MEENAKSHI COLLEGE OF ENGINEERING

WEST K.K. NAGAR, CHENNAI-600 078.

(Approved by AICTE and Affiliated to ANNA UNIVERSITY)

NM1090 - NATURAL LANGUAGE PROCESSING TECHNIQUES

NAME :

REG. NO. :

BRANCH :

YEAR :

SEMESTER :
 NAN MUDHALVAN PROJECT
NM1090 - NATURAL LANGUAGE PROCESSING TECHNIQUES

Movie Recommendation Project

Submitted by the team members of III Year CSE ‘A’

G.Mahalakshmi: 311422104053
S.Janasri: 311422104033
A.Divyadharshini:311422104016

COMPUTER SCIENCE AND ENGINEERING

​​ ​ MEENAKSHI COLLEGE OF ENGINEERING
KK NAGAR WEST CHENNAI - 600078
Abstract

In the age of digital entertainment, users are often overwhelmed by the vast number of movie
options available across various streaming platforms. To enhance user experience and
facilitate personalized content discovery, recommendation systems play a crucial role. This
project focuses on developing a Content-Based Movie Recommendation System that
intelligently suggests movies similar to a user’s selection by analyzing the content features of
the movies.

The proposed system utilizes Natural Language Processing (NLP) techniques and machine
learning concepts to understand and extract meaningful features from movie metadata such as
genres, cast, crew, keywords, and overview. These features are combined into a single
composite text field called "tags", which is then vectorized using Count Vectorizer to
transform textual data into numerical form. To determine similarity between movies, cosine
similarity is applied to the generated vectors.

Unlike collaborative filtering, which depends on user behavior and ratings, the content-based
method used in this system is independent of user history and performs well even for new or
less-reviewed movies. The model provides a fast and scalable solution for recommending
movies by focusing on intrinsic movie attributes, making it ideal for cold-start scenarios.

This approach not only improves user engagement by presenting relevant recommendations
but also demonstrates the potential of combining NLP with vector space models for
intelligent decision support.
 TABLE OF CONTENT

1. PROBLEM STATEMENT

2.
APPROACH

3.
PROPOSED ARCHITECTURE

4. DATA SET DESCRIPTION

5. IMPLEMENTATION

6. CODING

7. RESULT AND EVALUATION

8. CONCLUSION
MOVIE RECOMMENDATION:
1.Problem Statement

In today’s digital era, entertainment platforms such as Netflix, Amazon Prime, and Disney+
host thousands of movies and TV shows spanning across various genres, languages, and
cultures. While this wide range of content offers users immense choice, it also presents a
significant challenge—information overload. Users often spend a considerable amount of
time browsing through countless titles, struggling to find content that aligns with their
personal preferences.

Traditional search systems are typically keyword-based and lack the capability to understand
user preferences, context, or content similarity. These systems may return irrelevant or
generic results, leading to a frustrating user experience. Additionally, users may not always
be aware of titles that match their interests but use different tags or less popular keywords.

The inability to filter and recommend personalized content diminishes user satisfaction and
reduces platform engagement. Moreover, newly released or less popular movies might go
unnoticed due to the lack of effective recommendation mechanisms. This scenario calls for an
intelligent recommendation system that can intuitively understand the user's likes and
recommend content accordingly.

The core problem lies in designing a system that:

●​ Understands and processes unstructured metadata such as plot summaries, genres, and
cast information.
●​ Identifies content similarity between movies using meaningful patterns.
●​ Provides personalized suggestions without relying on user ratings or viewing history.

To address this, the project proposes a content-based movie recommendation system that
leverages machine learning and natural language processing techniques. The goal is to build a
model that can analyze movie attributes and deliver recommendations that are contextually
and thematically relevant to the user’s choice, thereby enhancing discoverability and
satisfaction

2. APPROACH:

The architecture of the content-based recommendation system involves the following

components:

1.​ Data Preprocessing:

o​ Merge necessary features like genres, cast, crew, keywords, etc.
o​ Clean and normalize text (e.g., lowercase, remove punctuation).
2.​ Feature Engineering:
o​ Combine selected features into a single "tags" column.
o​ Convert text to numerical vectors using CountVectorizer.
3.​ Similarity Computation:
o​ Compute pairwise cosine similarity between movie vectors.
 4.​ Recommendation Generation:
o​ Sort movies based on similarity scores.
o​ Return top N most similar movies.

3.PROPOSED ARCHITECTURE:

4.DATA SET DESCRIPTION:

The dataset used for this movie recommendation system is movies (1).csv. This CSV file
contains comprehensive information about movies, including the following key columns:
 ●​ genres: The genres associated with the movie.

●​ keywords: Important keywords describing the movie's themes or plot.

●​ overview: A brief summary of the movie's storyline.

●​ cast: The main actors in the movie.

●​ crew: Key crew members, including the director.

●​ director: The director of the movie.

Other fields such as budget, homepage, id, original_languagE,, original_title, popularity,
production_companies, production_countries, release_date, revenue, runtime,
spoken_languages, status, tagline, title, vote_average, and vote_count are also present.
The dataset is loaded into a Pandas DataFrame using pd.read_csv('/content/movies (1).csv').
For the recommendation system, specific textual features are selected, including genres,
keywords, tagline, cast, and director. Prior to processing, any missing values within these
selected features are handled by replacing them with empty strings, ensuring data
consistency. These cleaned features are then concatenated to create a single
combined_features string for each movie, which serves as the input for the feature extraction
process.
5.IMPLEMENTATION:

The implementation involves several steps as outlined in the movie_recommendation.ipynb

notebook:

1. Importing Dependencies: The necessary libraries are imported for numerical operations,
data manipulation, string matching, text feature extraction, and similarity calculation.

Python
import numpy as np [cite: 1]
import pandas as pd [cite: 1]
import difflib [cite: 1]
from sklearn.feature_extraction.text import TfidfVectorizer [cite: 1]
from sklearn.metrics.pairwise import cosine_similarity [cite: 1]

2. Data Collection and Pre-Processing: The movie dataset is loaded. Irrelevant features are
dropped, and relevant text features (genres, keywords, tagline, cast, director) are selected and
combined into a single string for each movie. Missing values are handled by replacing them
with empty strings.

3. Feature Extraction (TF-IDF Vectorization): The combined text content is converted into
numerical TF-IDF vectors. This step quantifies the importance of words in describing each
movie.

4. Similarity Calculation (Cosine Similarity): Cosine similarity is computed between all

movie TF-IDF vectors. This creates a similarity matrix where each entry represents the cosine
similarity between two movies.
5. Recommendation Logic:

●​ When a user inputs a movie title, difflib.get_close_matches is used to find the closest
match in the dataset, handling potential misspellings.
●​ The index of the matched movie is retrieved.
●​ The similarity scores for that movie with all other movies are extracted from the
cosine similarity matrix.
●​ The movies are sorted based on these similarity scores in descending order.
●​ The top N most similar movies (excluding the input movie itself) are then
recommended to the user.

Pseudo Code:

function recommend_movies(movie_title):
Load movie dataset (movies_data)
Select relevant features: genres, keywords, tagline, cast, director
Combine selected features into a single string for each movie
Handle missing values by replacing with empty strings

Initialize TfidfVectorizer (vectorizer)

Transform combined features into TF-IDF vectors (feature_vectors)

Calculate cosine_similarity between all feature_vectors (similarity_scores)

Find closest match for movie_title in movies_data.title (find_close_match)

If no close match found, return "Movie not found"

Get index of the movie_title from movies_data (movie_index)

Get similarity scores of movie_index with all other movies (movie_similarity_scores)
Sort movie_similarity_scores in descending order

Initialize an empty list for recommended_movies

For each movie in sorted movie_similarity_scores:
If movie is not the input movie:
Add movie title to recommended_movies
If count of recommended_movies reaches 10, break

Return recommended_movies
6.CODING
7.RESULT AND EVALUATION:

The recommendation system successfully analyzes movie metadata and provides

personalized recommendations. The cosine similarity-based content filtering technique
demonstrates high relevance and accuracy, especially in suggesting thematically similar
movies. The system is scalable and can be integrated into larger entertainment platforms to
enhance user experience.
To validate the system, various sample inputs were tested:
Example:

●​ Input: "Avatar"

●​ Output: "John Carter", "Aliens", "Titan A.E.", "The Helix... Loaded", "Battle Los
Angeles"
The recommendations are contextually aligned with the input movie's genre, storyline, and
themes.
8.CONCLUSION:
This movie recommendation system successfully addresses the challenge of movie discovery
by leveraging content-based filtering techniques. By utilizing TF-IDF for feature extraction
and cosine similarity for measuring movie resemblance, the system effectively analyzes
diverse movie attributes such as genres, keywords, cast, and director. The implementation
demonstrates a robust approach to data pre-processing, feature engineering, and similarity
computation, resulting in the ability to provide accurate and personalized movie suggestions.
The developed system offers a valuable tool for users to explore new cinematic content
tailored to their preferences, ultimately enhancing their movie-watching experience.