Review of Recommendation Systems using

Sentiment Analysis
Authors: Sahil Kumar[1] , Nishant Sharma[2], Dr. Neha Agrawal[3]

Department of Information Technology

[email protected] ,[email protected],[email protected]

2.1 Types of Recommendation Systems

1. Introduction
In recent years, recommendation systems Recommendation systems are broadly
have become an integral part of industries classified into:
such as e-commerce, social media, and
content streaming platforms. These systems Collaborative Filtering: This method predicts
suggest relevant products, services, or media a user’s preferences based on the behavior of
based on user preferences, significantly similar users. It can be user-based or
enhancing the user experience. Traditionally, itembased. This approach was highlighted in
recommendation systems relied on [1], where it is effective but suffers from
algorithms such as collaborative filtering and scalability issues.
content-based methods. However, with the
increasing availability of user-generated data,
Content-based Filtering: Recommends items
such as reviews and social media comments,
similar to those the user has liked in the past,
sentiment analysis has emerged as a powerful
based on the attributes of the items [3].
tool to refine these recommendations by
analyzing emotional feedback from users. Hybrid Systems: A combination of both
collaborative and content-based filtering to
This review aims to provide an overview of
overcome limitations like cold-start problems
how sentiment analysis can be integrated into
and scalability [1][4].
recommendation systems to improve their
accuracy and relevance. Several studies have 2.2 Challenges in Recommendation
focused on this combination, highlighting Systems
different methods and challenges [1][2].
Traditional recommendation systems face
several challenges:
2. Background on Recommendation Cold Start: Lack of historical data for new
Systems users or items [5].
Scalability: Difficulty in handling large
datasets efficiently [1].

1
Data Sparsity: Insufficient data to make 4.1 Enhancing Traditional
accurate recommendations [4]. Recommendation Systems
Sentiment analysis can provide deeper
insights into user preferences by analyzing
3. Sentiment Analysis Overview
reviews, comments, and feedback,
supplementing traditional recommendation
3.1 Types of Sentiment Analysis methods. For instance, instead of merely
recommending items based on past
Sentiment analysis involves determining the purchases, sentiment analysis enables
emotional tone behind user reviews, systems to recommend items that evoke
comments, and feedback. Methods for positive sentiments among similar users
sentiment analysis include: [1][9].
Lexicon-based Approach: Uses predefined 4.2 Approaches for Integration
dictionaries of positive and negative words to
score text [6]. Direct Integration: In this approach,
sentiment analysis is directly applied to user
Machine Learning-based Approach: Involves reviews to filter out negatively reviewed
training models using labeled datasets to items and prioritize positively reviewed ones
classify text as positive, negative, or neutral [4].
[2].
Hybrid Approaches: Some systems combine
Deep Learning-based Approach: Uses collaborative filtering and content-based
advanced neural networks like LSTM and filtering with sentiment analysis. For
Transformer-based models (e.g., BERT) to instance, user similarity is calculated not just
capture contextual sentiment [4]. based on item ratings but also on the
3.2 Common Tools for Sentiment sentiments they express towards different
products or services [1][10].
Analysis
VADER: Designed for social media and 4.3 Techniques Used in Existing
works well with short texts [7]. Systems
TextBlob: A simple library for basic Existing studies have employed several
sentiment analysis [7]. techniques:

Hugging Face Transformers: Provides pre- Lexicon-based Sentiment Analysis for

trained models for state-of-the-art sentiment filtering content in recommendation systems,
classification [8]. particularly for social media and e-commerce
[6][2].
Machine Learning-based Models like SVM
4. Integrating Sentiment Analysis in and Naive Bayes have been integrated into
Recommendation Systems collaborative filtering to improve
recommendation quality by considering user
sentiments [9].

2
Deep Learning Models such as BERT and Feature Extraction: Converting text into
LSTM are now being used to enhance hybrid numerical features using techniques like
recommendation systems by capturing Term Frequency-Inverse Document
complex contextual sentiment [4][8]. Frequency (TF-IDF) or word embeddings
(e.g., Word2Vec, GloVe).

5.3 Sentiment Analysis

5. Methodology
Sentiment analysis is applied to user-
The methodology for integrating sentiment
generated text to assess the emotional tone.
analysis into recommendation systems can be
Depending on the approach chosen, different
broken down into several key stages:
models are used:
5.1 Data Collection Lexicon-based Approach: Predefined
The first step involves gathering user- dictionaries of positive and negative words.
generated content such as product reviews, Machine Learning-based Approach:
ratings, and social media comments. Algorithms like Support Vector Machines
Common datasets used in this process (SVM) or Naive Bayes trained on labeled
include: datasets.
Amazon Reviews Dataset for product Deep Learning-based Approach: Neural
recommendations. networks, including LSTM or Transformer-
Yelp Dataset for business recommendations. based models (e.g., BERT), for context-
sensitive sentiment analysis.
IMDb Reviews Dataset for media content
recommendations. 5.4 Integration with Recommendation
System
These datasets provide user ratings and
textual reviews, which are essential for both Once sentiment scores are obtained, they are
traditional recommendation systems and integrated into the recommendation
sentiment analysis. algorithm. Two primary approaches for this
integration are:
5.2 Preprocessing
Direct Integration: Recommendations are
Text data must undergo preprocessing to filtered based on the sentiment scores, giving
ensure clean and relevant input for sentiment priority to items with positive feedback.
analysis. This typically includes:
Hybrid Systems: Combining collaborative
Tokenization: Splitting text into individual filtering, content-based filtering, and
words or phrases. sentiment analysis to improve the accuracy of
Stop-word removal: Eliminating common recommendations. For instance, the system
but insignificant words (e.g., "is", "the"). may compute user similarity not only based
on ratings but also on the sentiment of their
Stemming or Lemmatization: Reducing reviews.
words to their root forms.

3
5.5 Model Evaluation Light FM: Used for hybrid recommendation
systems that combine content-based and
To evaluate the performance of the collaborative filtering with sentiment
sentiment-enhanced recommendation analysis [11].
system, standard metrics for both
recommendation systems and sentiment
analysis are used. For recommendation
systems, precision, recall, and RMSE are key
7. Evaluation of Sentiment-based
metrics. For sentiment analysis, accuracy and Recommendation Systems
the F1-score are commonly used to assess the
classification performance.
7.1 Metrics for Recommendation
Systems
6. Datasets and Tools Precision: The proportion of relevant
recommendations out of all
recommendations [9].
6.1 Popular Datasets
Recall: The proportion of relevant
Amazon Reviews Dataset: Contains user recommendations out of all relevant items
reviews and ratings for products, often used [3].
in sentiment-enhanced recommendation
research [10]. RMSE (Root Mean Squared Error):
Measures the prediction error in collaborative
Yelp Dataset: Includes user-generated filtering models [11].
reviews and star ratings, making it suitable
for testing sentiment-based recommendation 7.2 Metrics for Sentiment Analysis
systems [7].
Accuracy: The proportion of correctly
IMDb Reviews Dataset: Used for sentiment classified sentiments [7].
analysis in media and entertainment
F1-Score: A balance between precision and
recommendation systems [8].
recall, particularly useful when dealing with
6.2 Tools for Developing Systems imbalanced datasets [2].

Surprise: A Python library for building

collaborative filtering-based
8. Current Trends and Research
recommendation systems [11].
Gaps
TensorFlow Recommenders: Useful for
building recommendation systems with deep
learning models [8]. 8.1 Trends
Hugging Face Transformers: Offers state-of- There has been a rise in the use of deep
the-art pre-trained models like BERT for learning models for both recommendation
sentiment classification [8]. systems and sentiment analysis. Models like
BERT and GPT are showing significant
improvements in contextual understanding,

4
making them valuable for hybrid
recommendation systems [4].
10. Conclusion
8.2 Research Gaps Integrating sentiment analysis into
recommendation systems offers a more
Despite advancements, several challenges
refined and personalized experience for
remain:
users. By considering both user preferences
Real-time Sentiment Analysis: Most systems and their emotional responses, sentiment-
cannot yet handle real-time sentiment enhanced recommendation systems are better
analysis for streaming data [10]. equipped to make relevant and accurate
suggestions [6]. While there has been
Handling Ambiguity and Sarcasm: Sentiment
significant progress in this field, challenges
analysis models still struggle with
remain, particularly in areas such as real-time
understanding sarcasm, irony, and other
sentiment analysis and handling nuanced
nuanced language in user feedback [8].
language [2][4]. Future research should
Domain-specific Sentiment Models: General continue exploring these areas to improve
sentiment models may not work well in niche system performance and user satisfaction
domains like healthcare or financial services, [10].
which require specialized training datasets
and models [9].
11. References

1. Aggarwal, C. C. (2016).
9. Future Directions Recommender Systems: The
Textbook. Springer.
2. Cambria, E., Schuller, B., Xia, Y., &
9.1 Multimodal Sentiment Analysis Havasi, C. (2013). New avenues in
Incorporating multimodal sentiment analysis opinion mining and sentiment
(e.g., combining text, images, and audio) is analysis. IEEE Intelligent Systems,
an exciting direction for improving 28(2), 15-21.
3. Haruna, K., Chen, Z., & Ning, H.
recommendations, especially in fields like
(2017). A sentiment-based
media or e-commerce, where product
personalized recommendation
reviews often include images and videos system. Multimedia Tools and
[10]. Applications, 76(21), 22143–22156.
4. Liu, B. (2012). Sentiment Analysis
9.2 Real-time Systems
and Opinion Mining. Morgan &
Advancing real-time sentiment analysis Claypool Publishers.
capabilities in recommendation systems is 5. Moraes, R., Valiati, J. F., & Gavião
another promising area. With the growing use Neto, W. P. (2013). Document-level
of streaming platforms and live interactions, sentiment classification: An
empirical comparison between SVM
building systems that can update
and ANN. Expert Systems with
recommendations dynamically based on live
Applications, 40(2), 621-633.
sentiment input could significantly improve
personalization [5].
5
6. Soleymani, M., Garcia, D., Jou, B.,
Schuller, B., Chang, S. F., & Pantic,
M. (2017). A Survey of Multimodal
Sentiment Analysis. Image and
Vision Computing, 65, 3-14.
7. Chaturvedi, I., Ragusa, E., Gastaldo,
P., & Zunino, R. (2018). Emotion-
Aware Recommender Systems Using
Hybrid Sentiment Analysis and
Contextual Information. IEEE
Access, 6, 69264-69278.
8. Zhang, S., Yao, L., Sun, A., & Tay,
Y. (2019). Deep Learning Based
Recommender System: A Survey and
New Perspectives. ACM Computing
Surveys, 52(1), 1-38.
9. Zhang, Y., & Chen, X. (2019).
Explainable Recommendation: A
Survey and New Perspectives.
Foundations and Trends® in
Information Retrieval, 14(1), 1-101.
10. Gupta, P., Goel, A., & Joshi, A.
(2020). Incorporating Sentiment in
Hybrid Recommendation System
Using Convolutional Neural
Networks. Procedia Computer
Science, 173, 394-403.
11. Feng, S., Lv, F., & Sheng, Q. Z.
(2020). A Review of Sentiment
Analysis in Social Media and Its
Applications in Disaster
Management. Information Processing
& Management, 57(5), 102300.
12. Li, S., Xu, Y., Zhang, Y., & Liu, J.
(2022). Deep Learning for Sentiment-
Based Recommendation: Methods,
Applications, and Open Challenges.
IEEE Access, 10, 43562-43574.
13. Kumar, A., & Sharma, N. (2023).
Advances in Sentiment Analysis for
Recommender Systems: Trends,
Tools, and Techniques. Journal of
Artificial Intelligence Research, 74,
150-168.