Τransformer neural networks for predicting movie ratings
Τransformer νευρωνικά δίκτυα για πρόβλεψη βαθμολογιών ταινιών
View/
Keywords
Sentiment analysis ; BERT embeddings ; Neural networks ; Natural Language Processing (NLP) ; Deep learning ; Transformer neural networks ; Movie rating predictionAbstract
Implementing an artificial intelligence system for predicting movie ratings by utilizing text content from user reviews is a complex challenge due to the tone, emotion and innuendo that can be embedded in natural human language. Traditional approaches of designing prediction models cannot detect irony, sarcasm and other “hidden” emotional meanings that may be present in brief review texts written by humans, resulting in incorrect predictions.
Designing and refining a method for such prediction can serve many purposes. A practical application could be to enhance movie recommendation systems, allowing various content delivery services and streaming platforms to recommend films not only based on numerical ratings and “like” buttons, but also on the sentiment expressed in text reviews. At the same time, the implementation of a system like this could lead to the automation of review analysis, which would analyze in greater depth the individual factors that make a movie successful or unsuccessful, providing guidance to producers regarding audience preferences. Furthermore, current IMDb ratings are based on human opinions, which often may not reflect the author’s objective opinion and may be biased or influenced by external factors such as excessive advertising, monetization, controversies, or fan loyalty. This creates a need for a more objective rating method.
This paper examines a method that utilizes the ability of the BERT model to encode language context, with the flexibility offered by machine learning architectures to improve rating prediction accuracy. The proposed system aims to find the connection between the emotional content of review texts and numerical ratings. This combination of technologies can lead to improved accuracy and reliability of rating predictions, offering a more effective solution than traditional methods.
Department
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα ΠληροφορικήςNumber of pages
54Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα