Movie Recommendation System with Fake Review Detection Using Deep Learning

Abstract

With the rapid development of network technology and entertainment creation, the diversity of movie genres has expanded significantly. Users often face the challenge of selecting movies that align with their preferences. Deep learning, a field that has garnered substantial attention from scholars, offers the potential to model more complex structures due to its deep architecture. This study proposes a personalized movie recommendation system integrating deep learning for enhanced accuracy and tailored suggestions. We combine personalized recommendation and collaborative filtering algorithms, leveraging users’ historical behaviors and big data from the mobile Internet and social media. Collaborative filtering techniques are employed within the context of movie recommendations. Specifically, we utilize the Seq2Seq model based on LSTM recurrent neural networks for deep learning. We also address the issue of fake reviews by integrating deep learning techniques to identify and filter out unreliable reviews. Overall, our system aims to create a more reliable and enjoyable movie-watching experience for users.

Introduction

I. INTRODUCTION

The need for recommendation systems has increased recently in the fast evolving media consumption environment, where consumers are faced with a difficult decision because of the multitude of content that is available. These mechanisms are essential to the process of directing viewers throughout the plethora of cinematic options and guaranteeing that each viewer finds material that complements their own tastes and inclinations. Against this backdrop, our research attempts to address this imperative by presenting a comprehensive, technically rigorous exploration of an advanced movie recommendation framework characterized by the integration of the latest deep learning methodologies.

The collaboration of personalized recommendation techniques with collaborative filtering algorithms, with underpin- ning provided by comprehensive analysis of vast repositories of user-generated data sourced from the dynamic ecosystem of the mobile, Internet and social media platforms is key to our approach. We meticulously scrutinize the historical viewing behaviors, interactions, and preferences of users to distill intricate patterns and discernible trends that underpin individualized content preferences. Through the power of big data analytics, our system embarks on a journey of discovery, crossing the vast terrain of user-generated data to unveil hidden correlations and latent affinities that inform the recommendation process.

Modern deep learning architectures, most notably the Seq2Seq model based on the strong foundations of Long Short-Term Memory (LSTM) recurrent neural networks, are deployed at the core of our approach. By means of this advanced architecture, our system is enabled to absorb, an- alyze, and contextualize the multitude of signals and cues present in user interactions, therefore enabling comprehensive comprehension and interpretation of user preferences. Our solution overcomes the constraints of traditional recommendation models by using the temporal dynamics stored within LSTM networks to capture the changing character of user tastes and preferences over time.

Our research goes further, tackling perhaps the most universal challenge in recommendation systems: fake reviews—an issue threatening the integrity and reliability of recommendation systems. We attempt to create robust mechanisms for identifying and filtering out unreliable feedback through the formidable capabilities of deep learning techniques in a bid to strengthen the trustworthiness and credibility of our recommendation framework. Our study, therefore, hopes to bring about a sea change in the movie recommendation systems—moving toward a paradigm in which more personalization, reliability, and satisfaction of users are promised.

In summary, our research proposes an advanced movie recommendation framework integrating deep learning methodologies to meet the escalating demand for personalized content selection. By merging personalized recommendation techniques with collaborative filtering algorithms and harnessing extensive user-generated data, our system delivers tailored suggestions aligned with individual preferences.

Powered by modern deep learning architectures like the Seq2Seq model with LSTM recurrent neural networks, our approach analyzes user interactions to adapt recommendations over time. Furthermore, we tackle the challenge of fake reviews using  deep  learning  techniques, enhancing  the trustworthiness of our system. Ultimately, our study aims to revolutionize movie recommendation systems, ensuring increased personalization, reliability, and user satisfaction.

II. RELATED WORKS

A. Deep Learning-Based Recommendation System

[1]Careful examination of different studies published from 2018 to February 2023 with emphasis on term classification shows that there is a systematic merging between deep learning techniques and recommendation systems. From the findings, it appears that recommendation systems depend on deep learning techniques like graph neural networks (GNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). Additionally, domain classification emerges in the study as an emerging field of research. This suggested area of investigation involves the application of deep learning methods in domains such as social networks, e-commerce or e-learning. In summary, this review article is very important because it bridges the gap in literature and is helpful for scholars and professionals who aim at using deep learning technologies towards recommendation systems.

B. Multimodal Movie Recommendation System

A personalized multimodal movie recommendation system based on deep learning and multimodal data analysis. The authors used the [2] MovieLens 100K and 1M datasets to test the effectiveness of their proposed algorithm, which outperformed traditional collaborative filtering algorithms such as user-based, item-based, and singular-value decomposition approaches. The key contributions of this work include: (1) introducing a novel movie recommendation system that combines deep learning and multimodal data analysis, (2) reporting experimental results that demonstrate the good performance of the proposed system, and (3) suggesting that incorporating multimodal data, such as movie posters, can further improve the recommendation system’s performance. The authors conclude that integrating deep learning and multimodal data analysis can provide better personalized recommendations and alleviate the sparse data problem in movie recommendation systems.

???????C. Parameter Tuning Via Co-Clustering

A movie recommendation system on the basis of parameter tuning of user and movie neighborhoods through co- clustering. The proposed approach is framed to forecast missing ratings in movie recommender system, where users and movies are grouped together to improve personalized recommendations. The experiments were conducted with the MovieLens datasets and [3] the results show that the proposed system performs better than existing systems with MAE and RMSE of 0.7461 and 0.9537 respectively. Additionally, the authors also discuss clustering, co-clustering, collaborative filtering and evaluation metrics for movie recommender systems in related work. Finally, potentiality of this approach is highlighted and future directions for improvement such as incorporating co-clustering with deep learning techniques for better performance are discussed in conclusion section.

???????D. Higher-Order Tensor Singular-Value-Decomposition

The paper “Personalized Multimedia Recommendation Systems Using Higher-Order Tensor Singular-Value- Decomposition” introduces [4] a new technique for personalized recommendation systems, which uses HOT-SVD to improve the precision of recommendations. This model solves the problem of sparse data and allows developing context-aware solutions by merging information from different modalities such as text, images, audio. With respect to multimedia recommendation tasks, this approach explores multimodal data fusion and deep learning techniques. The authors explain that the model is efficient in handling sparse data; it can also be used to provide context-aware suggestions by capturing complex relationships among users, items and features at higher levels of interactions through tensor-based methods.

???????E. Personalized Meal Recommendation System

The innovative framework for meal suggestions and menu planning represents a leap further forward in dietary systems by including human feelings and preferences in its core methodology. [5] By applying both social-affective computing and EEG signal analysis, the system has effectively captured and analyzed emotional responses to food items. This pre- dictive capability is enhanced with the use of a complex hierarchical ensemble approach that integrates various feature extraction methods for accuracy.

The system then proceeds to generate personalized meal suggestions through the use of the TOPSIS method, which is used for very careful item ranking on account of affectivity. This feature is especially useful for people in critical conditions who may not be able to express what they prefer.

Furthermore, the system automatically does menu planning, considering parameters such as energy intake requirements, emotional reactions, and nutrition values.

Optimization is achieved by the application of the bin-packing algorithm to ensure not only nutritional balance in each meal and snack but also that the food is acceptable emotionally. This approach makes good promises to provide a dining experience that harmonizes with physiological and emotional levels of wellness.

???????F. Actor Based Matrix Computations

The proposal of the actor-based recommendation system enhances movie suggestions by using content-based filtering. It employs [7] detailed filmography information and genre metadata from a dataset of 509 South Korean movies. User preferences are analyzed by studying their historical interactions such as ratings, watch history to identify patterns in favored genres, actors, and directors.

The system extracts various movie features with emphasis on the actors themselves and their associated meta data. These characteristics are used for actor-based calculations that estimate the likeness between movies . This process involves computing similarity scores based on these extracted features but with greater stress on common actors and genres. Then movies that have higher similarity scores are recommended to the user.

Its performance is then compared against traditional genre- oriented recommendation systems to ascertain if it is effective in giving more personalized and accurate recommendations.

???????G. Hybrid Recommendation System

The text discusses the importance of recommendation systems in the modern digital age, where the abundance of information can lead to information overload. It introduces [8] a novel hybrid recommendation framework that combines collaborative filtering, content-based filtering, and self- organizing map neural network techniques to improve the accuracy and efficiency of traditional collaborative filtering methods. The proposed system is evaluated on a subset of the Movies Database, and the results demonstrate that it outperforms state-of-the-art methods in terms of accuracy and precision.

???????H. Recommendation with Collaborative Filtering

Collaborative filtering provides insight into user preferences and behaviours. There are two main methodologies in the recommendation system: content-based and collaborative filtering. [10] On one hand, content-based filtering extracts recommendations from a user’s past interactions with the platform, such as his or her search history or purchase records. On the other hand, collaborative filtering suggests items based on the choices made by other users with similar tastes and is more like seeking movie recommendations from friends who like the same movies as you. Unlike content-based approaches, collaborative filtering is totally blind to item features and depends solely on user behavior to bring up a set of personalized recommendations. The biggest strength it has is in exposing users to a broad range of choices outside their historical preferences and helping discover new genres and movies. [9]There are also models that combine both content-based and collaborative filtering approaches. Collaborative filtering can thus add to the user experience through serendipitous discoveries while broadening horizons of choice in entertainment preferences.

III. PROPOSED MODEL

This system uses collaborative filtering for personalized movie recommendations to users based on past interaction histories and reviews across various genres of movies. Users also get much more enriched movie experiences because the recommendations made are much more personalized to the tastes and preferences of the users. Another added advantage of our proposed recommendation system is that it has a fake review detection system that makes our recommendation system more honest and reliable.

The system can detect and remove fake reviews, keeping recommendations that it will make to users of a higher quality and more trustworthy. Yet another added advantage is the availability of a user- centric module in our proposed system. This module is very important as far as user engagement is concerned and acts as an interacting module of the system with users. This fa- cilitates seamless communication between the system and the users for gathering feedback, preferences, and other relevant information about recommendation processing. Through this approach, we ensure recommendation accuracy and relevance to much-extended satisfaction and value to our users.

???????A. Data Collection

The project commenced with the collection of relevant datasets to facilitate the development of the movie recommendation system and the fake review detection component. Datasets were gathered from reputable sources such as MovieLens, IMDb, and Rotten Tomatoes. These datasets contain user ratings for various movies, which served as the basis for collaborative filtering-based recommendation algorithms. Additionally, critic review datasets from online review platforms like Rotten Tomatoes were acquired for training the seq2seq model for fake review detection.

???????B. Data Preprocessing

Prior to implementing the recommendation system and fake review detection component, extensive data preprocessing was conducted to clean and prepare the datasets for analysis. In the case of movie ratings datasets, preprocessing steps included removing duplicate entries, handling missing values, and normalizing ratings to ensure consistency across the dataset. For critic review datasets from Rotten Tomatoes, text preprocessing techniques such as tokenization, stop word removal, and stemming or lemmatization were applied to standardize the textual data and improve model performance. Furthermore, the datasets were split into training, validation, and testing sets to facilitate model training and evaluation.

???????C. Movie Recommendation System Development

The movie recommendation system employs collaborative filtering to give personalized movie suggestions. With user ratings, the system finds patterns and preferences and compute similarity scores to generate relevant recommendations. Collaborative filtering involves a comparison of a person’s rating of the movie with that of others to come to similar tastes and thereby enhance the recommendation accuracy. This method captures the nuanced preferences of users and makes the system recommend movies that are most consonant with the interests of the viewers. The system architecture allows dynamic interaction of users, and the recommendations are refined continuously whenever new ratings are added. Coupled with that, the user interface is designed for maximum user engagement, making navigation easy and rating submission efficient. All these combinations of more advanced algorithmic techniques with user-centric design deliver a very high-quality, personalized experience of movie recommendation.

???????D. Sequence-To-Sequence Model

The seq2seq model is a style of architecture of neural networks transformed from one sequence of data into another. It comprises an encoder and a decoder. The former processes an input sequence and encodes it into a context vector of fixed length. The decoder uses the context vector to generate the output sequence. This model is applied in many areas, from machine translation and text summarization to conversational agents. In a movie recommendation system, the seq2seq model can be used to enhance recommendations by analyzing user reviews. The model can be trained to uncover fake reviews by learning patterns that indicate real and fake content. By developing fake review detection, the system can filter out unreliable reviews and provide recommendations only based on genuine feedback from the users. Thus we are left with recommendations that are more accurate and reliable, ultimately improving user satisfaction and trust in the recommendation system. There are various applications concerning seq2seq model including that of [6]Human Action Recognition, chatbots and virtual assistants, image captioning and so on.

???????IV. RESULTS

The main outcomes of the project include development of the movie recommendation system and the implementation of the fake review detection component. Data collection and preprocessing prepared datasets from reputed sources such as MovieLens, IMDb, and Rotten Tomatoes for analysis, makes the data both reliable and of good quality. The implemented movie recommendation system, with collaborative filtering, aimed at offering personalized movie recommendations based on user ratings. The user experience was comfortable with easy navigation and functionality for submitting ratings. The set of movies similar to those already rated by users was correctly identified by collaborative filtering. Even though the full fake review detection component by using a seq2seq model, trained on a dataset of critic reviews, was not implemented in full, good progress was made in the training of the model and reviews classified as fake or real, based on certain keywords in the reviews. While this paper primarily focuses on the successful aspects of the movie recommendation system, iterations of future projects may be carried out to include refinement and improvement of the fake review detection.

???????V. FUTURE SCOPE

Integrating collaborative filtering with fake review detection will lead to more accurate recommendations. It filters out fake or biased reviews and ensures that user preferences are based on genuine feedback to bring out more reliable recommendations.

1. Integration with Home Monitoring Systems: Extend the project to seamlessly integrate with home monitoring systems, enabling elderly users to access our software directly on monitor screens within their homes. Enhance accessibility and promote continuous engagement with our platform by integrating it into the daily lives of elderly individuals.
2. Improved Personalization: The integration of collabora- tive filtering and fake review detection will let the system move a step ahead in providing recommendations person- alized for each user. It can analyze genuine interactions and behaviors of the users while filtering out the fake reviews to understand the tastes of each individual better and recommend movies that are more in tune with their interests.
3. Mitigating Information Bias: Fake review detection helps mitigate the influence of biased or manipulated informa- tion on recommendation accuracy. In this way, the system will be able to avoid recommendations being decided through fake, deceitful, or misleading information.
4. Adoption of Advanced Deep Learning Techniques: Fur- ther development of the fake review detection component could involve leveraging more advanced deep learning techniques, such as reinforcement learning or attention mechanisms. Reinforcement learning could make the system adapt and optimize the fake review detection strategies based on user feedback and changing patterns of deception over time. Attention mechanisms could make the model more accurate and reliable by improving its capability to identify key features or patterns in the texts of reviews.
5. Hybrid Recommendation Approaches: The exploration of hybrid recommendation approaches that combine the col- laborative filtering approach with content-based filtering or hybrid models can further enhance recommendation accuracy and diversity. Integrating multiple recommendation strategies will make it possible for the system to leverage the strengths of each approach into more comprehensive and nuanced recommendations.
6. User Trust and Satisfaction: In all, the integration of collaborative filtering with fake review detection tries to improve user trust and satisfaction in the recommendation system. With better accuracy in personalized recommendations based on genuine user feedback, the system is more likely to enhance the overall user experience and foster long-term engagement and loyalty.

To summarize, combining collaborative filtering with fake review detection can greatly enhance users’ trust and satisfaction in the recommendation system. The system tries to refine this experience with better accuracy in personalized recommendations derived from genuine user feedback, hence improving the quality of overall user experiences and achieving long-term engagement and loyalty. This integration will help attenuate the influence of biased or manipulated information, which improves the reliability and effectiveness of the recommendation system. It, therefore, tries to inculcate a greater sense of trust and confidence for the users, which sustains user engagement and brings about perpetual success.

Conclusion

We have created a recommendation system that embodies both personalized movie recommendations with fake review detection. Our approach applies collaborative filtering to provide movies recommended in a manner that is congenial and suitable to individual tastes and rating history. This approach not only helps to find accurate recommendations but also makes the movie experience more engaging and satisfying for every user. We have also incorporated fake review detection using seq2seq models, which assure to filter out fake reviews. This makes the recommendations really pegged on actual user feedback, thus making the system more reliable and trustworthy. With these techniques combined, there is a highly reliable recommendation system that satisfies the user’s preferences with great precision and ensures the feedback used to generate such recommendations upholds integrity to make the users more satisfied and trust the platform.

References

[1] C. Li, I. Ishak, H. Ibrahim, M. Zolkepli, F. Sidi and C. Li, ”Deep Learning-Based Recommendation System: Systematic Review and Classification,” in IEEE Access, vol. 11, pp. 113790-113835, 2023, doi: 10.1109/ACCESS.2023.3323353. keywords: Recommender systems; Deep learning; Systematics; Surveys; Measurement; Electronic learning; Electronic commerce; Classification algorithms; Systematics; Deep learning; term classification; recommendation system; systematic review; state-of-the-art techniques, [2] Mu, Y.; Wu, Y. Multimodal Movie Recommendation System Using Deep Learning. Mathematics 2023, 11, 895. https://doi.org/10.3390/math11040895 , Sonu Airen, Jitendra Agrawal, Movie Recommender System Using Parameter Tuning of User and Movie Neighbourhood via Co- Clustering, Procedia Computer Science, Volume 218, 2023, Pages 1176-1183, ISSN 1877-0509, https://doi.org/10.1016/j.procs.2023.01.096. (https://www.sciencedirect.com/science/article/pii/S1877050923000960), S. Y. Chang, H. -C. Wu, K. Yan, X. Chen and Y. Wu, ”Novel Personalized Multimedia Recommendation Systems Using Tensor Singular-Value-Decomposition,” 2023 IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), Beijing, China, 2023, pp. 1-7, doi: 10.1109/BMSB58369.2023.10211188. [3] keywords: Tensors;Social networking (online);Multimedia systems;Multimedia databases;Motion pictures;Robustness;Internet of Things;Personalized multimedia recommendation system (PMRS);tensor singular-value-decomposition;normalized root-mean-square error (NRMSE);E-commerce, [4] T. Islam, A. R. Joyita, M. G. R. Alam, M. Mehedi Hassan, M. R. Hassan and R. Gravina, ”Human-Behavior-Based Personalized Meal Recommendation and Menu Planning Social System,” in IEEE Transactions on Computational Social Systems, vol. 10, no. 4, pp. 2099-2110, Aug. 2023, doi: 10.1109/TCSS.2022.3213506. keywords:

Copyright

Copyright © 2024 Varsha Harikumar, Vighnesh O K, Vignesh Vijayan, Jeena Joy, Rotney Roy Meckamalil. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.