Our Project

In response to a tweet reviewing a movie, recommend a movie based on:

• ratings in the MovieLens Database

• overall twitter reviews of the initial movie

Data Sources

• Twitter feed @FilmReviewIn140 : for initial tweets

Data Sources

• General twitter search (I rated ‘movie-name’ #imdb)

Data Sources

• Standford’s large movie review dataset (http://ai.stanford.edu/~amaas/data/sentiment/) : to train sentiment analysis model

– several text files identifed as positive or negative movie reviews

• Movielens large database (https://grouplens.org/datasets/movielens/): to select the top 10 rated movies (based on genre)

– multiple zipped csv files (for project used movies and ratings)

Process

The process to Recommend was as follows:

Results - Sentiment Analysis

• Tested Accuracy against the letter grade assigned to each review
• Achieved 75% accuracy

Result - Recommendation

Godzilla (2014) had negative sentiment and is tagged with the following genres: Action|Adventure|Sci-Fi|IMAX

gave rise to the following selections from movielens database:

Result - Recommendation

And the top 3 with respect to twitter popularity were:

Another Result - Recommendation

Olympus Has Fallen had a positive sentiment and is tagged with the following genres: Action|Thriller

gave rise to the following selections from the movielens database:

Another Result - Recommendation

And the top 3 with respect to twitter popularity were:

Challenges

• How to analyse sentiment?

– Solved by using additional dataset maintained by Standford University to train a prediction model

• How to find tweets with reviews?

– Solved by using a particular twitter handle which had tweets in particular format

– For choosing top three searched for particular phrasing of “I rated movie #imdb”

• For general searches twitter API only gives access to tweets going back one week.

Conclusion

• Project enabled an exploration of recommendation systems where not much is known about the person recommending to (other than one tweet).

• Project helped to understand how to get data from twitter

• Project allowed exploration of text analysis including sentiment analysis

• Further improvements include analysing additional meta data about the movie such as Directors, Actors, Producers, and even plot analysis for bettter recommendation

Questions

THANK YOU