Article: ST-LSTM-sports mind: a multimodal deep learning framework for intelligent sports analytics and automated journalism Journal: International Journal of Computer Applications in Technology (IJCAT) 2026 Vol.78 No.7 pp.42 - 55 Abstract: This research introduces an AI-driven framework designed to automate the generation of sports highlights, optimising content creation for digital platforms. The framework utilises advanced deep learning techniques, including Spatial-Temporal Long Short-Term Memory (ST-LSTM) networks and convolutional neural networks (CNN), to address key challenges in sports classification, event detection, and player tracking. By incorporating multimodal data (audio and visual cues), the model achieves high accuracy rates, with 93% for goals, 90% for substitutions, and 86% for cards. However, further work is necessary to achieve 100% prediction accuracy for officially sanctioned events. The study also explores the integration of audio features to improve detection, particularly for dynamic events with strong audio cues, while acknowledging challenges with weak or ambiguous audio cues. Additionally, the research develops smart cropping techniques, automatic player detection, and an innovative multimodal game summarisation system aimed at enhancing sports content creation efficiency and engagement on digital platforms. Inderscience Publishers - linking academia, business and industry through research

Title: ST-LSTM-sports mind: a multimodal deep learning framework for intelligent sports analytics and automated journalism

Authors: Fangni Li; Yang He

Addresses: School of International Studies, Communication University of China, Beijing, China ' National Key Lab for Media Convergence and Communication, Communication University of China, Beijing, China

Abstract: This research introduces an AI-driven framework designed to automate the generation of sports highlights, optimising content creation for digital platforms. The framework utilises advanced deep learning techniques, including Spatial-Temporal Long Short-Term Memory (ST-LSTM) networks and convolutional neural networks (CNN), to address key challenges in sports classification, event detection, and player tracking. By incorporating multimodal data (audio and visual cues), the model achieves high accuracy rates, with 93% for goals, 90% for substitutions, and 86% for cards. However, further work is necessary to achieve 100% prediction accuracy for officially sanctioned events. The study also explores the integration of audio features to improve detection, particularly for dynamic events with strong audio cues, while acknowledging challenges with weak or ambiguous audio cues. Additionally, the research develops smart cropping techniques, automatic player detection, and an innovative multimodal game summarisation system aimed at enhancing sports content creation efficiency and engagement on digital platforms.

Keywords: artificial intelligence; sports highlight generation; ST-LSTM; audio-visual fusion; automated journalism; multimodal analytics.

DOI: 10.1504/IJCAT.2026.153788

International Journal of Computer Applications in Technology, 2026 Vol.78 No.7, pp.42 - 55

Received: 07 Jun 2025
Accepted: 28 Jan 2026
Published online: 26 May 2026 *

Title: ST-LSTM-sports mind: a multimodal deep learning framework for intelligent sports analytics and automated journalism

Keep up-to-date