Keyword-Based Ad Selection Using SVM Classifiers

The system performs preliminary action by pre-generating classifiers for multiple search keywords before actual advertisement selection is needed. These classifiers are trained in advance using positive and negative examples, allowing the system to quickly evaluate web page relevance without performing complex analysis at selection time. This pre-computation resolves the contradiction by preparing matching criteria beforehand, enabling both accurate relevance assessment and efficient scaling.

The patent introduces classifiers as intermediary components between web pages and advertisements. Instead of directly comparing web page content with advertisement content (which would be time-consuming), the system uses classifiers as mediators that have been trained to recognize relevant patterns. These classifiers quickly score web pages against search keywords, enabling scalable automated selection while maintaining relevance accuracy through their trained discrimination capabilities.

The system segments the advertisement selection process into distinct components: keyword identification, classifier evaluation, and advertisement matching. By breaking down the complex relevance determination into separate manageable tasks, each component can be optimized for both speed and accuracy. The classifiers are segmented by search keyword, allowing parallel processing and efficient evaluation of multiple criteria simultaneously, thus maintaining accuracy while improving throughput.

The system incorporates feedback mechanisms where classifiers are trained using positive examples (relevant web pages) and negative examples (irrelevant web pages). This feedback loop allows the automated system to learn from correct and incorrect classifications, continuously improving its relevance determination accuracy. The feedback-driven training ensures that automated selection maintains high precision by learning from actual relevance patterns rather than relying on simple rules.