MovieClub

Movie Recommendation System based on Content Based Filtering extracting 7 features out of Dataset

Load the data set and extract features

selected_features = ['genres','keywords','tagline','cast','director','vote_average','popularity']

Used TfidfVectorizer() to create a vector space of floats using concatenated features variable for each movie

combined_features = movies_data['genres']+' '+movies_data['keywords']+' '+movies_data['tagline']+' '+movies_data['cast']+' '+movies_data['director']+' '+str(movies_data['vote_average'])+' '+ str(movies_data['popularity'])

vectorizer = TfidfVectorizer()

feature_vectors = vectorizer.fit_transform(combined_features)

Found array of closed matches amoung all movies in the list for given user input using difflib library

list_of_all_titles = movies_data['title'].tolist()

find_close_match = difflib.get_close_matches(movie_name, list_of_all_titles)

Used the Algorithm of Cosine Similarity (Angular distance between two values on 2D plot) to create similarity matrix according to values of array of close matches to User Input

similarity = cosine_similarity(feature_vectors)

index_of_the_movie = movies_data[movies_data.title == close_match]['index'].values[0]

similarity_score = list(enumerate(similarity[index_of_the_movie]))

Sort the movies according to similarity value present at 1st index i.e x[1] in Descending order (Higher similarity first)

sorted_similar_movies = sorted(similarity_score, key = lambda x:x[1], reverse = True)

Create three lists and append the information of top 15 movies recommended their name, genre, homepage website values in it, and pass the lists to HTML frontend via Flask backend

return flask.render_template('positive.html',movie_names=names,movie_date=genre,movie_dir=dir, search_name=m_name)

Future Scope

To create separate pages of lists of movies recommended according to SELECVTIVE features choosen by user like separate lists of recommendations based on Ratings, separate list for recommendations based on genre and so on!

Challenge

Cold start problem which persists in all content based filter models as for new user we cannot reecommend anything as we dont know his/her preference.

Viable Solution

Take User's location data from his/her Internet adderess and based on language spoken in that location recommend the movies!

Open for Collab!