Movie Recommendation System using Sentiment Analysis from Microblogging Data (1811.10804v1)

Abstract: Recommendation systems are important intelligent systems that play a vital role in providing selective information to users. Traditional approaches in recommendation systems include collaborative filtering and content-based filtering. However, these approaches have certain limitations like the necessity of prior user history and habits for performing the task of recommendation. In order to reduce the effect of such dependencies, this paper proposes a hybrid recommendation system which combines the collaborative filtering, content-based filtering with sentiment analysis of movie tweets. The movie tweets have been collected from microblogging websites to understand the current trends and user response of the movie. Experiments conducted on public database produce promising results.

Movie Recommendation System using Sentiment Analysis from Microblogging Data

The paper presents a comprehensive approach to movie recommendation systems by integrating sentiment analysis obtained from microblogging data with traditional collaborative filtering (CF) and content-based filtering (CBF) techniques. The primary aim is to enhance the accuracy and reliability of movie recommendations through a hybrid system. It utilizes publicly available databases, particularly the MovieTweetings database, to extract relevant data for sentiment analysis and recommend movies based on user preferences and emerging trends.

Key Contributions

1. Hybrid Recommendation System: The paper introduces a novel hybrid recommendation system that combines CF, CBF, and sentiment analysis derived from social media interactions. This system is designed to mitigate the limitations of relying solely on user historical data, which is typically a requirement for CF and CBF, by incorporating real-time sentiment data from social media.
2. Sentiment Analysis Integration: A key innovation in this work is the use of sentiment analysis from movie-related tweets to boost recommendation accuracy. By evaluating sentiment polarity—using the Valence Aware Dictionary and Sentiment Reasoner (VADER)—the authors derive a sentiment score that enhances movie recommendations beyond static user rating data.
3. Optimization of Feature Weights: The approach described includes deriving optimal weights for content-based feature attributes through social graph analysis, specifically using Moore-Penrose pseudoinversion. This allows for a more nuanced and contextually aware evaluation of movie similarity.

Methodology

The authors propose a hybrid framework that synthesizes collaborative, content, and sentiment data for recommendations. The system leverages a modified MovieTweetings dataset, ensuring the inclusion of metadata fetched via The Movie Database (TMDb) API for a comprehensive similarity computation. Sentiment scores are computed and integrated using a compound scoring methodology that translates social media sentiment into a relatable movie rating scale.

The hybrid model computes a final composite similarity score, which factors in movie attributes and user-linked sentiment analysis. This is achieved by formulating a weighted sum of hybrid similarity and sentiment similarity, where weight optimization is guided by empirical precision at various recommendation levels.

Results

Empirically, the hybrid model demonstrated superior performance over individualized and purely sentiment-based models. Notably, average precision metrics for Top-5 and Top-10 recommendations significantly surpassed alternative models. This suggests a successful augmentation of traditional recommendation techniques with dynamic, sentiment-based layers.

Implications and Future Directions

The integration of sentiment analysis into recommendation systems presents significant potential for dynamic and contextually relevant product suggestions. By utilizing real-time data from vast user-generated social interactions, such systems can potentially reduce cold-start issues and improve personalization without extensive prior user data.

Future work can explore further enhancements, such as fine-tuning sentiment analysis granularity or extending the model for real-time applicability by incorporating a broader set of movies, including metadata updates in real-time environments. Furthermore, it may be valuable to explore emotional tone interpretation from diverse social media platforms to refine recommendation precision across different digital commerce applications.

Overall, the paper sets a strong foundation for the utility of sentiment analysis in enriching recommender systems, advocating a forward-thinking integration of traditional models and real-time social data processing.