Pipeline for document scoring
One or more techniques and/or systems are provided for implementing a pipeline used to generate, train, test, and implement a document scoring model for assigning document scores to documents. Features from various sources are combined to create a joined page level feature set, a joined domain level feature set, and a host level feature set. Numerical features and content features are extracted from ground truth documents and random documents. The numerical features are joined with the joined feature sets to create a set of joined features. The document scoring model is trained using the set of joined features and a training technique. A document is scored with a document score using the document scoring model based upon the content features and the set of joined features with document scores obtained during training.
1. A method comprising:
combining feature types within one or more levels of feature sets into a joined host level feature set, wherein the one or more levels of feature sets comprise page level features joined into a joined page level feature set, domain level features joined into a joined domain level feature set, and host level features joined into a joined host level feature set, wherein a domain level feature corresponds to a feature of a domain associated with a target document, and wherein a host level feature corresponds to a feature of a host associated with a target document;
extracting numerical features and content features from ground truth documents and random documents, wherein a numerical feature corresponds to a ratio of an amount of a first type of content within a target document to an amount of a second type of content within the target document;
joining the numerical features with the one or more levels of feature sets to create a set of joined features for the ground truth documents and the random documents;
training a document scoring model utilizing machine learning to score documents using the set of joined features;
scoring documents with document scores using the document scoring model based upon the content features and the set of joined features with document scores obtained during training; and
selectively indexing a subset of the documents based upon the document scores of the documents.
2. The method of claim 1 , wherein a set of search results for a query comprises a document, and the method comprising:
assigning a rank to a document within the set of search results based upon a document score assigned to the document.
3. The method of claim 2 , comprising:
displaying the set of search results in response to receiving the query, wherein a document is populated within the set of search results based upon the rank.
4. The method of claim 1 , comprising:
indexing a document based upon a document score exceeding a threshold.
5. The method of claim 1 , comprising:
refraining from indexing a document based upon a document score not exceeding a threshold.
6. The method of claim 1 , wherein the document score is indicative of an importance of the document.
7. The method of claim 1 , wherein the document score is indicative of at least one of an importance or quality of the document.
8. The method of claim 1 , wherein the document score is indicative of a relevancy of the document.
9. The method of claim 1 , wherein a numerical feature corresponds to a numerical statistic of a target document.
10. The method of claim 1 , wherein a document comprises a webpage.
11. The method of claim 1 , wherein a document comprises a text document.
12. The method of claim 1 , wherein a numerical feature corresponds to a number of times a target document is linked to.
13. The method of claim 1 , wherein the document score is indicative of a quality of the document.
14. The method of claim 2 , comprising displaying the set of search results.
15. The method of claim 1 , wherein the machine learning comprises a gradient boosted decision tree regression technique.
16. The method of claim 1 , comprising:
merging the numerical features with the content features for scoring a document using the document scoring model.
17. The method of claim 1 , wherein the document score is indicative of an importance and a quality of the document.
18. The method of claim 1 , wherein the content features comprise textual features of a target document.
19. A non-transitory machine readable medium comprising instructions for performing a method, which when executed by a machine, causes the machine to:
combine feature types within one or more levels of feature sets into a joined host level feature set, wherein the one or more levels of feature sets comprise page level features joined into a joined page level feature set, domain level features joined into a joined domain level feature set, and host level features joined into a joined host level feature set, wherein a domain level feature corresponds to a feature of a domain associated with a target document, and wherein a host level feature corresponds to a feature of a host associated with a target document;
extract numerical features and content features from ground truth documents and random documents, wherein a numerical feature corresponds to a ratio of an amount of a first type of content within a target document to an amount of a second type of content within the target document;
join the numerical features with the one or more levels of feature sets to create a set of joined features for the ground truth documents and the random documents;
train a document scoring model utilizing machine learning to score documents using the set of joined features;
score documents with document scores using the document scoring model based upon the content features and the set of joined features with document scores obtained during training; and
selectively index a subset of the documents based upon the document scores of the documents.
20. A computing device comprising:
a memory comprising instructions; and
a processor coupled to the memory, the processor configured to execute the instructions to cause the processor to:
combine feature types within one or more levels of feature sets into a joined host level feature set, wherein the one or more levels of feature sets comprise page level features joined into a joined page level feature set, domain level features joined into a joined domain level feature set, and host level features joined into a joined host level feature set, wherein a domain level feature corresponds to a feature of a domain associated with a target document, and wherein a host level feature corresponds to a feature of a host associated with a target document;
extract numerical features and content features from ground truth documents and random documents, wherein a numerical feature corresponds to a ratio of an amount of a first type of content within a target document to an amount of a second type of content within the target document;
join the numerical features with the one or more levels of feature sets to create a set of joined features for the ground truth documents and the random documents;
train a document scoring model utilizing machine learning to score documents using the set of joined features;
score documents with document scores using the document scoring model based upon the content features and the set of joined features with document scores obtained during training; and
selectively index a subset of the documents based upon the document scores of the documents.