Journal of Web Engineering

Statistical Signal Amplification for Watermark Verification in Low-Rank Diffusion Adaptations

Jinseok Kim — 2026-08-22

As the World Wide Web evolves into the central infrastructure for AI-generated content (AIGC), ensuring the provenance of assets distributed via online platforms has become a critical challenge in Web Engineering. The uncontrolled propagation of Low-Rank Adaptation (LoRA) models facilitates unauthorized style mimicry, yet existing watermarks often fail to survive LoRA’s parameter compression. To safeguard digital trust and creator rights, we propose an Adaptation-Agnostic Trace Verification method optimized for secure web ecosystems. Our approach combines deep learning-based watermarking with a Statistical Resonance Amplifier (SRA) to induce the transfer of high-frequency signals into model weights. Furthermore, to overcome the noise limitations of single-image analysis in distributed web applications, we introduce an ensemble-based detection technique. Experimental results validate the method’s robustness, achieving an AUC-ROC of 0.891 even in highly restricted Rank 32 environments (using 100 generated images) and a near-perfect 0.999 at Rank 128, without degrading generation quality. This study presents a practical technology for AI governance and copyright protection, essential for ensuring the trustworthiness of AI-enhanced Web services.

A Robust Audio DNA-Based Method for OTT Content Recognition in Noisy Web Streaming Environments

Byeongchan Park — 2026-08-22

With the proliferation of Over-The-Top (OTT) platforms via web browsers and mobile web apps, the demand for real-time copyright protection in web streaming environments has surged. However, unstructured noise generated when users consume web content in public environments (e.g., subways, cafes) increases the false positive rate of existing clean-audio-based identification systems and heavily burdens web server computations. This paper proposes a robust, audio DNA-based content recognition method capable of fast and accurate retrieval from large-scale media databases even in noisy web streaming environments. The proposed method extracts dual-stage (Coarse-Fine) features based on Mel-spectrograms at the client side and performs a highly efficient three-stage matching pipeline (Coarse Matching, Fine Matching, and Post-verification) at the web server side. Experimental results on 649 noisy audio samples demonstrated that applying web-optimized parameters (FFT length of 4096, Hop length of 1470) achieved a precision of 0.9965, a recall of 0.8814, and an F1-score of 0.9354. The proposed method significantly accelerates retrieval speed for large-scale web streaming data through binary hash matching in the Coarse stage while maintaining high accuracy, proving its effectiveness for real-time OTT copyright protection and web media monitoring systems.

An Event-Based Platform-Independent Verification Method for OTT Content Usage History

Ruziev Ulugbek Shukhrat Ugli — 2026-08-22

In OTT (Over-The-Top) service environments, reliable recording and verification of content usage history are becoming increasingly important for copyright protection, royalty settlement, and dispute resolution. However, conventional platform-specific log management methods have limitations in interoperability, integrity assurance, and external verifiability. In addition, simple indicators such as the number of views or total playback time are insufficient to accurately reflect actual OTT viewing behavior, which often includes nonlinear interactions such as pause, seek, replay, and early stop.

This paper proposes a platform-independent event-based verification method for OTT content usage history. The proposed method defines usage history metadata consisting of content information, session information, playback event information, and usage result information, and derives valid usage intervals from playback events such as play, pause, seek, replay, and stop. In addition, hash-based verification information and linked recording structures are applied to the generated usage history to enhance integrity and traceability.

To examine the applicability of the proposed method, processing latency in the usage history aggregation stage was measured according to changes in the number of sessions and platforms. The experimental results show that the processing time increased in a stable manner as the session scale increased, indicating that the proposed method can be applied to large-scale OTT usage environments. This study is meaningful in that it presents a platform-independent and verifiable method for organizing OTT content usage history, and it can be further extended through future performance evaluation of blockchain recording and verification processes in diverse service environments.

A Web-Based Streaming Video Identification Method Using Self-Supervised Structural Embedding and Vector Similarity Search

Injae Yoo — 2026-08-22

Streaming video services have become a major channel for content distribution owing to the rapid growth of over-the-top (OTT) platforms and web-based media services. As a result, accurate video identification is increasingly required for copyright protection, audience measurement, content management, and search and recommendation services. However, conventional feature point-based methods often show limited robustness against transformations frequently observed in streaming environments, including compression, resolution change, re-encoding, frame-rate reduction, aspect-ratio conversion, overlays, rotation, and flipping. They also require the storage and comparison of many local descriptors, which can limit retrieval efficiency in large-scale content databases. In this paper, we propose a web-based streaming video identification method using self-supervised structural embedding and vector similarity search. The proposed method extracts representative frames from original and query videos, applies preprocessing, generates 512-dimensional structural embeddings using a self-supervised copy detection model, and performs cosine similarity search in a vector database. To improve robustness against geometric transformations, transformed reference frames are additionally registered. For query videos, multiple representative frames are used, and the final video-level identification result is determined by majority voting over frame-level retrieval results. Each frame is represented by a fixed-length 512-dimensional vector, requiring 2048 bytes per frame and 10,240 bytes for a five-frame query. Experiments on 4000 query videos covering 14 transformation categories and 40 detailed transformation settings show that the proposed method achieves a recognition rate of 99.23%, a missed recognition rate of 0.38%, and a false recognition rate of 0.40%. These results demonstrate the feasibility of the proposed method for identifying transformed copies of the same source video in web-based streaming environments.

Unsupervised Cross-Modal Hashing Algorithms for Web Multimedia Retrieval

Yang-hao Li — 2026-08-22

With the Web witnessing a rapid surge in multimodal data, it’s becoming increasingly vital to develop efficient and budget-friendly cross-modal (CM) retrieval techniques to elevate the user experience in Web applications. Traditional hashing methods, however, often neglect the valuable semantic information hidden within the text descriptions that come with Web images. Moreover, they tend to lean heavily on supervised learning, which poses a challenge when it comes to adapting to real-world Web scenarios where annotations are often in short supply. To tackle this, this research introduces an unsupervised CM hashing algorithm for Web multimedia retrieval. By mining the semantic structure of text associated with Web images and utilizing a deep network to achieve semantic transfer from text to vision, a unified and efficient hashing learning framework is constructed. Experiments indicate that the introduced approach achieves mAP values of 0.3370 and 0.6990 with 16-bit hash codes (HCs). When the HC length is increased to 128 bits, the mAP increases to 0.3632 and 0.7575, representing an absolute improvement of 13.0% and 4.04% compared to the best performing baseline method. Further analysis shows that the semantic transfer mechanism significantly improves the semantic representation ability of the HCs. Even in a semi-supervised setting using only 20% of labeled data, the retrieval mAP can still reach 0.8871. The method requires only a portion of image-text pairs during the training phase and supports pure image queries during the retrieval phase, achieving millisecond-level response times under Hamming distance calculation. This provides an efficient and practical solution for Web-scale multimedia retrieval.

Immersive Web Virtual Environment for Language Learning: Multimodal Interaction and Error Correction Based on Speech, Gesture and Eye Tracking

Shuwen Yu — 2026-08-22

In order to solve the problems of existing Web language learning tools, such as single interaction dimension, lagging error correction feedback and insufficient attention perception, this paper designs and verifies a multimodal immersive Web virtual environment optimization scheme which integrates voice, gesture and eye tracking. The core breakthroughs are as follows: (1) a dual-mechanism adaptive fusion algorithm of “attention weight+dynamic threshold” is proposed, which solves the problem of error correction failure of the traditional fixed threshold scheme in high-fluctuation interactive scenarios (such as voice discontinuity, gesture occlusion, attention drift); (2) a four-stage closed-loop control mechanism of perception-decision-feedback-adjustment” is constructed, which realizes the end-side lightweight deployment. Based on WebXR, MediaPipe and other native technologies, the cross-end compatible architecture is constructed, and the comparative experiments on three data sets and 10 typical working conditions show that, compared with the traditional Web platform, the collaborative error correction accuracy of the scheme is 93. 2% (improved by 41%), and the response time is compressed to 0.15 seconds (shortened by 62%). End-side resource occupancy is reduced to 16.8%, and learning efficiency is 4.8 times per minute. The research confirms that the scheme effectively balances the immersion, error correction accuracy and end-to-end adaptability, and provides technical support for the large-scale application of immersive Web language education.

Adaptive Learning Driven by Web Technologies: A Model-Driven Development Framework for AI Enhanced English Intelligent Tutoring Web Applications

Yuqing Ge — 2026-08-22

There are some problems in the existing English teaching network applications, such as low personalized adaptability, outdated teaching strategies and weak multimodal interaction, limited teaching paths, single feedback mechanism, weak adaptability of learners, low learning efficiency and low retention rate. In order to solve these problems, this paper proposes a Model-Driven Development (MDD) framework that integrates Web technology and artificial intelligence (Web-AI-I), and builds a lightweight, cross-device English intelligent teaching application through the Web native technology stack. Multi-modal data perception, AI adaptive decision-making and real-time feedback optimization are integrated to solve the coupling problem between teaching logic and Web application architecture through MDD mode. Through the experiment, the response time of learning path adaptation can be effectively shortened to 0.2 s, the English vocabulary mastery rate is 52% higher than that of traditional Web teaching applications, the grammar error correction accuracy rate is 91.8%, the average resource occupancy rate in cross-device scenarios is only 18.3%, and the MTA is 412, which is helpful to realize the large-scale landing of English intelligent teaching Web applications.