Financial market movements are affected by not only previous prices and trade indicators but also by different types of information that is received from different sources such as social media, financial news, corporate/company announcement, analyst reports and economic events and so on. This gives a motivation to researchers for financial sentiment analysis to find signals from unstructured text materials and add this to the financial prediction models. The objective here is to understand financial behaviour after taking sentiments of a trader into account. Machine learning, deep learning and recently transformer based models are greatly improving the capability of financial sentiment detection and also helping to predict market movement effectively. If we look into the research literature from 2020 to 2026, we can see the growing trend of using machine learning and deep learning models for financial sentiment analysis. Most of the research papers in the previous period were based on traditional machine learning algorithms and traditional text classification approaches. But there is a growing trend in the recent research literature of using LSTM networks, transformer based models like FinBERT, ensemble approaches and large language models, which are capable of understanding context and specific peculiarities of financial language. Although these complex models are growing in complexity, it raises concerns like resource requirement, model transparency and interpretability of the predictions. From the literature, we can see that there is a common theme that suggests that information in textual sentiment may be able to capture some information that is not easily captured by numerical market variables. Sometimes, the variation of results could be attributed to difference in characteristics of the market, data sources, time period of the studies, prediction horizons, modelling approach and so on at that point in time. This is a strong indication that the characteristics of the data are playing a very strong role in the effectiveness of the sentiment analysis. From the existing literature it seems that it is more effective to use sentiment information along with traditional financial variables. As a standalone predictor it is not so effective. But there are some important issues that have not been explored enough for this field despite advancements. These are integration of multi source information, appropriate temporal alignment of heterogeneous data, avoiding information leakage, model robustness evaluation for varying market conditions, handling computational requirements and at the same time model interpretability. These problems open up a detailed discussion on combined financial prediction models. If we develop an integrated framework which can use financial text sentiment and market variables together, we may get a more robust approach to predict financial markets. This framework should take into account information timing and also model interpretability. Indian financial market can be a good setup to carry out this type of investigation. We can get multiple views of changes in market sentiment using company disclosures, news, investor talks, market data etc. which can be studied along with market behaviour. After which it can be collected and processed for a prediction model. We need to consider how different sources of information are combined in a thoughtful way. For example, market data is captured at regular time intervals whereas news articles, corporate announcements, social media posts are available at non-regular time points. Information timing of release of textual information is important for understanding the linkage with future market movements. If we mistakenly consider information which is not available at prediction time in the model then it may give wrong figures for model performance. The issue is not just technical. It is more important to have model transparency when these models are used for decision making. Also when it is being used as a base for research.