A hybrid recommendation system and method based on multi-task depth learning

Abstract

The invention provides a hybrid recommendation system based on multi-task depth learning, it consists of three stages in turn, phases I and II, phases II and III, the first stage is a hybrid recommendation model construction, the second stage generating a set of training samples, the third stage is hybrid recommendation model training, after three stages of calculations, the hybrid recommendationmodel construction comprises a convolution neural network and a hybrid recommendation model of metric learning. As the invention mainly consists of three stages, the implementation process of each stage is simple and easy to realize, and is not limited by specific development tools and programming software, and can be rapidly extended to a distributed and parallel development environment.

Description

technical field [0001] The invention relates to the field of computers, in particular to a hybrid recommendation system based on multi-task deep learning and a method thereof. Background technique [0002] With the advent of the era of big data, people need to face more and more data information, how to extract valuable information from massive data has become a great challenge. [0003] The recommendation system can extract the user's interests and preferences from the user's historical information, and recommend items that may be of interest to the user, which has gradually become a hot spot of attention. [0004] Traditional recommendation methods mainly include content-based recommendation methods, collaborative filtering recommendation methods, and hybrid recommendation methods. Among them, content-based recommendation methods recommend similar items to users according to their historical favorite items, and collaborative filtering recommendation recommends similar item...

AI Technical Summary

Problems solved by technology

[0005] However, the high-quality recommendation of traditional recommendation methods is based on heavy feature processing. A large number of feature extraction, feature combination, and feature selection need to be processed by technical personnel, and this processing can only capture the shallow level of users and items. Sub-relationships, unable to dig deep into the deep features of users and items

[0006] At the same time, it is also found that using metric learning to predict user preferences can help improve the recommendation effect, but this recommendation method only focuses on the relationship between users and items, and does not effectively mine the internal relationship between items and items. When the user's historical feedback data on items is sparse, good recommendation results cannot be achieved

Images