
Abstract: Stock market price prediction entails forecasting future stock prices and trends based on historical data and various market indicators. Financial news and user generated text are also essential for investors to make informed investment decisions. Stock market price prediction presents a complex and challenging task due to the continuously evolving, vast, and generated data, companies’ performance, global events, and sensitivity to human emotions. Deep learning approaches have demonstrated promising results in various fields, including financial forecasting. This paper reviews the application of deep learning models for stock price prediction, surveying 58 studies published between 2009 and 2025, analyzing their capacity to discern intricate patterns and dependencies in financial time series data. Different deep learning architectures, and hybrid models thereof, their strengths, limitations, and suitability for this task are presented. Furthermore, the various factors influencing stock prices, the challenges associated with accurate prediction, and potential future research directions, as well as practical recommendations for stock investors, are presented.
Trust systems in decentralized blockchain networks commonly treat reputation as permanent, creating a vulnerability in which a user can accumulate trust through honest behavior, go dormant, and later return to exploit their stored reputation. This paper proposes a temporal trust-adaptive differential privacy mechanism that ties the privacy noise budget to both a user's verified trust score and their elapsed inactivity time. The noise budget decays exponentially with dormancy, ensuring that long-absent users receive maximum noise on re-entry regardless of historical reputation. We formally prove that the mechanism satisfies (ε, δ)-differential privacy for all users at all times, and that it suppresses both new-identity and dormant-identity Sybil attacks as a direct mathematical consequence, without requiring explicit adversary detection. On-chain enforcement is achieved through a Groth16 zk-SNARK proof that each update must carry, attesting to correct noise application and interaction freshness. Experiments on a real-world peer-to-peer trust dataset confirm negligible accuracy loss at the recommended operating point, empirical validation of the suppression guarantee on naturally dormant users, and minimal on-chain overhead introduced by the freshness constraint.
E-commerce has become widely used in recent years. Online E-commerce platforms use online text reviews and rates of products for marketing these products. New online buyers and customers mainly consider these online reviews and rates before their online purchases. Unfortunately, many of these online reviews are fake and computer / AI-generated reviews. Fake reviews give wrong description of product's quality. So, fake text reviews filtering is a major task in E-commerce. In this paper, fake reviews filtering system is proposed based on deep learning and Natural Language Processing (NLP). The proposed system used transformer-based model that used deep learning for learning labelled filtered fake reviews datasets. Also, non-labelled products' reviews were handled using the proposed system using pre-trained fine-tuned transformer model based on NLP. The pre-trained transformer applied self-learning using many large fake reviews datasets (domain-specific) and acquired its own domain (fake reviews) knowledge. The new non-filtered / non-labelled data were labelled using the fine-tuned model based on the acquired fake reviews knowledge. A filtered publicly available fake reviews dataset was used. The proposed system achieved accuracy of 92.14%.
Drug repurposing offers a faster, lower-cost alternative to de novo drug development, particularly for chronic and neurodegenerative diseases where conventional discovery pipelines are slow and resource-intensive. This paper proposes OTC-DTI (Optimized Transformer-CNN Drug-Target Interaction), a multi-stage framework that combines a Transformer-CNN encoder-decoder for drug-target interaction (DTI) prediction with Bayesian hyperparameter optimization and transfer learning, and applies the resulting model to repurpose existing drugs for Parkinson's disease, a neurodegenerative disorder with substantial unmet therapeutic need. On the Davis and KIBA benchmarks, OTC-DTI achieves a concordance index (CI) of 0.882 and mean squared error (MSE) of 0.255 on Davis, and a CI of 0.885 and MSE of 0.174 on KIBA, improving on the DeepDTA baseline by approximately 2.3% (MSE) and 0.5% (CI) on Davis and 10.3% (MSE) and 2.6% (CI) on KIBA, while remaining competitive with more recent graph- and attention-based architectures. Applying the optimized model to known Parkinson's disease signature genes via the Broad Repurposing Hub, OTC-DTI identifies several candidate compounds, including rifamycin and lycopene, with literature-supported plausibility for further investigation. These results demonstrate that combining hybrid deep learning architectures with Bayesian optimization and transfer learning is a practical and effective strategy for DTI-driven drug repurposing.
Recent advances in Large Language Models (LLMs) have significantly improved the quality of abstractive text summarization. However, the high computational cost and memory requirements of these models limit their practical adoption, particularly in resource-constrained environments. This focused survey reviews recent progress in efficient LLM-based abstractive text summarization, emphasizing methods designed to operate under realistic hardware constraints. The survey first provides a concise background on abstractive and extreme summarization, highlighting challenges introduced by long contexts and aggressive compression. A structured taxonomy is then proposed to categorize existing approaches according to model architecture, training strategy, adaptation technique, and hardware awareness. Building on this taxonomy, the paper analyzes key efficiency-oriented techniques, including parameter-efficient fine-tuning, low-bit quantization, precision optimization, and inference-time acceleration. Furthermore, commonly used evaluation protocols are reviewed, emphasizing the limitations of traditional automated metrics and the growing need for faithfulness- and cost-aware evaluation. A comparative discussion summarizes trade-offs between summarization quality and computational efficiency, offering practical guidelines for model selection and deployment. Finally, open challenges and future research directions are outlined, including faithful extreme summarization, long-context processing, multimodal summarization, and standardized efficiency-aware benchmarks. Overall, this survey provides researchers and practitioners with a clear, structured overview of practical considerations for efficient LLM-based text summarization.