/Movie-Recommendation

A Novel Movie Recommendation System that aims to recommend movies based off of available services

Primary LanguagePython

#Development Concepts

  • ML Algorithms

    • Content Based Filtering
      • Based on users past views
    • Collaboritve Based Filtering
      • Uses other users ratings
    • Cosine Similarity

      • Used to measure how similar two items are. Mathematically, it measures the cosine of the angle between two vectors projected in a multi dimensional space. The output value is between 0 - 1 Where 0 is no similarity where 1 is 100%

      • Basically what we can do is compare all movies to a movie users like and choose a minimum value for the cosine value to be recommended to the user

  • Recommendation Systems

    • Most recommendation systems typically use a similarity matrix to make recommendations

##Ratings Algorithm

  • My first thought was to simply sort movies based on ratings
  • I would assign a weighted rating based on TMDB rating algorithm
    • TMDB wr = (v/v+m *R) + (m/v+m * C)
    • m is the minimum votes required to be listed in the chart, 95 percentile is recommended
    • R is the average rating of the movie
    • C is the mean vote across the whole report

##Concepts used to code

  • I needed to vectorize some data to use the similarity matrix
    • Python has a module called TF-IDF Vectorizer
      • A TF-IDF is an abbreviation for Term Frequency Inverse Document Frequency
      • A common algorithm to transform text into meaningful representation of numbers which is used to fit machine learning algorith for prediction
      • The algorithm looks at important words only, usually excludes prepositions

###Cosine Sim in Python

  • So cosine sim function is part of the sklearn.metrics.pairwise
  • This cosine function takes in a list as a paramater and assigns the cosine value as a dot product, this row x col (like a matrix)
  • The equation for Cosine Sim is
    • cos(x, y) = x * y / ||x|| * ||y||
    • ||x|| is the length of the vector
    • ||x|| * ||y|| is the cross product

###TF IDF Vectorize Matrix

  • Need to do more research on what this returns and what it means
  • However, what I need it to put into the cosine_sim function to get cos value
  • The function for Tf-IDF Vectorizer is TF * IDF
    • TF is Term Frequency TF() = log2(Freq(i, j)+1)/log2(L)
      • L = Total Number of documents in J
      • Freq(i, j) frequecny of term i in document j
    • Term frequency determines a word relative frequency within a document
    • IDF is the inverse document frequency IDF() = log2(1+Nd/f)
      • Nd = Number of documents considered
      • f = number of documents containing term i
  • The function returns a sparse matrix
  • Hard to demonstrate with real data since the values change based on how large our dataset is
  • A way to improve this would be to include movie ratings from the movies that have similar scores
    • An issue arrises though when a similar movie (like in a series such as the dark knight and batman and robin) when the similar doesn't have a high enough rated score. This is a fault of the dataset we use and how we have weighted scores

###Anothe way to improve

  • My theorize right now is the more data will result in higher similarity scores. If i include things like director, keywords, genres, cast will I get a better score for movies?
  • I am also giving more weight to directors since thats what people like and I am doing this by placing the name of the director three times in a list
  • For keywords we want words that are greater than 1 and change words to its stem

###Collaboritive Filtering

  • Making a hybird interpreter