Will I "Really Like" this Movie?

Navigating Movie Website Ratings to Select More Enjoyable Movies

Archive for the category “MovieLens”

This One Is All About You and Movielens

A few months ago my daughter texted me for recommendations for good movies on Netflix or Amazon Prime. I recommended a hidden treasure of a movie, Begin Again, but I couldn’t remember if it was on Netflix or Amazon. I knew it was on one of them. I had to go to each site to find the movie to nail down which streaming service it was on.

My daughter, and others like her, will no longer need to search blindly for movies on the streaming services they subscribe to if they’ve signed up to use my favorite movie recommender site, Movielens. Aside from being a very reliable movie recommender site, it is also the most useful in terms of finding movies to watch.

Within the last couple of months Movielens has added its best feature to date. Not only can you get pages and pages of recommended movies, once you’ve taken the time to rate the movies you’ve seen, but now you can filter them by the most popular streaming services.

Movielens allows you to filter recommendations by movies currently on Netflix, Amazon, Hulu, HBO, and Showtime. You can filter them individually or in combination. In my case, I filter by Netflix, Amazon and HBO. This means that you can get a list of movies that you can watch right now, ranked by the probability that you will “really like” them.

If I go to the Home Page of Movielens right now and go to Top Picks, I can click on the filter’s drop down menu and select Streaming Services. This will provide me with a list of the five services mentioned previously. By clicking on Netflix, Amazon, and HBO, I get a list of movies that I can watch now that I haven’t seen before. There are 5,256 movies available for me to watch right now ranked from the one I’m most likely to enjoy, last summer’s box office surprise Me Before You (Amazon), to the movie I’m least likely to enjoy, The Admirer (Amazon). You’ve never heard of The Admirer? Neither have I. It is a 2012 Russian movie based on the love between Anton Chekhov and a young writer, Lidiya Avilova. ZZZ.

More often than not my posts are about my experiences in finding movies that I will “really like”. This one’s for you. If you only have time to track one movie recommender website, go to Movielens. It will be fun and it will save you time scrolling through lines and lines of movies searching for movies that you might like.

Advertisements

Playing Tag with Movielens, Redux

Last July I wrote an article introducing my use of tags in Movielens to organize my movies. I can’t impress upon you enough how useful this tool is to someone as manic about movies as I am.

Regular readers of this blog know that I’ve shifted my focus to Oscar nominated movies. My research revealed that movies that haven’t receive a single Academy Award nomination have only a 35.8% chance of being a “really like” movie. On the other hand, even a single minor nomination increases the “really like” odds to around 55%. My algorithm now incorporates Oscar recognition.

Based on this finding, I’ve altered my tagging strategy. I created an “Oscar” tag that I attach to any movie I run across that has received even a single nomination. Many of these movies are older without enough credible data in the movie ratings websites to earn reliable recommendations. The probabilities in my algorithm for these Quintile 1 & 2 movies are driven by their Oscar performance.

Movies that pique my interest that weren’t Oscar nominated are tagged separately. Now, because these movies have no Oscar nominations, a Quintile 1 or 2 movie is going to have a “really like” probability closer to the 35% mark that reflects its “no nomination” status. It can only climb to a high enough probability to be considered for my weekly watch list if it is highly recommended by the movie websites and it falls into a high enough credibility quintile that its Oscar status doesn’t matter much.

I apply one of two tags to non-Oscar nominated movies. If they have fewer than 25 ratings in Movielens, I tag them as “might like”. Realistically, they have no chance of being highly recommended in my algorithm until the number of ratings received from Movielens raters becomes more robust.

Those non-Oscar nominated movies that have more than 25 ratings are tagged as “prospect”. Movies with the “prospect” tag that are highly rated by the websites and have enough ratings to reach higher credibility quintiles can reach a “really like” probability high enough to be considered for the watch list. For example, a quintile 5 movie like The American President can earn a 75% “really like” probability even though it was never nominated for an Academy Award.

I also have created tags for movies I don’t want to see even though they are highly rated. If I’ve already seen a movie and I don’t want to see it again, I tag it “not again”. If I’ve never seen a movie but it’s just not for me, I tag it “not interested”. Movielens also has the capability of hiding movies that you don’t want to see in any of your searches for movies. I take advantage of this feature to hide my “not again” and “not interested” tagged movies.

So, I’ve tagged all of these movies. Now what do I do with them. That will be included in next week’s topic “Building a Watch List”.

In Pursuit of “Really Like” Movies, Playing Tag with Movielens Can Be Helpful

You may wonder where my list of Top Ten Movies Available To Watch This Week comes from. Or, you may be curious how far Mad Movie Man madness extends. Or, maybe you neither wonder nor are curious, but are instead just happy to have another place to go to for movie recommendations. Whether you have questions unanswered or questions unasked, today is your lucky day. This is the day that you discover how obsessive I can be in pursuit of movies that I will “really like”.

Before I plumb the depths of this madness, a few words about a Movielens tool that helps me organize the mania, would be appropriate. Movielens has a tagging system. There are Community Tags that are available for everyone to use. For example, here are the Community Tags for There’s Something About Mary.

 

You can use these tags as an additional screen when searching for a movie to watch. The number next to the tag tells you how often the tag has been used for this movie. By looking at these tags you could conclude that this movie isn’t for everyone and it probably wouldn’t meet the cringeworthy test. There is a plus sign next to the tag that allows you to agree or disagree with the tag. Also, if you thought the movie was “hilarious”, you can click on the tag and all of the movies will come up that have been tagged “hilarious” Try it on the list above.

If you want to keep track of all of the movies that you thought were “hilarious” there is a box where you can make “hilarious” one of your own personal tags for this movie. If you want to add a tag that isn’t listed for this movie, you can do that too. It is a very dynamic system for gaining additional insights into a movie and for helping you to organize your movies if you are so inclined.

Which brings me back to my manic inclinations. I keep a list of approximately 450 movies for which I’ve calculated “really like” probabilities. Each of these movies I assign the Movielens tag “reviewed” to. This allows me to keep track of movies that I’ve already put on my list of 450. These movies stay on the list as long as they qualify as a recommended movie on one of the five movie websites I use. The data for each movie is refreshed every 90 days.

Movies that I come across that aren’t on the list of 450 that I’m intrigued by are tagged “prospect” in Movielens. Whenever I watch a movie from the list of 450, or it is removed from the list because one of the websites no longer recommends it, I replace it with a movie from the “prospect” list. Movielens allows you to go to Your Tags and sort the movies you’ve tagged. For example, I take the movies that I’ve tagged “prospect” and Movielens sorts them by Movielens Recommended movie, from the highest to the lowest. The highest recommended “prospect” movie moves to the list of 450 to replace the movie removed from the list.

Each Wednesday, after reviewing which movies from the list of 450 are available to watch on the various movie outlets available to me, I rank them by their “really like” probabilities, with the top ten making my list.

Now you understand why I’m the “Mad” Movie Man and how playing tag with Movielens enables my madness.

 

 

 

In the Battle of Memory vs. Movie Website, Netflix is Still the Champ

On Monday I posed the question, is your memory of a movie that you’ve already seen the best predictor of “really like” movies. Based on Monday’s analysis memory certainly comes out on top against IMDB and Rotten Tomatoes. Today, I’m extending the analysis to Criticker, Movielens, and Netflix. By reconfiguring the data used in Monday’s post, you also can measure the relative effectiveness of each site. For example, let’s look again at IMDB.

Probability I Will “Really Like” Based on IMDB Recommendation
Recommended Not Recommended Percentage Point Spread
Seen Before 80.1% 69.2%                            0.11
Never Seen Before 50.6% 33.6%                            0.17

It’s not surprising that the probabilities are higher for the movies that were seen before. After all it wouldn’t make sense to watch again the movies you wished you hadn’t seen the first time. But by looking at the gap between the probability of a recommended movie and a non-recommended movie, you begin to see how effectively the movie recommender is at sorting high probability movies from low probability movies. In this instance, the small 11 point spread for Seen Before movies suggests that IMDB is only sorting these movies into small departures from average. The low probabilities for the Never Seen Before movies suggest that, without the benefit of the memory of a movie seen before, IMDB doesn’t do a very good job of identifying “really like” movies.

Rotten Tomatoes follows a similar pattern.

Probability I Will “Really Like” Based on Rotten Tomatoes Recommendation
Recommended Not Recommended Percentage Point Spread
Seen Before 80.5% 65.1%                            0.15
Never Seen Before 49.8% 31.8%                            0.18

Rotten Tomatoes is a little better than IMDB at sorting movies. The point spreads are a little broader. But, like IMDB, Rotten Tomatoes doesn’t effectively identify “really like” movies for the Never Seen Before group.

Theoretically, when we look at the same data for the remaining three sites, the Percentage Point Spread should be broader to reflect the more personalized nature of the ratings. Certainly, that is the case with Criticker.

Probability I Will “Really Like” Based on Criticker Recommendation
Recommended Not Recommended Percentage Point Spread
Seen Before 79.3% 56.4%                            0.23
Never Seen Before 45.3% 18.9%                            0.26

Like IMDB and Rotten Tomatoes, though, Criticker isn’t very effective at identifying “really like” movies for those movies in the Never Seen Before group.

When you review the results for Movielens, you can begin to see why I’m so high on it as a movie recommender.

Probability I Will “Really Like” Based on Movielens Recommendation
Recommended Not Recommended Percentage Point Spread
Seen Before 86.6% 59.6%                            0.27
Never Seen Before 65.1% 22.3%                            0.43

Unlike the three sites we’ve looked at so far, Movielens is a good predictor of “really like” movies for Never Seen Before movies. And, the spread of 43 points for the Never Seen Before movies is dramatically better than the three previous sites. It is a very effective sorter of movies.

Last, but certainly not least, here are the results for Netflix.

Probability I Will “Really Like” Based on Netflix Recommendation
Recommended Not Recommended Percentage Point Spread
Seen Before 89.8% 45.7%                            0.44
Never Seen Before 65.7% 21.4%                            0.44

What jumps off the page is that there is no memory advantage in the allocation of movies for Netflix. As expected, the Seen Before probabilities are higher. But, there is an identical 44 point gap for Seen Before movies and movies Never Seen Before. It is the only site where you have a less than 50% chance that you will “really like” a movie you’ve already seen if Netflix doesn’t recommend it.

“If memory serves me correctly, I “really liked” this movie the first time I saw it.” That is an instinct worth following even if the movie websites suggest otherwise. But, if Netflix doesn’t recommend it, you might think twice.

***

6/24/2016 Addendum

I’ve finalized my forecast for the last three movies on my June Prospect list. My optimism is turning to pessimism regarding my hopes that Independence Day: Resurgence and Free State of Jones would be “really like movies”. Unfavorable reviews from the critics and less than enthusiastic response from audiences suggest that they could be disappointments. Of my five June prospects, Finding Dory seems to be the only safe bet for theater viewing, with Me Before You a possibility for female moviegoers. The IMDB gender split is pronounced for Me Before You with female voters giving it an 8.1 rating and males a 7.3 rating. It is also one of those rare movies with more female IMDB voters than males.

Movielens Knows About Your Ratings

In the connected world that we live in, everyone knows us. The NSA, TSA, political campaigns, and Netflix are among the organizations who have profiles on us based on information we’ve supplied on the internet. If you have doubts about this there is a website called AboutTheData.com that will provide you with the data that has been collected on you for the use of various marketing organizations. In the age of Big Data we have sacrificed privacy for the convenience internet access provides us.

Movielens also has a profile on you based on the movie ratings you have provided the website. Unlike so many of the Big Data organizations out there, Movielens shares with you what they know. When you open their website there is a drop down menu right next to their logo. ( MovieLens logo )   The second column in the menu is titled YOUR ACTIVITY. Click on the link titled ABOUT YOUR RATINGS and a page full of graphs and information is provided to you that will offer you insights into your taste in movies.

A review of my ABOUT YOUR RATINGS supplies the following information:

  • December 22, 2006 was the first day I entered movie ratings. I rated 49 movies that day. On May 13, 2016 I rated my 2,009th movie.
  • My most common movie rating is a 4.0 (a “really like” rating) which I’ve given to 494 out of the 2,009 rated movies.
  • 433 of my rated movies were released between 2005 and 20015.
  • Drama is my most watched and favorite genre with 1,371 movies and an average rating of 3.71. The next closest genres were Comedy, 588 movies with 3.48 avg. rating, and Romance, 580 movies with 3.61 average rating. Of genres with at least 100 rated movies, Comedy is my lowest average rated.

ABOUT YOUR RATINGS also provides you with a list of movies that you dislike the most when compared to the average rating. These are my candidates for Razzies (the awards for the worst movies presented the day before the Oscars):

  • Requiem for a Dream (2000)
  • Dancer in the Dark (2000)
  • The Browning Version (1951)
  • Twin Peaks: Fire Walk with Me (1992)
  • There Will be Blood (2007)
  • The Other Side of the Mountain (2001)
  • The Magnificent Ambersons (1942)
  • How Green was my Valley (1941)

My unusual likes (or guilty pleasures) when compared to the average rating are:

  • Saturday Night Fever (1977)
  • The Karate Kid (1984)
  • Titanic (1997)
  • Pretty Woman (1990)
  • Ghost (1990)
  • Notting Hill (1999)
  • Somethings Gotta Give (2003)
  • A League of their Own (1992)

Now you know a lot about my taste in movies. Start rating movies in Movielens and you can discover a great deal about your taste in movies as well.

In the age of Big Data it is refreshing to find an organization with the transparency of Movielens.

 

When It Comes to Movie Rating Websites, There is Strength in Numbers.

If you can only use one website to help you select movies that you will “really like”, which should you choose? That’s a tougher question than you might think. Because I have used all five of the websites recommended here to select movies to watch, my data has been heavily influenced by their synergy. I have no data to suggest how effective using only one site would be. Here’s what I do have:

Probability I Will “Really Like”
Recommendation Standard When Recommended in Combination with Other Sites When Recommended by This Site Only
MovieLens > 3.73 70.2% 2.8%
Netflix > 3.8 69.9% 8.4%
Criticker > 76 66.4% 10.1%
IMDB > 7.4 64.1% 0.3%
Rotten Tomatoes Certified Fresh 62.7% 4.3%

When MovieLens recommends a movie, in synergy with other websites, it produces the highest probability. When Criticker recommends a movie but the other four sites don’t recommend the movie, then Criticker has the highest probability. Netflix is second in both groups. Which one is the best is unclear. What is clear is that the three sites that recommend movies based on your personal taste in movies, MovieLens, Netflix, & Criticker, outperform the two sites that are based on third party feedback, Rotten Tomatoes and IMDB. When Netflix, MovieLens, & Criticker recommend the same movie there is an 89.9% chance I’ll “really like” it. When both IMDB & Rotten Tomatoes recommend the same movie the probability is 75.8% I’ll “really like” it.

What also is clear is that if four websites are recommending that you don’t watch a movie and one is recommending that you do, the probability is that you won’t “really like” the movie no matter how good that one website is overall. The progression of probabilities in the example below gives some perspective of how combining websites works:

Websites Recommending a Movie Probability I Will “Really Like”
None 3.9%
Netflix Only 8.4%
Netflix & MovieLens Only 31.9%
Netflix, MovieLens, & Criticker Only 50.9%
Netflix, MovieLens, Criticker & IMDB Only 71.1%
All Five 96.6%

Stated Simply, your odds increase with each website that recommends a particular movie. If, for example, you were to only use Netflix for your movie recommendations, the probability of “really liking” a movie might be 69.9% but, in reality, it could be any one of the probabilities in the table above with the exception of the 3.9% for no recommendations. You wouldn’t know if other websites had recommended the movie.

So, if I had to choose one website, I’d choose Netflix-DVD if I were one of their 5,000,000 DVD subscribers. If I’m not already a subscriber I’d go with MovieLens. It would be a reluctant recommendation, though, because the strength in numbers provided by using multiple websites is just so compelling.

***

You’ll notice in the Top Ten Movies Available to Watch This Week that there are a number of movies on the list that are available on Starz. I’m taking advantage of the Comcast Watchathon Week which provides for free Starz, HBO, & Cinemax. Some of my highly rated movies which would ordinarily be unavailable are available for the short duration of this promotion. Bonus movies. Wahoo!!

 

 

Movielens: The Reliable Alternative

In previous posts I’ve expressed my concern with corporate interests impacting the integrity of movie recommender algorithms. IMDB is owned by Amazon. Rotten Tomatoes is owned by Fandango. Netflix is owned by, well, …Netflix. Criticker isn’t corporately owned but is partially funded by commercial advertising. Now I present to you, Movielens, which isn’t owned by a corporation and doesn’t advertise on its website. Movielens is operated by GroupLens Research at the University of Minnesota. It exists for the benefit of students at the University who are researching predictive modeling. In other words, it exists because it wants to build the best recommender of movies that you will “really like” that you can possibly build. There is no corporate bottom line. There is just the goal of building a better mousetrap.

So far, it’s done a pretty good job. My benchmark for movies that I will “really like” is 4 out of 5 stars, or 7.5 out of 10, or 75 out of 100, depending on the rating scale used. When I calculate the probability that I will “really like” a movie that meets that benchmark for each individual website, I get the following results:

Website Recommend Criteria Probability I Will “Really Like” 
Netflix > 3.8 94.4%
MovieLens > 3.75 93.7%
IMDB > 7.4 90.2%
Criticker > 76 89.8%
Rotten Tomatoes Certified Fresh 89.0%

Movielens holds its own with Netflix and, unlike Netflix, its algorithm is not held hostage to their corporate interests, and its free. All you have to do is click on the MovieLens link above, sign up, and begin rating movies that you’ve seen. Even though MovieLens uses a five star scale, you can enter half stars. You will, at times, be torn between whether  you “really like” a movie or just “like” it. MovieLens lets you enter 3 1/2 stars for that situation.

I encourage you to use MovieLens. You can pat yourself on the back for making a contribution to science.

***

Geek Alert!! Geek Alert!!

If you look at the Movie Lists I updated yesterday, you may be puzzled why so many movies have the same probability. Each month I recalibrate the probabilities in my Bayesian model. I’m constantly experimenting to get the right balance between a model that has many probability differences among movies, but more uncertainty about reliability, and a model that has fewer probability differences among individual movies but is more reliable. Too many probability groups create the risk of randomness creeping into the probabilities. The Bayesian Model recognizes this and shifts the probabilities closer to the probability of a random movie selection. This happened last month when I used 20 groups. When fewer groups are used, the larger groups that result  are more credible and produce less randomness. The model recognizes this and allows for probabilities closer to the tendencies of the group. This month I went back to 5 groups which produces more reliable probability results but with more movies with the same probability. Just in case you were wondering:)

Post Navigation