Stock price prediction has always been an important application in time series predictions. Recently, deep neural networks have been employed extensively for financial time series tasks. The network typically requires a large amount of training samples to achieve high accuracy. However, in the stock market, the number of data points collected on a daily basis is limited in one year, which leads to insufficient training samples and accordingly results in an overfitting problem. Moreover, predicting stock price movement is affected by various factors in the stock market. Therefore, choosing appropriate input features for prediction models should be taken into account. To address these problems, this paper proposes a novel framework, named deep transfer with related stock information (DTRSI), which takes advantage of a deep neural network and transfer learning. First, a base model using long short-term memory (LSTM) cells is pre-trained based on a large amount of data, which are obtained from a number of different stocks, to optimize initial training parameters. Second, the base model is fine-tuned by using a small amount data from a target stock and different types of input features (constructed based on the relationship between stocks) in order to enhance performance. Experiments are conducted with data from top-five companies in the Korean market and the United States (US) market from 2012 to 2018 in terms of the highest market capitalization. Experimental results demonstrate the effectiveness of transfer learning and using stock relationship information in helping to improve model performance, and the proposed approach shows remarkable performance (compared to other baselines) in terms of prediction accuracy.

Stock price forecasting remains a persistent challenge in time series analysis due to complex inter-stock relationships and dynamic textual signals such as financial news. While Graph Neural Networks (GNNs) can model relational structures, they often struggle with capturing higher-order dependencies and are sensitive to noise. Moreover, sentiment signals are typically aggregated using fixed time windows, which may introduce temporal bias. To address these issues, we propose DASF-Net (Diffusion-Aware Sentiment Fusion Network), a multimodal framework that integrates structural and textual information for robust prediction. DASF-Net leverages diffusion processes over two complementary financial graphs—one based on industry relationships, the other on fundamental indicators—to learn richer stock representations. Simultaneously, sentiment embeddings extracted from financial news using FinBERT are aggregated over an empirically optimized window to preserve temporal relevance. These modalities are fused via a multi-head attention mechanism and passed to a temporal forecasting module. DASF-Net integrates daily stock prices and news sentiment, using a 3-day sentiment aggregation window, to forecast stock prices over daily horizons (1–3 days). Experiments on 12 large-cap S&P 500 stocks over four years demonstrate that DASF-Net outperforms competitive baselines, achieving up to 91.6% relative reduction in Mean Squared Error (MSE). Results highlight the effectiveness of combining graph diffusion and sentiment-aware features for improved financial forecasting.