Influences of Deep Learning on Recommendation Systems

Author Name(s): Neha R. Kasture
Author Email: kasturenr@rknec.edu

Abstract

Deep structured learning is a noteworthy advancement in the subset of machine learning. Recommendation not only requires domain knowledge but also needs data science intuition to cope up with the abundance and complexity of data in the age of information overload. The plethora of information available on internet and especially on social networking sites can be exploited to provide some personalized recommendations to the users. These recommendations could be predicting the next point of interest (POI) or stop over in the domain of tourism, judging the future preferences of the users from their past choices or predicting their socio-historical inclination from the data available through location-based social networks (LBSN’s). Lot of research has been conducted recently which gives effective recommendation. The goal of this article is to provide a comprehensive survey and comparative analysis of the state-of-art research techniques based extensively on recommendation systems used for applications related to sequence learning.

Keywords Deep Learning, Recommendation Systems, Location Based Social Networks, POI Recommendation, Sequence Learning, Social Media Analysis.

Introduction

Recommendation systems are a tool to filter the data of user’s choice from dearth of online information. Such systems can ease the user’s decision making process related to the products or services which they are interested in but have not seen from a huge array of available options. Lately recommendations are highly used in the tourism industry to predict the tourist’s next stop over or point of interest based on spatial and temporal data like check-ins available through location based social networks (LBSN’s). Recommendation systems have thus become the core area of industry to boost online websites and sales services with increased customer satisfaction. Deep learning is recently gaining remarkable attention not only in field of natural language processing but is also dominating in the area of recommendation systems to provide effective results which can enhance user online experience. Combining the techniques of deep learning with the architecture of recommendation systems is better adapted to capture the non linear characteristics of user behavior or predicting the next point of interest for the tourist to flash the likable advertisements. Recommendation models are mainly classified as collaborative filtering, content based recommender system and hybrid recommender system as shown in Figure 1 based on the type of data given as input to the system [1].

 

Conclusion

This article gives a review based on three significant techniques in the domain of deep learning for POI recommendation. The review consists of the analysis of the techniques along with its characteristics and comparison with other stated methods. The novel matrix factorization method FPMC-LR investigates the spatial-temporal properties of LBSN datasets. Although localization constraint plays a vital role in this model for successive POI recommendation in FPMC-LR, it reduces the computation cost and also discards the noisy information which improves the quality of recommendation. The RNN based approach can model and predict the POI categories to recommend a personalized tourist path. This RNN model shows one loophole and that is indefinite interpretation in case of different user clusters where some cases show improvement while in some cases performance drops.Word2Vec techniques although preliminary but the analysis shows that the performance of skip-gram techniques is better than the results achieved through CBOW. Deep learning methods have a huge potential and can definitely enhance the traditional recommender systems. The reviewed work shows that still there is a scope to use heterogeneous unstructured data samples like visual, audio and video features as an input for the deep learning methods. Feature engineering is in widespread use in industries. But since it is very tedious to manually craft the selected features, deep learning can play a vital role to automate feature engineering.

792 total views, no views today

Download PDF File

About the author: admin