Method for selecting electronic advertisements using machine translation techniques
A system for selecting electronic advertisements from an advertisement pool to match the surrounding content is disclosed. To select advertisements, the system takes an approach to content match that takes advantage of machine translation technologies. The system of the present invention implements this goal by means of simple and efficient machine translation features that are extracted from the surrounding context to match with the pool of potential advertisements. Machine translation features used as features for training a machine learning model. In one embodiment, a ranking SVM (Support Vector Machines) trained to identify advertisements relevant to a particular context. The trained machine learning model can then be used to rank advertisements for a particular context by supplying the machine learning model with the machine translation features measures for the advertisements and the surrounding context.
1. A method of determining relevance of an electronic advertisement to a target content, said method comprising:
extracting a set of terms from said electronic advertisement and said target content;
calculating a first content match feature using said set of terms, the first content match feature comprising a translation evaluation feature indicating a degree to which n-grams of the electronic advertisement and the target content match;
calculating a second content match feature using said set of terms, the second content match feature comprising a translation probability feature indicating a probability that one or more terms of the electronic advertisement are related to one or more terms of the target content, wherein the one or more terms of the electronic advertisement do not match the one or more terms of the target content; and
processing said first content match feature and said second content match feature with a machine learning model to output a relevance score indicating the relevance of the electronic advertisement to the target content, the machine learning model comprising individual weights for the first and second content match features, the machine learning model being trained using said first and second content match features and machine learning techniques.
2. The method of claim 1 , wherein:
the first content match feature comprises a BLEU metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
3. The method of claim 1 , wherein:
the first content match feature comprises a NIST metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
4. The method of claim 1 , wherein said machine learning model comprises a support vector machine.
5. The method of claim 1 , wherein said electronic advertisement and target content comprise different vocabularies, said first and second content match features providing a translation between said electronic advertisement and target content vocabularies.
6. A system, comprising at least one processor and memory, for determining relevance of an electronic advertisement to a target content, said system comprising:
a server system configured for:
extracting a set of terms from said electronic advertisement and said target content;
calculating a first content match feature using said set of terms, the first content match feature comprising a translation evaluation feature indicating a degree to which n-grams of the electronic advertisement and the target content match;
calculating a second content match feature using said set of terms, the second content match feature comprising a translation probability feature indicating a probability that one or more terms of the electronic advertisement are related to one or more terms of the target content, wherein the one or more terms of the electronic advertisement do not match the one or more terms of the target content; and
processing said first content match feature and said second content match feature with a machine learning model to output a relevance score indicating the relevance of the electronic advertisement to the target content, the machine learning model comprising individual weights for the first and second content match features, the machine learning model being trained using said first and second content match features and machine learning techniques.
7. The system of claim 6 , wherein:
the first content match feature comprises a BLEU metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
8. The system of claim 6 , wherein:
the first content match feature comprises a NIST metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
9. The system of claim 6 , wherein said machine learning model comprises a support vector machine.
10. The system of claim 6 , wherein said electronic advertisement and target content comprise different vocabularies, said first and second content match features providing a translation between said electronic advertisement and target content vocabularies.
11. A method of determining relevance of an electronic advertisement to a target content, said method comprising:
extracting a set of terms from said electronic advertisement and said target content;
calculating a first content match feature using said set of terms, the first content match feature comprising a translation evaluation feature indicating a degree to which n-grams of the electronic advertisement and the target content match;
calculating a second content match feature using said set of terms, said second content match feature comprising a translation proportion feature indicating a proportion of related terms in the electronic advertisement and the target content; and
processing said first content match feature and said second content match feature with a machine learning model to output a relevance score indicating the relevance of the electronic advertisement to the target content, the machine learning model comprising individual weights for the first and second content match features, the machine learning model being trained using said first and second content match features and machine learning techniques.
12. The method of claim 11 , wherein:
the first content match feature comprises a BLEU metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
13. The method of claim 11 , wherein:
the first content match feature comprises a NIST metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
14. The method of claim 11 , wherein said machine learning model comprises a support vector machine.
15. The method of claim 11 , wherein said electronic advertisement and target content comprise different vocabularies, said first and second content match features providing a translation between said electronic advertisement and target content vocabularies.
16. A system, comprising at least one processor and memory, for determining relevance of an electronic advertisement to a target content, said system comprising:
a server system configured for:
extracting a set of terms from said electronic advertisement and said target content;
calculating a first content match feature using said set of terms, the first content match feature comprising a translation evaluation feature indicating a degree to which n-grams of the electronic advertisement and the target content match;
calculating a second content match feature using said set of terms, said second content match feature comprising a translation proportion feature indicating a proportion of related terms in the electronic advertisement and the target content; and
processing said first content match feature and said second content match feature with a machine learning model to output a relevance score indicating the relevance of the electronic advertisement to the target content, the machine learning model comprising individual weights for the first and second content match features, the machine learning model being trained using said first and second content match features and machine learning techniques.
17. The system of claim 16 , wherein:
the first content match feature comprises a BLEU metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
18. The system of claim 16 , wherein:
the first content match feature comprises a NIST metric; and
an n-gram comprises a bigram, trigram, or 4-gram.
19. The system of claim 16 , wherein said machine learning model comprises a support vector machine.
20. The system of claim 16 , wherein said electronic advertisement and target content comprise different vocabularies, said first and second content match features providing a translation between said electronic advertisement and target content vocabularies.