Method, apparatus and computer readable medium for indexing advertisements to combine relevance with consumer click feedback
A method and apparatus are provided for better web ad matching by combining relevance with consumer click feedback. In one example, the method includes receiving a query page, extracting features from the query page, re-weighting the query page, evaluating the query page in light of each ad in order to score each ad and pick substantially best ad matches of the indexed ads, and returning the substantially best ad matches to the consumer computer.
1. A system for indexing an advertisement in order to match the advertisement to a webpage, the system comprising:
a communications interface configured to receive click-through rate (CTR) data for an advertisement;
a processor and memory communicatively coupled to the communications interface configured to:
extract an advertisement feature from the advertisement and a webpage feature from the webpage that are relevant to each other, via a variable selection technique that includes at least one of: selection of the advertisement feature and the webpage feature using CTR data and views of the advertisement, or selection of the advertisement feature and the webpage feature using relevance scores of words of the advertisement that are independent of click feedback;
calculate a coefficient representative of the relevancy between the advertisement feature and the webpage feature;
compute a static score for the advertisement by using the CTR data and the coefficient as input to a logistic regression function, wherein the static score provides relevance of the advertisement feature with respect to the webpage feature, and wherein the logistic regression function uses a monotonically increasing linear function;
generate a unique identification for the advertisement feature; and
generate a unique advertisement identification for the advertisement according to the static score relative to other advertisements.
2. The system of claim 1 , wherein the monotonically increasing linear function is optimized by iterative scaling function.
3. The system of claim 1 , wherein the monotonically increasing linear function is optimized by a nonlinear conjugate gradient function.
4. The system of claim 1 , wherein the monotonically increasing linear function is optimized by a quasi-Newton function.
5. The system of claim 1 , wherein the monotonically increasing linear function is optimized by an iteratively-reweighted least squares function.
6. The system of claim 1 , wherein the monotonically increasing linear function is optimized by a truncated Newton function or trust-region Newton function.
7. A method for indexing advertisements in order to match an advertisement to a webpage, the method comprising:
receiving click-through rate (CTR) data for an advertisement;
extracting an advertisement feature from the advertisement and a webpage feature from the webpage that are relevant to each other, via a variable selection technique that includes at least one of: selection of the advertisement feature and the webpage feature using CTR data and views of the advertisement, or selection of the advertisement feature and the webpage feature using relevance scores of words of the advertisement that are independent of click feedback;
calculating a coefficient representative of the relevancy between the advertisement feature and the webpage feature;
computing a static score for the advertisement by using the CTR data and the coefficient as input to a logistic regression function, wherein the static score provides relevance of the advertisement feature with respect to the webpage feature, and wherein the logistic regression function uses a monotonically increasing linear function;
generating a unique identification for the advertisement feature; and
generating a unique advertisement identification for the advertisement according to the static score relative to other advertisements.
8. The method of claim 7 , wherein the advertisement feature or the webpage feature includes a unigram.
9. The method of claim 7 , wherein the advertisement feature or the webpage feature includes a phrase.
10. The method of claim 7 , wherein the linear function is a monotonically increasing linear function, including one or more of an iterative scaling, a nonlinear conjugate gradient, a quasi-Newton, an iteratively-reweighted least squares, a truncated Newton, or a trust-region Newton function.
11. A non-transitory computer readable medium that stores a set of instructions executable by a computer to:
receive click-through rate (CTR) data for an advertisement;
extract an advertisement feature from the advertisement and a webpage feature from the webpage that are relevant to each other, via a variable selection technique that includes at least one of: selection of the advertisement feature and the webpage feature using CTR data and views of the advertisement, or selection of the advertisement feature and the webpage feature using relevance scores of words of the advertisement that are independent of click feedback;
calculate a coefficient representative of the relevancy between the advertisement feature and the webpage feature;
compute a static score for the advertisement by using the CTR data and the coefficient as input to a logistic regression function, wherein the static score provides relevance of the advertisement feature with respect to the webpage feature, and wherein the logistic regression function uses a monotonically increasing linear function including at least one of an iterative scaling, a nonlinear conjugate gradient, a quasi-Newton, an iteratively-reweighted least squares, a truncated Newton, or a trust-region Newton function;
generate a unique identification for the advertisement feature; and
generate a unique advertisement identification for the advertisement according to the static score relative to other advertisements.