Movie Recommender System for Netflix

Introduction

The rapid increase in data generated in the internet is startling and poses a challenge to online businesses. The problem is having customers going searching through lots of items and services before making their choice (Subramaniyaswamy, V. et al. 2017). The fact that consumers can practically not go through every possible product of an online business (with large number of products or services) is a challenge because these businesses risk losing customers. due to them being stressed out by spending time looking for the product of their choice. Streaming entertainment service companies like Netflix is one of such online businesses with large number of both products (movies) and customers. There has been a remarkable change in the way people watch movies in last decade. In the fourth quarter of 2010, Netflix attracted about 3.1 million net customers which increased their customer base to over 27 million. In addition, with $2.2 billion in sales, Netflix had a 61 percent share of the digital video industry in 2010 (Welter, B.S., 2012). As part of the bounties of a swift improvement in technology, mobile electronic gadgets including portable DVD players, smart phones, and cheap/free software became popular as a more convenient and cheap channel for watching movies and entertainment (Smith, A.D., 2008). Given the enormous number of movies accessible through the internet, it becomes difficult for a user to identify and find movies they are or would be best interested in. S. Halder et al., (2012) Argued that since users’ choice of movie and actor varies, it is expedient that a mechanism for filtering out only movies that a user would be interested in be developed.  

Businesses that regularly offer diverse variety of items, find recommender system very useful. Recommender systems assist firms in actively engaging customers by evaluating their customers’ preferences and suggesting items that match these estimated preferences to the customers. According to Amatriain et al. (2009a), recommender systems use algorithm that suggests items to users based on the ratings that users have previously given to items. According to Amatriain, X. and Basilico, J., (2015), researches have long debated about feasibility of video recommender systems because they make use of metadata instead of the videos themselves. However, Subramaniyaswamy, V. et al (2017) argued that the effectiveness of recommender system is based on the fact that they can manage tens of thousands of ratings and give real time predictions. They also argued that the more data a recommender system is trained on, the more accurate the its predictions would be.

Recommendation systems differ from other types of expert systems because they incorporate the knowledge of an expert in a particular area with the user's preferences, to filter available data and suggest the most relevant information (to a particular user) to the user (Eyjolfsdottir, E.A. et al. 2010). The three common filtration algorithm recommender systems use are content-based, collaborative filtering, and hybrid of the later and the former (Eyjolfsdottir, E.A. et al. 2010). Recommender system is common use case of mainstream large-scale data mining and machine learning in Internet, music, e- commerce, gaming and video. According to Amatriain, X. and Basilico, J., (2015), recommender systems have been effective in improving customer happiness and revenue of companies that use them. This project focuses on using machine learning to build a video recommender system that would suggest movies to customers on Neflix.

Problem Statement

As the size of data in the Internet grows, product categories are become more diverse. According to (Yi, N et al. 2017), it is impossible to present all product categories to customers of large scale online businesses on a web page for example and thus, these businesses risk losing their customers. To some extent, search engines can solve this problem but it however has a downside of being passive. In addition, it also requires users to have knowledge of terms and key words to use for a search. Recommender systems however are a plausible solution to the passive nature and domain knowledge requirements of search engines. Recommender systems are therefore a way to increase the revenue of large scale online businesses by improving users experience (Liang X. 2012).

Aim and Objectives

The aim of this project is to explore, predict and evaluate a movie recommender system that would recommend videos on Netflix. The objectives of this research work are as follows:

1. To better understand into the algorithms used in a movie recommender system.
2. To gain insight into the variation of customers’ movie preferences on Netflix.
3. To build a machine learning model that would recommend movies on Netflix to customers.

Research Questions

The research questions for this project are as follows:

1. How do recommender systems’ algorithms differ?
2. Can a machine learning model explain the variation them movie preferences of users on Netflix?
3. Can machine learning effectively use only meta data to recommend movies to Netflix users?

Deliverables

The deliverables of this project are exploratory data analysis, data mining, data munging, modeling, evaluation, model deployment and a project report. The report should thoroughly explain the source of the data used, exploratory data analysis, code implementation, feature selection, model selection as well as model performance.

Relevance

This project is mainly focused on building a movie recommender system that would recommend videos to users on Netflix.

Methodology

This project would consider a secondary research and modeling. They are further buttressed below:

Secondary research

The secondary research in this project will utilize a systematic approach (Johnson et al., 2016) to review the works of literature. The steps involved in the systematic review of the literature are provided below:

• Step 1: Identify the research questions that can be used for the project.
• Step 2: Identify the keywords that should be used to research the works of literature.
• Step 3: Extract the journals and books that are appropriate for this project.
• Step 4: Write the literature review chapter.

Modeling

This section involves sourcing for and making sense out of data. The steps involved are as follows:

• Data collection
• Data Munging
• Exploratory Data Analysis
• Modelling and model selection
• Model optimization
• Model evaluation

Evaluation

The risk assessment conducted for this project is provided in the table below:

Table 1:  Risk assessment