Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101037
Artur F. Tomeczek
Microsoft’s acquisition of Activision Blizzard valued at $68.7 billion ($95 per share) has drastically altered the landscape of the video game industry. At the time of the takeover, the intellectual properties of Activision Blizzard included World of Warcraft, Diablo, Hearthstone, StarCraft, Overwatch, Battle.net, Candy Crush Saga, and Call of Duty. This article aims to explore the patenting activity of Activision Blizzard between 2008 (the original merger) and 2023 (the Microsoft acquisition). Four IPC code co-occurrence networks (co-classification maps) are constructed and analyzed based on the patent data downloaded from the WIPO Patentscope database. International Patent Classification (IPC) codes are a language agnostic system for the classification of patents. When multiple IPC codes co-occur in a patent, it shows that the technologies are connected. These relationships can be used for patent mapping. The analysis identifies the prolific and bridging technologies of Activision Blizzard and explores its synergistic role as a subsidiary of Microsoft Corporation.
{"title":"Innovative activities of activision blizzard: A patent network analysis","authors":"Artur F. Tomeczek","doi":"10.1016/j.entcom.2025.101037","DOIUrl":"10.1016/j.entcom.2025.101037","url":null,"abstract":"<div><div>Microsoft’s acquisition of Activision Blizzard valued at $68.7 billion ($95 per share) has drastically altered the landscape of the video game industry. At the time of the takeover, the intellectual properties of Activision Blizzard included World of Warcraft, Diablo, Hearthstone, StarCraft, Overwatch, Battle.net, Candy Crush Saga, and Call of Duty. This article aims to explore the patenting activity of Activision Blizzard between 2008 (the original merger) and 2023 (the Microsoft acquisition). Four IPC code co-occurrence networks (co-classification maps) are constructed and analyzed based on the patent data downloaded from the WIPO Patentscope database. International Patent Classification (IPC) codes are a language agnostic system for the classification of patents. When multiple IPC codes co-occur in a patent, it shows that the technologies are connected. These relationships can be used for patent mapping. The analysis identifies the prolific and bridging technologies of Activision Blizzard and explores its synergistic role as a subsidiary of Microsoft Corporation.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101037"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145361390","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101035
Zhixing Guo , Xiangshi Ren , Xinxin Ma
Game difficulty is a significant focus within game design research. However, how to effectively measure players’ perceived difficulty (subjective game difficulty, SGD) remains unresolved. Currently, SGD is primarily assessed by the player’s difficulty rating, self-reporting, and physiological measurements. However, these measuring methods are limited in their ability to capture the complex structure of SGD for complete and precise evaluation. Therefore, this study develops a new scale to measure SGD. We first identified and classified the structure of SGD in six dimensions. On this basis, the Subjective Game Difficulty Scale (SGDS) was developed and validated through a standard three-stage scale development method. Sixty related items were generated, and thirty-three items were selected in the first two stages. In the third stage, an international survey with 326 American (USA), Chinese, and Japanese participants was conducted to test the scale. The results indicated that our final 25-item SGDS is reliable and valid. We discussed the usability of the SGDS, compared it with other measurement methods, and we provided design implications and the guidance regarding how to apply the SGDS. This work presents a promising instrument for game designers and researchers to support game difficulty evaluation and design.
{"title":"Measuring perceived difficulty in video games: Development of the subjective game difficulty scale","authors":"Zhixing Guo , Xiangshi Ren , Xinxin Ma","doi":"10.1016/j.entcom.2025.101035","DOIUrl":"10.1016/j.entcom.2025.101035","url":null,"abstract":"<div><div>Game difficulty is a significant focus within game design research. However, how to effectively measure players’ perceived difficulty (subjective game difficulty, SGD) remains unresolved. Currently, SGD is primarily assessed by the player’s difficulty rating, self-reporting, and physiological measurements. However, these measuring methods are limited in their ability to capture the complex structure of SGD for complete and precise evaluation. Therefore, this study develops a new scale to measure SGD. We first identified and classified the structure of SGD in six dimensions. On this basis, the Subjective Game Difficulty Scale (SGDS) was developed and validated through a standard three-stage scale development method. Sixty related items were generated, and thirty-three items were selected in the first two stages. In the third stage, an international survey with 326 American (USA), Chinese, and Japanese participants was conducted to test the scale. The results indicated that our final 25-item SGDS is reliable and valid. We discussed the usability of the SGDS, compared it with other measurement methods, and we provided design implications and the guidance regarding how to apply the SGDS. This work presents a promising instrument for game designers and researchers to support game difficulty evaluation and design.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101035"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145361394","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101020
Patrícia Alves , João Trindade , Gonçalo Monteiro , Pedro Campos , Pedro Saraiva , Goreti Marreiros , Paulo Novais
Accurately determining someone’s personality is complex and often requires lengthy questionnaires, which are subject to social desirability bias, or a great amount of users’ interactions with the system. Also, most existing research focuses on broader personality dimensions rather than more granular personality traits, which better characterize a person.
In this work, we propose to implicitly acquire the users’ granular personality traits using mobile short-duration serious games, in < 5 min and in a single play interaction, namely cautiousness and achievement-striving as concept proof, to replace personality questionnaires.
Two platform mobile games were developed, one for each trait, Which Way and Time Travel, respectively. Then, an experiment with real participants (n = 100) was conducted. Time Travel proved to be capable of detecting achievers (get all coins, diamonds, and better scores), while Which Way couldn’t effectively measure cautiousness, although following hard paths could be related to less cautious persons. As expected, significant correlations with other personality traits were also found (15 out of 30), such as anger, modesty, excitement seeking, and adventurousness. Contrary to other types of (serious) games, the results show short-duration mobile minigames are a viable way of unobtrusively determining the users’ granular personality, being the path to replacing personality questionnaires.
{"title":"“You Want to Play a Game?” Detecting Two Personality Traits with Short-Duration Mobile Games","authors":"Patrícia Alves , João Trindade , Gonçalo Monteiro , Pedro Campos , Pedro Saraiva , Goreti Marreiros , Paulo Novais","doi":"10.1016/j.entcom.2025.101020","DOIUrl":"10.1016/j.entcom.2025.101020","url":null,"abstract":"<div><div>Accurately determining someone’s personality is complex and often requires lengthy questionnaires, which are subject to social desirability bias, or a great amount of users’ interactions with the system. Also, most existing research focuses on broader personality dimensions rather than more granular personality traits, which better characterize a person.</div><div>In this work, we propose to implicitly acquire the users’ granular personality traits using mobile short-duration serious games, in < 5 min and in a single play interaction, namely cautiousness and achievement-striving as concept proof, to replace personality questionnaires.</div><div>Two platform mobile games were developed, one for each trait, Which Way and Time Travel, respectively. Then, an experiment with real participants (n = 100) was conducted. Time Travel proved to be capable of detecting achievers (get all coins, diamonds, and better scores), while Which Way couldn’t effectively measure cautiousness, although following hard paths could be related to less cautious persons. As expected, significant correlations with other personality traits were also found (15 out of 30), such as anger, modesty, excitement seeking, and adventurousness. Contrary to other types of (serious) games, the results show short-duration mobile minigames are a viable way of unobtrusively determining the users’ granular personality, being the path to replacing personality questionnaires.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101020"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145157426","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101015
Fengtian Shao
This study explores the application of artificial intelligence to enhance the representation and evaluation of Chinese cultural elements in animated films, emphasizing both cultural significance and market potential while redefining intellectual property (IP) value in the industry. A major challenge addressed is the accurate assessment of cultural content, as traditional Back Propagation Neural Networks (BPNNs) often suffer from slow convergence and local minima issues. To overcome these limitations, the research proposes an improved GA-BP model, combining BPNN’s localized optimization with the global search capabilities of Genetic Algorithms (GA). The paper reviews cultural development theories and examines the status of Chinese and international animation IPs. Experimental results show that the GA-BP model achieves higher accuracy and stability than standard BPNNs, closely matching expert evaluations. This validates its effectiveness in supporting intelligent cultural evaluation and creative design in animation. By applying AI techniques to cultural evaluation, the research applies artificial intelligence methods to evaluate and support the structured integration of cultural elements into animated film design, laying a methodological groundwork for innovation in Chinese animated films. It supports cultural sustainability and strengthens national cultural identity through digital storytelling, contributing to both academic inquiry and industry practice.
{"title":"The expression method of Chinese creative elements in animation films based on artificial intelligence technology","authors":"Fengtian Shao","doi":"10.1016/j.entcom.2025.101015","DOIUrl":"10.1016/j.entcom.2025.101015","url":null,"abstract":"<div><div>This study explores the application of artificial intelligence to enhance the representation and evaluation of Chinese cultural elements in animated films, emphasizing both cultural significance and market potential while redefining intellectual property (IP) value in the industry. A major challenge addressed is the accurate assessment of cultural content, as traditional Back Propagation Neural Networks (BPNNs) often suffer from slow convergence and local minima issues. To overcome these limitations, the research proposes an improved GA-BP model, combining BPNN’s localized optimization with the global search capabilities of Genetic Algorithms (GA). The paper reviews cultural development theories and examines the status of Chinese and international animation IPs. Experimental results show that the GA-BP model achieves higher accuracy and stability than standard BPNNs, closely matching expert evaluations. This validates its effectiveness in supporting intelligent cultural evaluation and creative design in animation. By applying AI techniques to cultural evaluation, the research applies artificial intelligence methods to evaluate and support the structured integration of cultural elements into animated film design, laying a methodological groundwork for innovation in Chinese animated films. It supports cultural sustainability and strengthens national cultural identity through digital storytelling, contributing to both academic inquiry and industry practice.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101015"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144922365","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101043
Karl Cini , John Abela
The film industry is an important entertainment avenue for audiences of all ages. Demand for good quality scripts remains a core element of this industry, rendering the screenplay a pivotal tool at the green lighting stage.
While previous work addressed isolated elements influencing the performance of a movie, this research aims to bring together known influential factors and some novel approaches by applying Natural Language Processing (NLP) and Machine Learning (ML) techniques to analyse movie scripts, with the aim of extracting valuable insights and patterns that are able to predict the audience rating as collated by the Internet Movie Database (IMDb).
This research helps producers determine which movies are most viable for financing. By providing a sound method to sift through and rank the various script projects presented to them, they can focus on scripts that are likely to perform better.
Methods adopted in this research include the use of lexicons for the extraction of linguistic features, the analysis of emotional arcs in movies, embedding strategies for the script and statistical features generated from sentiment analysis. These features are concatenated to cast and crew specific factors to train various regression models by using a forward rolling window training strategy.
{"title":"Forecasting film audience ratings: A natural language processing approach to script and production data","authors":"Karl Cini , John Abela","doi":"10.1016/j.entcom.2025.101043","DOIUrl":"10.1016/j.entcom.2025.101043","url":null,"abstract":"<div><div>The film industry is an important entertainment avenue for audiences of all ages. Demand for good quality scripts remains a core element of this industry, rendering the screenplay a pivotal tool at the green lighting stage.</div><div>While previous work addressed isolated elements influencing the performance of a movie, this research aims to bring together known influential factors and some novel approaches by applying Natural Language Processing (NLP) and Machine Learning (ML) techniques to analyse movie scripts, with the aim of extracting valuable insights and patterns that are able to predict the audience rating as collated by the Internet Movie Database (IMDb).</div><div>This research helps producers determine which movies are most viable for financing. By providing a sound method to sift through and rank the various script projects presented to them, they can focus on scripts that are likely to perform better.</div><div>Methods adopted in this research include the use of lexicons for the extraction of linguistic features, the analysis of emotional arcs in movies, embedding strategies for the script and statistical features generated from sentiment analysis. These features are concatenated to cast and crew specific factors to train various regression models by using a forward rolling window training strategy.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101043"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145361395","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101046
Xinge Tong , Ian Willcock , Yi Sun
Online game reviews are a category of customer feedback for video games that contain valuable information for either a game’s development team or its potential players. Currently, Steam is the most popular video game distribution platform, encouraging players to provide feedback by leaving reviews on the store page. However, little research has been conducted to determine the structures and characteristics of these user reviews on Steam. This paper aims to identify the types of information contained within these reviews, as well as how this information is structured. It takes the game No Man’s Sky as a case study and employs both qualitative textual analysis and the latent Dirichlet allocation topic model method. The contribution of this study is to propose a baseline model of a game review by identifying their generalisable characteristics to support better natural language analysis of Steam game reviews. Our results show that game reviews on Steam are characterised by their capacity to include mixed information. The review data analysis should account for this diversity of meaning to accurately summarise players’ views.
{"title":"What is a game review: A case study approach to defining player reviews","authors":"Xinge Tong , Ian Willcock , Yi Sun","doi":"10.1016/j.entcom.2025.101046","DOIUrl":"10.1016/j.entcom.2025.101046","url":null,"abstract":"<div><div>Online game reviews are a category of customer feedback for video games that contain valuable information for either a game’s development team or its potential players. Currently, Steam is the most popular video game distribution platform, encouraging players to provide feedback by leaving reviews on the store page. However, little research has been conducted to determine the structures and characteristics of these user reviews on Steam. This paper aims to identify the types of information contained within these reviews, as well as how this information is structured. It takes the game <em>No Man’s Sky</em> as a case study and employs both qualitative textual analysis and the latent Dirichlet allocation topic model method. The contribution of this study is to propose a baseline model of a game review by identifying their generalisable characteristics to support better natural language analysis of Steam game reviews. Our results show that game reviews on Steam are characterised by their capacity to include mixed information. The review data analysis should account for this diversity of meaning to accurately summarise players’ views.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101046"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145415066","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101044
Chang Ma, Jingbo Shao, Pengyu Li
As a dominant monetization mechanism in mobile gaming, gacha systems have raised significant concerns regarding their behavioral impacts due to the probabilistic nature of virtual item acquisition. While previous studies have drawn parallels between gacha engagement and gambling-related disorders, this research adopts a behavioral economics lens to investigate the determinants of user participation in this prevalent virtual economy model. Through a randomized controlled trial (N = 457), we systematically examine how pricing strategies (single-pull gacha cost) and pity systems (guaranteed prize mechanisms) interact to shape players’ intention to pay. Empirical evidence reveals that both pricing strategies of single gacha and pity systems significantly impact the spending intentions through cognitive reappraisal of perceived risk. Notably, individual differences in belief of luck patterns emerged as critical moderators. This study quantifies the economic interplay between system architecture and superstitious cognition, and provides evidence-based recommendations for gacha mechanisms design and regulatory interventions targeting compulsive spending patterns in digital environments.
{"title":"Monetization mechanisms in gacha games: The behavioral triad of pricing strategies, pity systems, and belief of luck","authors":"Chang Ma, Jingbo Shao, Pengyu Li","doi":"10.1016/j.entcom.2025.101044","DOIUrl":"10.1016/j.entcom.2025.101044","url":null,"abstract":"<div><div>As a dominant monetization mechanism in mobile gaming, gacha systems have raised significant concerns regarding their behavioral impacts due to the probabilistic nature of virtual item acquisition. While previous studies have drawn parallels between gacha engagement and gambling-related disorders, this research adopts a behavioral economics lens to investigate the determinants of user participation in this prevalent virtual economy model. Through a randomized controlled trial (N = 457), we systematically examine how pricing strategies (single-pull gacha cost) and pity systems (guaranteed prize mechanisms) interact to shape players’ intention to pay. Empirical evidence reveals that both pricing strategies of single gacha and pity systems significantly impact the spending intentions through cognitive reappraisal of perceived risk. Notably, individual differences in belief of luck patterns emerged as critical moderators. This study quantifies the economic interplay between system architecture and superstitious cognition, and provides evidence-based recommendations for gacha mechanisms design and regulatory interventions targeting compulsive spending patterns in digital environments.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101044"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145415067","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The recreation of looks established by favorite movie characters or fashion icons is a popular trend in this decade. It is difficult to find out the dresses and accessories required to develop that look as current product recommendations are mostly based on history of users’ choices. There exists computer vision-based solutions that check image-wise similarities between the desired looks and available fashion products from e-commerce stores. However, this is a resource hungry complex process as plenty of product images would be analyzed. In this work an NLP-based lightweight look recommendation system is proposed. In the proposed approach, multiple text descriptions of trendy looks are collected from different websites to build the training dataset. A subset of two benchmark datasets (Myntra Products Dataset and Ajio Products Dataset) have been used for recommendation. Using the bag of words technique, text datasets are embedded, and a set of looks is recommended for each product. The system is validated using Cosine similarity and Cohen’s kappa metrics. Products in the test dataset have been mapped to their 1st and 2nd highest recommended looks with positive scores. We observed a minimum score of 0.6 and 0.2 for Cosine similarity and Cohen’s kappa respectively, representing appreciable performance.
{"title":"Lock the look: Recommending trendy looks for fashion products using natural language processing","authors":"Manjarini Mallik , Tushti Thakur , Chandreyee Chowdhury","doi":"10.1016/j.entcom.2025.101049","DOIUrl":"10.1016/j.entcom.2025.101049","url":null,"abstract":"<div><div>The recreation of looks established by favorite movie characters or fashion icons is a popular trend in this decade. It is difficult to find out the dresses and accessories required to develop that look as current product recommendations are mostly based on history of users’ choices. There exists computer vision-based solutions that check image-wise similarities between the desired looks and available fashion products from e-commerce stores. However, this is a resource hungry complex process as plenty of product images would be analyzed. In this work an NLP-based lightweight look recommendation system is proposed. In the proposed approach, multiple text descriptions of trendy looks are collected from different websites to build the training dataset. A subset of two benchmark datasets (Myntra Products Dataset and Ajio Products Dataset) have been used for recommendation. Using the bag of words technique, text datasets are embedded, and a set of looks is recommended for each product. The system is validated using Cosine similarity and Cohen’s kappa metrics. Products in the test dataset have been mapped to their 1st and 2nd highest recommended looks with positive scores. We observed a minimum score of 0.6 and 0.2 for Cosine similarity and Cohen’s kappa respectively, representing appreciable performance.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101049"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145465462","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Social media hate speech is an emerging issue, and there is a need to create automatic systems to identify and mitigate its effects. The rapid expansion of social media platforms, especially Twitter, has facilitated the dissemination of hate speech, presenting a major challenge for online communities. Such speech can have severe social and psychological consequences, including inciting violence, promoting extremism, and affecting mental health. Thus, it is essential to manage hateful content on Twitter. This paper presents an ensemble deep learning model that combines BiLSTM and BiGRU to enhance prediction accuracy and robustness. The model achieved 98.56% accuracy rate and demonstrated better generalization than existing methods, proving its effectiveness in identifying hate speech with fewer false positives. This paper offers a powerful tool for detecting and preventing harmful online behavior, contributing to a safer and more inclusive digital space.
{"title":"Hateful tweet detection using a BiLSTM-BiGRU: An ensemble perspective","authors":"Imandi Tejaswini , Venkata Gayathri Ganivada , Appala Srinuvasu Muttipati","doi":"10.1016/j.entcom.2025.101019","DOIUrl":"10.1016/j.entcom.2025.101019","url":null,"abstract":"<div><div>Social media hate speech is an emerging issue, and there is a need to create automatic systems to identify and mitigate its effects. The rapid expansion of social media platforms, especially Twitter, has facilitated the dissemination of hate speech, presenting a major challenge for online communities. Such speech can have severe social and psychological consequences, including inciting violence, promoting extremism, and affecting mental health. Thus, it is essential to manage hateful content on Twitter. This paper presents an ensemble deep learning model that combines BiLSTM and BiGRU to enhance prediction accuracy and robustness. The model achieved 98.56% accuracy rate and demonstrated better generalization than existing methods, proving its effectiveness in identifying hate speech with fewer false positives. This paper offers a powerful tool for detecting and preventing harmful online behavior, contributing to a safer and more inclusive digital space.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101019"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145120833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-01DOI: 10.1016/j.entcom.2025.101052
Firat Ismailoglu
Turkey is the third largest exporter of TV series worldwide. However, half of these series are cancelled early leading to economic and social consequences. In this study, we explore whether the success of these series can be predicted from the scripts of their first episodes using LLMs. We built a dataset of first-episode scripts from recently aired Turkish series and trained LLM-based models on it. The main challenge we faced is that these scripts are very long, making them unsuitable for standard BERT models. This led to one of the key contributions of our study, as there is currently no research that specifically focuses on handling long Turkish texts. We pretrained a BigBird model from scratch for Turkish and fine-tuned it for our task. We also developed a Hierarchical Attention Network (HAN) model capable of processing long Turkish texts. While predicting the exact number of episodes is difficult, both HAN and BigBird achieve strong performance in binary classification setup, distinguishing successful series from unsuccessful ones. Additionally, we investigate whether audience preferences in Turkey have changed over time by testing our models on some iconic older Turkish series to see if they would still be classified as successful by today’s standards.
{"title":"Can LLMs predict the success of Turkish TV series from their first episodes?","authors":"Firat Ismailoglu","doi":"10.1016/j.entcom.2025.101052","DOIUrl":"10.1016/j.entcom.2025.101052","url":null,"abstract":"<div><div>Turkey is the third largest exporter of TV series worldwide. However, half of these series are cancelled early leading to economic and social consequences. In this study, we explore whether the success of these series can be predicted from the scripts of their first episodes using LLMs. We built a dataset of first-episode scripts from recently aired Turkish series and trained LLM-based models on it. The main challenge we faced is that these scripts are very long, making them unsuitable for standard BERT models. This led to one of the key contributions of our study, as there is currently no research that specifically focuses on handling long Turkish texts. We pretrained a BigBird model from scratch for Turkish and fine-tuned it for our task. We also developed a Hierarchical Attention Network (HAN) model capable of processing long Turkish texts. While predicting the exact number of episodes is difficult, both HAN and BigBird achieve strong performance in binary classification setup, distinguishing successful series from unsuccessful ones. Additionally, we investigate whether audience preferences in Turkey have changed over time by testing our models on some iconic older Turkish series to see if they would still be classified as successful by today’s standards.</div></div>","PeriodicalId":55997,"journal":{"name":"Entertainment Computing","volume":"55 ","pages":"Article 101052"},"PeriodicalIF":2.4,"publicationDate":"2025-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145519277","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号