Time and click based updatable static web page ranking

1 . A method comprising:

identifying a web page at a discovery time, wherein, at the discovery time, the web page is not included in a search index;

ranking, at discovery time, the web page, wherein ranking the web page comprises:

predicting, by a first ranking model, a first ranking of the web page using first features available at the discovery time, the first ranking model having a first set of weights learned by training on a training set in a masked state, the training set comprising discovery features and post-discovery features, the training set masked to include only the discovery features;

predicting, by a second ranking model, a second ranking of the web page using the first features, the second ranking model having a second set of weights learned by training on the training set in an unmasked state including the discovery features and the post-discovery features;

generating an index score from a combination of the first ranking and the second ranking; and

placing, at the discovery time, the web page within the search index at a position defined by the index score; and

re-ranking, after the discovery time, the web page, wherein re-ranking the web page comprises:

predicting, by the first ranking model after the discovery time, an updated first ranking of the web page using only the first features;

predicting, by the second ranking model after the discovery time, an updated second ranking of the web page using the first features and second features not available at the discovery time;

updating the index score from a combination of the updated first ranking and the updated second ranking; and

adjusting, after the discovery time, the position of the web page within the search index based in part on the updated index score.

2 . The method of claim 1 , further comprising:

determining, at the discovery time and from the index score, that the web page does not qualify to be placed within a search index; and

placing, after the discovery time and from the updated index score, the web page within the search index at a position defined by the updated index score.

3 . The method of claim 1 , further comprising:

placing, at the discovery time, the web page within a search index at a position defined by the index score; and

removing, after the discovery time and based at least in part on the updated index score, the web page from the search index.

4 . The method of claim 1 , wherein the first features comprise one or more of web page metadata including a host identifier for the web page and a domain for the web page, parent link metadata including a host identifier for a parent web page, a search index rank of the parent web page, and a page authority of the parent web page, and the discovery time.

5 . The method of claim 1 , wherein the second features comprise one or more of a time series of click data, usefulness heuristics including how often the web page is clicked when provided in search results, popularity heuristics including an observed or reported traffic to the web page and a trending measure, a page authority including a number of other web pages known to link to the respective web page and a quality of the other web pages, and click metadata including a local client time when clicked and a client location when clicked.

6 . The method of claim 1 , further comprising training the first ranking model and the second ranking model using labeled training data comprising a plurality of web pages having target rankings in a search index.

7 . The method of claim 6 , wherein training the first ranking model comprises partially masking the labeled training data to remove post-discovery data.

8 . The method of claim 7 , wherein weights of the first ranking model differ from weights of the second ranking model.

9 . A system having a memory comprising a non-transitory computer readable media, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:

receiving a first trigger comprising a discovery of a web page at a discovery time, wherein, at the discovery time, the web page is not included in a search index;

in response to the first trigger, ranking, at discovery time, the web page, wherein ranking the web page comprises:

predicting, by a first ranking model, a first ranking of the web page using first features available at the discovery time, the first ranking model having a first set of weights learned by training on a training set in a masked state, the training set comprising discovery features and post-discovery features, the training set masked to include only the discovery features;

predicting, by a second ranking model, a second ranking of the web page using the first features, the second ranking model having a second set of weights learned by training on the training set in an unmasked state including the discovery features and the post-discovery features;

generating an index score from a combination of the first ranking and the second ranking; and

placing, at the discovery time, the web page within the search index at a position defined by the index score; and

receiving a second trigger comprising one of a click event for the web page and a time elapsed for the web page;

in response to the second trigger, re-ranking, after the discovery time, the web page, wherein re-ranking the web page comprises:

predicting, by the first ranking model after the discovery time, an updated first ranking of the web page using only the first features;

predicting, by the second ranking model after the discovery time, an updated second ranking of the web page using the first features and second features not available at the discovery time;

updating the index score from a combination of the updated first ranking and the updated second ranking; and

adjusting, after the discovery time, the position of the web page within the search index based in part on the updated index score.

10 . The system of claim 9 , wherein the click event comprises detecting a click, by a user, of a link to the web page.

11 . The system of claim 9 , wherein the time elapsed for the web page comprises a predetermined interval after which the respective web page is re-evaluated.

12 . The system of claim 11 , wherein receiving the second trigger comprises determining that the time elapsed exceeds the predetermined interval.

13 . The system of claim 12 , wherein the time elapsed for the web page is reset after updating the index score.

14 . The system of claim 9 , wherein the first features comprise one or more of web page metadata including a host identifier for the web page and a domain for the web page, parent link metadata including a host identifier for a parent web page, a search index rank of the parent web page, and a page authority of the parent web page, and the discovery time; and

wherein the second features comprise one or more of a time series of click data, usefulness heuristics including how often the web page is clicked when provided in search results, popularity heuristics including an observed or reported traffic to the web page and a trending measure, a page authority including a number of other web pages known to link to the respective web page and a quality of the other web pages, and click metadata including a local client time when clicked and a client location when clicked.

15 . A system having a memory comprising a non-transitory computer readable media, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:

training a first ranking model to predict a first ranking of a web page using first features available at a discovery time, the first ranking model having a first set of weights learned by training on a training set in a masked state, the training set comprising discovery features and post-discovery features, the training set masked to include only the discovery features;

training a second ranking model to predict a second ranking of the web page, the second ranking model having a second set of weights learned by training on the training set in an unmasked state including the discovery features and the post-discovery features;

ranking, at discovery time, the web page, wherein, at the discovery time, the web page is not included in a search index, and wherein ranking the web page comprises:

predicting, by the first ranking model, a first ranking of the web page using first features available at the discovery time;

predicting, by the second ranking model, a second ranking of the web page using the first features;

generating an index score from a combination of the first ranking and the second ranking; and

placing, at the discovery time, the web page within the search index at a position defined by the index score; and

re-ranking, after the discovery time, the web page, wherein re-ranking the web page comprises:

predicting, by the first ranking model after the discovery time, an updated first ranking of the web page using only the first features;

predicting, by the second ranking model after the discovery time, an updated second ranking of the web page using the first features and second features not available at the discovery time;

updating the index score from a combination of the updated first ranking and the updated second ranking; and

adjusting, after the discovery time, the position of the web page within the search index based in part on the updated index score.

16 . The system of claim 15 , wherein weights of the first ranking model differ from weights of the second ranking model.

17 . The system of claim 15 , wherein the computer readable instructions further control the one or more processors to, in response to receiving a first trigger comprising a discovery of the web page at the discovery time:

predict, using the first ranking model, the first ranking of the web page using the first features;

predict, using the second ranking model, the second ranking of the web page using the first features; and

generate an index score from a combination of the first ranking and the second ranking.

18 . The system of claim 17 , wherein the computer readable instructions further control the one or more processors to, in response to receiving a second trigger comprising one of a click event for the web page and a time elapsed for the web page:

predict, using the first ranking model after the discovery time, an updated first ranking of the web page using the first features;

predict, using the second ranking model after the discovery time, an updated second ranking of the web page using the first features and the second features; and

update the index score from a combination of the updated first ranking and the updated second ranking.

19 . The system of claim 18 , wherein the computer readable instructions further control the one or more processors to:

placing, at the discovery time, the web page within a search index at a position defined by the index score; and

adjust, after the discovery time, the position of the web page within the search index based in part on the updated index score.