A standard model for Recommender Systems is the Matrix Completion setting: given partially known matrix of ratings given by users (rows) to items (columns), infer the unknown ratings. In the last decades, few attempts where done to handle that objective with Neural Networks, but recently an architecture based on Autoencoders proved to be a promising approach. In current paper, we enhanced that architecture (i) by using a loss function adapted to input data with missing values, and (ii) by incorporating side information. The experiments demonstrate that while side information only slightly improve the test error averaged on all users/items, it has more impact on cold users/items.

https://hal.inria.fr/hal-01336912/document

Hybrid Recommender System based on Autoencoders

The research work of the third author is partially funded by the WIQ-EI (IRSES grant n. 269180) and DIANA APPLICATIONS (TIN2012-38603-C02-01), and done in the framework of the VLC/Campus Microcluster on Multimodal Interaction in Intelligent Systems.

/pdf/twin-personality-based-intelligent-recommender-system-1bfxvd1zqq.pdf

TWIN: Personality-based Intelligent Recommender System

Collaborative Filtering aims at exploiting the feedback of users to provide personalised recommendations. Such algorithms look for latent variables in a large sparse matrix of ratings. They can be enhanced by adding side information to tackle the well-known cold start problem. While Neu-ral Networks have tremendous success in image and speech recognition, they have received less attention in Collaborative Filtering. This is all the more surprising that Neural Networks are able to discover latent variables in large and heterogeneous datasets. In this paper, we introduce a Collaborative Filtering Neural network architecture aka CFN which computes a non-linear Matrix Factorization from sparse rating inputs and side information. We show experimentally on the MovieLens and Douban dataset that CFN outper-forms the state of the art and benefits from side information. We provide an implementation of the algorithm as a reusable plugin for Torch, a popular Neural Network framework.

Hybrid Collaborative Filtering with Autoencoders

There are lots of issues existing in traditional collaborative filtering recommendation, such as data scarcities, cold start, recommendation accuracy and timeliness. And how to improve the efficiency and quality of recommendation is a key problem in collaborative recommendation. In the traditional collaborative filtering algorithms, the rating scale of different users for all projects sometimes may be neglected while calculating the similarity. Some algorithms such as adjusted cosine similarity algorithm and the Pearson similarity algorithm are proposed to optimize this problem, but there still exists the problem that the single rating scale is different for the same project with different users. It may result in similar resultant vector results when the users have significant differences for the score vectors on a common set. The substantial presence of this kind of phenomena has a direct impact on the accuracy of user similarity calculation. Furthermore, it will affect the target user's predicted score accuracy. To solve the problem, an improved collaborative recommendation algorithm based on optimized user similarity is proposed. A balancing factor is added to the traditional cosine similarity algorithm, which is used to calculate the project rating scale differences between different users. Also, the most appropriate balance factor threshold can be obtained by experiments, a series of reasonable experiments to validate the effectiveness of the proposed algorithm based on the threshold. Experimental results show that the proposed improved collaborative filtering algorithm based on user similarity can significantly optimize the accuracy of user similarity and get better recommendation results.

An improved collaborative recommendation algorithm based on optimized user similarity

Recommender systems have shown a lot of awareness in the past decade. Due to their great business value, recommender systems have also been successfully deployed in business, such as product recommendation at flipkart, HomeShop18, and music recommendation at Last.fm, Pandora, and movie recommendation at Flixstreet, MovieLens, and Jinni. In the past few years, the incredible growth of Web 2.0 web sites and applications constitute new challenges for Traditional recommender systems. Traditional recommender systems always ignore social interaction among users. But in our real life, when we are asking our friends or looking opinions, reviews for recommendations of Mobile or heart touching music, movies, electronic gadgets, restaurant, book, games, software Apps, we are actually using social information for recommendations. In this paper social popularity factor are incorporated in SVD++ factorization method as implicit feedback to improve accuracy and scalability of recommendations.

/pdf/social-popularity-based-svd-recommender-system-30ubieyh9s.pdf

Social Popularity based SVD++ Recommender System

Contexts and social web information have been recognized to be valuable information for making perfect recommender system. Context-aware recommender systems (CARS) have been implemented in different applications and domains which improve the performance of recommendations. Context-aware approaches have been successfully applied in various domains such as music, movies, mobile recommendations, personalized shopping assistants, conversational and interactional services, social rating services and multimedia. The recommender systems are widely being used for products, content and services recommendations. Successful deployment of recommender system in social web and many commercial website like Amazon.com, flipkart, HomeShop18 and numerous different sectors have already done. The growth of the social web has revolutionized the architecture of sharing and association in the web, making it essential to reiterate recommendation. If recommender systems have established their key role in providing the user access to resources on the web, when sharing resources has turn into social, it is likely for recommendation techniques in the social web should consider social popularity factor and the relationships among users to compute their predictions. In this paper contextual information are being included in social popularity based SVD++ model to improve accuracy and scalability of recommendations.

/pdf/context-aware-social-popularity-based-recommender-system-3epgwg3s11.pdf

Shyam Sunder Rastogi

Papers

Social Popularity based SVD++ Recommender System

Context-aware Social Popularity based Recommender System