Social network data is the basis of social network analysis that is why it's important to collect such data. To solve the problem of less collected data and low recall rate in current focused crawlers on social network, this paper proposes a method combining the based built-in search engine and general search engines to crawl topic messages, as well as applys the LDA model to extract the topic keywords from collected data iteratively and adds new topic keywords to the seed. Besides, an efficient expansion strategy based on users survival value is discussed. Our experiment shows that the methods proposed can improve the recall rate with a high precision.