Cryptocurrencies have become a vital asset class with rising investment across demographics. However, their high risk is a well-known stylized fact. Therefore, understanding the interconnected movements between cryptocurrencies and their connections with other asset classes is crucial for reducing portfolio risk, especially during market crashes. Building on this context, we first examine how cryptocurrency correlations change over time, especially between stable and turbulent periods. Next, we explore how investor sentiments (e.g. fear, neutral and greed) and investment behaviors (e.g. herding and contrarian) influence these correlations. Lastly, we extend the study to traditional assets, including stocks and US ETFs, to compare correlations and investment behaviors between cryptocurrency and traditional markets across various market conditions. Using graph-based methods such as Minimum Spanning Tree, Louvain and Girvan-Newman algorithms, we discover consistent changes in cryptocurrency correlations over time. However, within turbulent and stable periods, the correlations exhibit distinct patterns. Specifically, they increase during turbulent times like market crashes, global economic crises, political turmoil and bullish markets. Conversely, they remain low during stable times. Understanding these patterns helps investors make strategic investment decisions and optimize their portfolios. Furthermore, by adopting a popular herding detection model Cross-sectional Absolute Deviation, we find that the correlation in the cryptocurrency market is strongly influenced by investor sentiments and investment behaviors. In other words, the reactions of investors towards the market signal its upcoming correlation pattern as well as possible movements. Therefore, we suggest that investor sentiment is a good indicator that helps investors anticipate future changes and movements in the market, preparing them to adjust their portfolios if necessary. Notably, we discover that although the cryptocurrency market is gradually becoming similar to traditional markets, there remains a significant gap, not only in the correlation between them but also in statistical characteristics and in how investors engage with them. The distinction of cryptocurrencies from traditional markets suggests a great potential for them to be used as hedges in a portfolio. In addition to new findings, we introduce two novel techniques. First, a noise and trend removal scheme is applied to the assets’ correlations using the Random Matrix Theory and Market Component concept, which has never been considered in the quantitative finance literature. Secondly, this is the first study observing herding behavior at a community level, using the Louvain algorithm.