Predicting future queries from log data
A system, media, and method for selecting future queries are provided. The selected future queries are used to transmit appropriate online advertising to a user that issues queries to a search engine. The search engine is coupled to a prediction component that predicts what subject the user is going to be interested in and when the user will be interested in the subject. The prediction component returns a future query using statistical language models representing a query history of the user and aggregate query histories for a community of users.
1. A computing system configured to select a future query, the system comprising:
one or more search engines configured to receive queries from a user and to provide results to the user;
one or more query logs coupled to the one or more search engines and configured to store queries issued by users, who submit queries to the one or more search engines; and
one or more prediction components configured to select future queries that the user is likely to issue in a certain time period based on a query history for the user and aggregated behaviors for a group of users corresponding to the queries stored in the query log for the group of users.
2. The computing system of claim 1 , further comprising one or more advertising platforms configured to use the future queries to select proper advertisements as a function of time.
3. The computing system of claim 1 , wherein the selected advertisement is communicated to the user proximate to the received query.
4. The computing system of claim 1 , wherein the selected advertisement is communicated to the user after a period of time has expired since the query was received.
5. The computing system of claim 1 , wherein the future queries and user queries are related, the relation between the received query and the future query is a function of a time period.
6. The computing system of claim 5 , wherein the time period is one of a quarter year, a month, a bi-week, or a week.
7. One or more computer readable media storing instructions for performing a method of determining a likelihood of a future query, the method comprising:
collecting data from searches issued by users;
receiving a user query at a certain time;
generating prediction language models from the collected data; and
predicting a likelihood of a future query based on the model and a query history of the user.
8. The media of claim 7 , further comprising selecting proper advertisements corresponding to the future queries.
9. The media of claim 7 , wherein generating prediction language models from the collected data further comprises:
traversing the collected data to identify a background model for the prediction language model, wherein the background model comprises common queries included in the collection of data.
10. The media of claim 9 , wherein generating prediction language models from the collected data further comprises:
locating a query history of the user; and
comparing the collected data to the query history of the user to identify a lifetime model for the user.
11. The media of claim 10 , wherein generating prediction language models from the collected data further comprises:
tuning a combination of the background model and the lifetime model; and
using the tuned combination of the background model and lifetime model to calculate probabilities of future queries during a period of time.
12. The media of claim 7 , further comprising assigning probabilities to one or more future queries based on previous activities of the user observed through the received queries and common behaviors of all users obtained from the collected data.
13. The media of claim 7 , wherein the future query and user query are related, the relation between the user query and the future query is a function of time period and relevance.
14. A computer-implemented method to select a future query, the method comprising:
receiving a query from a user;
determining whether a user query history is available for the user;
when a query history of the user is unavailable, generating a background model query as the future query for the user; and
when a query history of the user is available, comparing the query history of the user to an aggregate query history for other users and statistically selecting a query as the future query for the user.
15. The method of claim 14 , wherein the queries issued by other users are stored in a query log.
16. The method of claim 15 , wherein the query log is partitioned and clustered to create the aggregate query history for other users.
17. The method of claim 14 , wherein statistically selecting a query as the future query uses a language model having observed probabilities corresponding with other users.
18. The method of claim 14 , wherein the future query is likely to be issued by the user within a period of time from the received query.
19. The method of claim 14 , further comprising:
selecting an advertisement corresponding to the future query; and
communicating the advertisement to the user.
20. The method of claim 19 , wherein the advertisement is communicated to the user proximate to a time the query was issued or the advertisement is communicated after a period of time has expired since the query was received, the period of time is one of a quarter year, a month, a bi-week, or a week.