Predicting customer purchase behaviour is an interesting and challenging task. In e-commerce context, to tackle the challenge will confront a lot of new problems different from those in traditional business. This study investigates three factors that affect purchasing decision-making of customers in online shopping: the needs of customers, the popularity of products and the preference of the customers. Furthermore, exploiting purchase data and ratings of products in the e-commerce website, we propose methods to quantify the strength of these factors: (1) using associations between products to predict the needs of customers; (2) combining collaborative filtering and a hierarchical Bayesian discrete choice model to learn preference of customers; (3) building a support vector regression based model, called Heat model, to calculate the popularity of products; (4) developing a crowdsourcing approach based experimental platform to generate train set for learning Heat model. Combining these factors, a model, called COREL, is proposed to make purchase behaviour prediction for customers. Submitted a purchased product of a customer, the model can return top n the most possible purchased products of the customer in future. Experiments show that these factors play key roles in predictive model and COREL can greatly outperform the baseline methods.