Yahoo! Inc.

Ramnath Balasubramanyan

Byron Edward Dom

Dmitry Pavlov

Bipin Suresh

Abhishek Goyal

Dmitri Y Pavlov

Hui Han

Techniques are described for assigning, to target categories of a target scheme, items that have been obtained from a plurality of sources. In situations in which one or more of the sources has organized its information according to a source scheme that differs from the target scheme, the assignment may be based, in part, on an estimate of the probability that items from a particular source category should be assigned to a particular target category. Such probability estimates may be based on how many training set items associated with the particular source category have been assigned to the particular target category. Source categories may be grouped into clusters. The probability estimates may also be based on how many training set items within the cluster to which the particular source category has been mapped, have been assigned the particular target category.

Assigning into one set of categories information that has been assigned to other sets of categories

Techniques are provided for automatic product categorization. In one aspect, the categorization is based on text and one or more other values associated with a product offering. In another aspect, a first categorization of a product offering is performed and, if the product category chosen is in a set of co-refinable product categories, then a second (or third, fourth and so on) categorization is performed among the set of co-refinable product categories. In a third aspect, products are categorized based on cost. In another aspect, after products are categorized, the products are flagged for further categorization processing if the cost for categorizing the product is beyond a predefined threshold.

Automatic product categorization

Techniques are described for reducing the false positive rate of regular expression attribute extractions via a specific data representation and a machine learning method that can be trained at a much lower cost (much fewer labeled examples) than would be required by a full scale machine learning solution. Attribute determinations made using the regular expression technique are represented as skeleton tokens. The skeleton tokens, along with accurate attribute determinations, are provided to a machine-learning mechanism to train the machine-learning mechanism. Once trained, the machine-learning mechanism is used to predict the accuracy of attribute determinations represented by skeleton tokens generated for not-yet-analyzed input text.

Growing community of inventors

Mountain View, CA, United States of America

Ramnath Balasubramanyan