Classifying search query traffic
A method for classifying search query traffic can involve receiving a plurality of labeled sample search query traffic and generating a feature set partitioned into human physical limit features and query stream behavioral features. A model can be generated using the plurality of labeled sample search query traffic and the feature set. Search query traffic can be received and the model can be utilized to classify the received search query traffic as generated by a human or automatically generated.
1. A computer-implemented method for classifying search query traffic, said method comprising:
receiving, from a search engine, labeled sample search query traffic, said labeled sample search traffic being labeled as human generated search query traffic or automatically generated search query traffic, said labeled sample search query traffic including one or more keywords for each search query submitted to said search engine and request times for a plurality of search queries submitted to said search engine within distinct user sessions;
extracting features from said labeled sample search query traffic in accordance with a set of feature definitions partitioned into physical features related to physical limitations of human generated search queries and behavioral features of automatically generated search queries and keywords of the automatically generated search queries;
generating a feature set comprising (i) said features extracted from said labeled sample search query traffic and (ii) a behavioral feature related to query word length entropy (WLE) that is calculated as:
WL
E
(
l
ij
)
=
-
∑
i
∑
j
l
ij
log
(
l
ij
)
,
i being an index for each separate query submitted to a search engine by a single user ID and I ij being a length of an individual query term j in the ith query;
generating a model using said labeled sample search query traffic and said feature set;
receiving, from said search engine, search query traffic associated with a plurality of search queries submitted by a particular user identifier;
classifying, using said model, said search query traffic associated with said plurality of search queries submitted by said particular user identifier as human generated search query traffic or automatically generated search query traffic; and
modifying a quality of service provided by said search engine to said particular user identifier when said search query traffic associated with said plurality of search queries submitted by said particular user identifier is classified as automatically generated search query traffic.
2. The computer-implemented method of claim 1 , further comprising:
classifying, using said model, said received search query traffic as legitimate human generated search query traffic or illegitimate human generated search query traffic.
3. The computer-implemented method of claim 1 , wherein said feature set comprises a query window feature that separates a distinct user session into a plurality of discrete time periods.
4. The computer-implemented method of claim 1 , further comprising:
servicing one or more search queries submitted by human users by modifying said quality of service provided by said search engine to said particular user identifier by delaying service of one or more search queries submitted by said particular user identifier.
5. The computer-implemented method of claim 1 , wherein said feature set comprises a behavioral feature related to chronological search queries submitted by a single user identifier in an alphabetical order.
6. The computer-implemented method of claim 1 , wherein said feature set comprises a behavioral feature related to keyword content associated with spam.
7. The computer-implemented method of claim 1 , wherein said feature set comprises a behavioral feature related to keyword content associated with adult content.
8. A system for classifying search query traffic, said system comprising:
memory storing computer-executable modules including:
a feature set module that receives, from a search engine, labeled sample search query traffic and search query traffic associated with a plurality of search queries submitted by a particular user identifier, said labeled sample search traffic being labeled as human generated search query traffic or automatically generated search query traffic, said labeled sample search query traffic including one or more keywords for each search query submitted to said search engine and request times for said plurality of search queries submitted to said search engine within distinct user sessions, said feature set module extracting features from said labeled sample search query traffic in accordance with a set of feature definitions partitioned into physical features related to physical limitations of human generated search queries and behavioral features of automatically generated search queries and keywords of the automatically generated search queries, said feature set module generating a feature set comprising said features extracted from said labeled sample search query traffic;
a classifier module that builds a model using said labeled sample search query traffic and said feature set, said classifier module using the model to classify said search query traffic associated with a said plurality of search queries submitted by said particular user identifier as human generated search query traffic or automatically generated search query traffic; and
a quality of service module that changes a quality of service provided by said search engine to said particular user identifier when said search query traffic associated with said plurality of search queries submitted by said particular user identifier is classified as automatically generated search query traffic; and
a processor that executes said computer-executable modules stored in said memory,
wherein said feature set comprises a behavioral feature related to query word length entropy (WLE) that is calculated as:
WL
E
(
l
ij
)
=
-
∑
i
∑
j
l
ij
log
(
l
ij
)
,
and
wherein i is an index for each separate query submitted to a search engine by a single user ID and I ij a length of an individual query term j in the ith query.
9. The system of claim 8 , wherein said classifier module uses said model to classify said search query traffic associated with said plurality of search queries submitted by said particular user identifier as legitimate human generated search query traffic or illegitimate human generated search query traffic.
10. The system of claim 8 , wherein said feature set comprises a behavioral feature related to number of open browser identifiers.
11. The system of claim 8 , wherein said feature set comprises a behavioral feature related to request time periodicity of automatically generated search queries.
12. The system of claim 8 , wherein said feature set comprises a behavioral feature related to entropy of categories associated with single user identifier.
13. The system of claim 8 , wherein said feature set comprises a behavioral feature related to one or more of blacklisted IP addresses, blacklisted user agents and particular country codes.
14. The system of claim 8 , wherein said feature set comprises a query window feature that separates a distinct user session into a plurality of discrete time periods.
15. A computer-readable storage medium storing computer-executable instructions that, when executed, cause a computer system to perform a method for classifying search query traffic, said method comprising:
receiving, from a search engine, labeled sample search query traffic, said labeled sample search traffic being labeled as human generated search query traffic or automatically generated search query traffic, said labeled sample search query traffic including one or more keywords for each search query submitted to said search engine and request times for a plurality of search queries submitted to said search engine within distinct user sessions;
extracting features from said labeled sample search query traffic in accordance with a set of feature definitions partitioned into physical features related to physical limitations of human generated search queries and behavioral features of automatically generated search queries and keywords of the automatically generated search queries;
generating a feature set comprising (i) said features extracted from said labeled sample search query traffic and (ii) a behavioral feature related to query word length entropy (WLE) that is calculated as:
WL
E
(
l
ij
)
=
-
∑
i
∑
j
l
ij
log
(
l
ij
)
,
i being an index for each separate query submitted to a search engine by a single user ID and I ij being a length of an individual query term j in the ith query;
generating a model using said labeled sample search query traffic and said feature set;
receiving, from said search engine, search query traffic associated with a plurality of search queries submitted by a particular user identifier;
classifying, using said model, said search query traffic associated with said plurality of search queries submitted by said particular user identifier as human generated search query traffic or automatically generated search query traffic; and
altering a quality of service provided by said search engine to said user identifier when said search query traffic associated with said plurality of search queries submitted by said particular user identifier is classified as automatically generated search query traffic.
16. The computer-readable storage medium of claim 15 , wherein said feature set comprises a behavioral feature related to number of open browser identifiers.
17. The computer-readable storage medium of claim 15 , wherein said feature set comprises a behavioral feature related to a click-through rate of query results.
18. The computer-readable storage medium of claim 15 , wherein said feature set comprises a behavioral feature related to a program specified field identifying a program source.