A mass of traces of human activities show diverse dynamic patterns. In this paper, we comprehensively investigate the dynamic pattern of human attention defined by the quantity of interests on subdisciplines in an online academic communication forum. Both the expansion and exploration of human attention have a power-law scaling relation with browsing actions, of which the exponent is close to that in one-dimension random walk. Furthermore, the memory effect of human attention is characterized by the power-law distributions of both the return interval time and return interval steps, which is reinforced by studying the attention shift that monotonically increase with the interval order between pairs of continuously segmental sequences of expansion. At last, the observing dynamic pattern of human attention in the browsing process is analytically described by a dynamic model whose generic mechanism is analogy to that of human spatial mobility. Thus, our work not only enlarges the research scope of human dynamics, but also provides an insight to understand the relationship between the interest transitivity in online activities and human spatial mobility in real world.