Less popular movie recommendation

Deshpande, Pradyumna and Desarkar, Maunendra Sankar (2017) Less popular movie recommendation. Masters thesis, Indian Institute of Technology Hyderabad.

Text
CS15MTECH11010.pdf - Submitted Version
Restricted to Registered users only until 2 July 2020.
Download (1MB) | Request a copy

Abstract

In recent years , recommender system have received attention and gained tremendous popularity because it has many applications in a real world and they can play very signi fi cant role in our today’s life whether if we need to make everyday’s decisions or in E-commerce. Users are provided the recommendations on variety of personalised and relevant items or activities with the help of recommendation systems. In these thesis we are implementing graph networks using collaborative fi ltering algorithm and pearson correlation for graph operations for a movie recommendation. Graph networks refer to the networks comprising single/multiple types of entities as well as their interaction relationships. They arise in a great variety of domains, for example, event-based social networks Meetup, Amazon, Net fl ix, DBLP etc. Ample number of movies are available all over the world, for one user it is impossible to watch all the movies as all the movies are not interesting or may not be of user’s interest. So we can suggest the relevant movies which would be interesting for users and can fi lter out the irrelevant movies by using movie recommendation system. Although many recommendation algorithms are proposed for the graph data in movie recommendation, none of them is able to explicitly recommend the newly added movies in the corpus, but also allowed the popular movie to recommend. So in order to in fl uence strength between di ⇤ erent types of entities(e.g. User and a Movie), which is a useful information not only for achieving higher recommendation accuracy but also better understanding the role of each entity type in recommendation problems. We are using movielens 10m dataset for movie recommendation. These dataset contains infor- mation about user , movie and the relationship between user and movie. User has id , age , gender , occupation and zipcode as attributes. Movie has id, title, year, genre, weight and number of users rated the movie as attributes. And there is a relationship between user and a movie of name reviewed is present having rating as an attribute. There can be di ⇤ erent storage choices are there for storing these information. In these thesis, we view the user-movie interaction as a heterogeneous graph and explore ways to use this graph for generating the recommendations. In particular, we use a graph database neo4j to store the information and implement collaborative fi ltering with the help of a sequence of query operations on this graph. Many recommendation systems have been used for movie/music recommendation but they let you watch/listen to popular item. Only a small fraction of the items are popular while many items are moderately popular or not popular at all. A system if recommends only popular items, than other items may not be shown and user doesn’t get chance to see it. In these way, the item does not get exposure to become popular through this channel. Also for RS perspective, its coverage becomes less which may dissatisfy the sellers/content uploaders.

[error in script]

IITH Creators: