommendation systems. "Autorec: Autoencoders meet collaborative filtering." Proceedings of the 24thInternational Conference on World Wide Web.ACM, 2015, written by Sedhain, Suvash et al. A collaborative filtering method based on an autoencoder is proposed, which uses an autoencoder to Directly learn the compressed vector representation of row or column da...

AI Technical Summary

Problems solved by technology

Although this method utilizes time series data, it only uses the previous item in a session to recommend users, and cannot discover the user's long-term preferences.

Images