Currently, millions of companies, organizations and individuals take advantage of the social media function of Twitter to promote themselves. One of the most important goals is to attract web traffic. In this paper, we study the problem of obtaining web traffic via Twitter. We approach this problem in two stages. First, we analyze the correlation between important factors and the click number of URLs in tweets. Through measurements, we find that the commonly accepted method, increasing followers by reciprocal exchanges of links, has limited effects on improving the number of clicks. And characteristics of tweets (such as the presence of hashtags and tweet length) exert different impacts on users with different influence levels for obtaining the click number. In our second stage, based on the analyses, we introduce the Multi-Task Learning (MTL) to build a model for predicting the number of clicks. This model takes into account the specific characters of users with different influence levels to improve the predictive accuracy. The experiments, based on Twitter data, show the predictive performance is significantly higher than the baseline.