Paid Search, also known as Search Engine Marketing (SEM), has been the largest channel in online advertising by revenue. Yet its causal effectiveness has been difficult to measure. At eBay marketplace we have developed a hybrid Geo+User experiment approach, and conducted a long-running field test over 17-month period to understand the incrementality of paid search on one of the largest e-commerce platforms in the U.S. The experiment results indicate that paid search campaigns drive statistically significant sales lift to the eBay site, and a portion of the direct sales made via paid search is incremental. More importantly, such incrementality demonstrates strong seasonality during the year, and do not necessarily correlate with the paid search spend levels. Our results also show that the new user acquisition lift is higher than the immediate sales lift on the eBay site from paid search campaigns. These findings provide critical insights for paid search and overall online advertising investment strategies.

A new paradigm for designing compressed image quality metric is proposed in this paper, of which the most significant characteristic is the use of mutual information, a key concept in information theory which measures statistical dependence between two random variables, to exploring the degree of similarity of spatial visual information distribution across different frequency bands in image. Visual information used in the calculation of mutual information is extracted by contrast sensitivity function (CSF) and local band-limited contrast definition proposed by E. Peli. Our paradigm is more consistent with human perceptual mechanism comparing with the traditional error-summation based ones. The effectiveness of our image quality assessment paradigm is validated by JPEG & JPEG2000 compressed images at different bit rates and images with various types of noises.

This paper studies the transformation mechanisms of multilayer perceptrons with sigmoid activation functions for classifications. The viewpoint is presented that in the input spaces the hyperplanes determined by the hidden basis functions with values of 0 do not play the role of separate hyperplanes, and furthermore such "hyperplanes" do not certainly go through the marginal regions between different classes. The number of hidden units is only related to the number of categories and the sample distribution shapes. The rank of output matrix of hidden units should be taken as the basis for pruning or growing the hidden nodes. As a result, an empirical formula for optimally determining the number of hidden neurons is proposed. Finally, two examples are given to verify it.