Cryptocurrency Investment Recommendation System for Users Based on Market Trend and Popularity Analysis

Abstract

The dynamic cryptocurrency market with thousands of digital assets poses significant challenges for investors in identifying investment opportunities that suit their preferences and risk profile. Conventional recommendation systems have not been optimal because they fail to accommodate the unique characteristics of digital assets such as high volatility, popularity metrics (market cap, trading volume), and current investment trends. The research developed a cryptocurrency recommendation system based on Feature-Enhanced Collaborative Filtering (FECF) and Neural Collaborative Filtering (NCF), as well as a Hybrid model that combines the two to improve the accuracy and adaptability of recommendations. Data obtained through CoinGecko's API (top 1,000 projects) includes market, social, and asset category metrics. The system development follows the Agile-Scrum and CRISP-DM methodologies, with PyTorch (NCF model) and FastAPI-Laravel (API/web application) technologies. Evaluation using the Precision, Recall, NDCG, and Hit Ratio metrics showed that the Hybrid model excelled in providing relevant personalized recommendations (NDCG@10: 0.3557), while FECF was more effective in handling cold-start problems (Hit Ratio: 63.73%) and data sparsity (98.77%). This system provides practical contributions for investors in decision-making and methodological contributions to the development of a blockchain-based digital asset recommendation system