Improve the performance of recommender systems using extraction User Reviews in big data

Improve the performance of recommender systems using extraction User Reviews in big data

khadijh esfandiari1 Alireza honarvar2 Nasrin jokar3

1) Iran university of science Technology
2) Iran university of science Technology
3) Iran university of science Technology

IOI Code : XYCZ-AAHFF
Publication : International Conference on Engineering and Applied Science(iceasconf.com)
Abstract :
    Recommender systems, due to the increasing social networking services online, has become an emerging research topic in the past few years, In the present, according to the continued rapid expansion of internet use, the need for a recommender system for filtering large amounts of information effectively has greatly increased, The aim of recommender systems is giving the user a list of favorite items, And due to the increased volume of available data, the tool used previously is not suitable to process this volume of data, So having a system that can store and process large data, has become a problem, In this research to solve the mentioned problems, a recommender system works by using Collaborative Filtering based-model, which uses spark processing engine in the context of Hadoop for more detailed recommendations to users by using user reviews, The method used in this study uses a combination of two steps for the user, In the first stage, for each item, all comments are classified into two classes, the positive and negative, according to dictionary-based approach and using ready dictionary Sentiwordnet, In the second stage, by using algorithms based on Collaborative Filtering [CF] and calculating similarities between user items and active user items, the most similar items to active user item placed on a list as suggestions to user, In this stage combines Recommendtion list of previous step, based on the first stage and will remove items that users have negative comments about them from the list, The results show that the present methods would be more efficient and effective than conventional recommended methods.

Keywords : recommender systems big data sentiment analysis hadoop spark