Optimizing search result snippet selection

1. A method performed by a computing system comprising one or more processors and storage media storing one or more programs executed by the one or more processors to perform the method, the method comprising:

receiving a query and identifying a set of documents matching the query;

selecting, for a snippet representing a corresponding document of the set of documents, a set of a plurality of snippet field types from a plurality of predefined snippet field types, the selecting based at least in part on a snippet selection model;

causing display of the snippet, the snippet display including a separate value for each snippet field type in the set of the plurality of snippet field types selected for the snippet

detecting a user click on the snippet;

determining a user dwell time, associated with the user click, on the corresponding document;

based at least in part on the user click and the user dwell time, adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types for future queries;

wherein, before the adjusting, the snippet selection model is configured to apply a first set of weights to a set of parameters of the snippet selection model;

wherein, after the adjusting, the snippet selection model is configured to apply a second set of weights to the set of parameters of the snippet selection model;

wherein the second set of weights are different than the first set of weights;

wherein the second set of weights is determined based at least in part on:

assigning a relevance label to the set of the plurality of snippet field types selected,

wherein the relevance label is assigned to the set of the plurality of snippet field types selected based at least in part on the user dwell time on the corresponding document,

determining a ground truth ranking of the plurality of predefined snippet field types;

wherein the ground truth ranking is determined based at least in part on the relevance label assigned, and

determining a set of weights that, when applied to the set of parameters of the snippet selection model, minimizes a loss function with respect to at least the ground truth ranking.

2. The method of claim 1 , further comprising:

based at least in part on the user click and the user dwell time, adjusting the snippet selection model to reinforce selection of the set of the plurality of snippet field types for future queries.

3. The method of claim 1 , further comprising:

based at least in part on the user click and the user dwell time, adjusting the snippet selection model to discourage selection of the set of the plurality of snippet field types for future queries.

4. The method of claim 1 , wherein:

the user dwell time is based at least in part on an amount of time that at least a portion of the corresponding document is presented in a graphical user interface in response to the user click on the snippet.

5. The method of claim 1 , wherein:

the adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types is further based at least in part on a ranking score for the corresponding document; and

the ranking score is based at least in part on a query-dependent score for at least the query and the corresponding document.

6. The method of claim 1 , wherein:

the adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types is further based at least in part on a ranking score for the corresponding document; and

the ranking score is based at least in part on a query-independent score for the corresponding document.

7. A computing system comprising:

one or more processors;

storage media; and

one or more programs stored in the storage media and configured for execution by the one or more processors, the one or more programs comprising instructions configured for:

receiving a query and identifying a set of documents matching the query;

selecting, for a snippet representing a corresponding document of the set of documents, a set of a plurality of snippet field types from a plurality of predefined snippet field types, the selecting based at least in part on a snippet selection model;

causing display of the snippet, the snippet display including a separate value for each snippet field type in the set of the plurality of snippet field types selected for the snippet;

detecting a user click on the snippet;

determining a user dwell time, associated with the user click, on the corresponding document; and

based at least in part on the user click and the user dwell time, adjusting the snippet selection model to discourage selection of the set of the plurality of snippet field types for future queries;

wherein, before the adjusting, the snippet selection model is configured to apply a first set of weights to a set of parameters of the snippet selection model;

wherein, after the adjusting, the snippet selection model is configured to apply a second set of weights to the set of parameters of the snippet selection model;

wherein the second set of weights are different than the first set of weights;

wherein the second set of weights is determined based at least in part on:

assigning a relevance label to the set of the plurality of snippet field types selected,

wherein the relevance label is assigned to the set of the plurality of snippet field types selected based at least in part on the user dwell time on the corresponding document,

determining a ground truth ranking of the plurality of predefined snippet field types;

wherein the ground truth ranking is determined based at least in part on the relevance label assigned, and

determining a set of weights that, when applied to the set of parameters of the snippet selection model, minimizes a loss function with respect to at least the ground truth ranking.

8. The computing system of claim 7 , wherein

the adjusting the snippet selection model to discourage selection of the set of the plurality of snippet field types is based at least in part on determining that the user dwell time on the corresponding document did not exceed a threshold amount of time; and

the user dwell time is based at least in part on an amount of time that at least a portion of the corresponding document is presented in a graphical user interface in response to the user click on the snippet.

9. The computing system of claim 7 , wherein:

the user click is a first user click;

the snippet is a first snippet;

the corresponding document is a first corresponding document;

the instructions are further configured for detecting a second user click, that is not the first user click, on a second snippet, that is not the first snippet, representing a second corresponding document of the set of documents, that is not the first corresponding document;

the adjusting the snippet selection model to discourage selection of the set of the plurality of snippet field types is based at least in part on determining that the second user click happened before the first user click.

10. The computing system of claim 9 , wherein:

the adjusting the snippet selection model to discourage selection of the set of the plurality of snippet field types is based at least in part on determining that a user dwell time on the first corresponding document exceeds a threshold amount of time; and

the user dwell time is based at least in part on an amount of time that at least a portion of the first corresponding document is presented in a graphical user interface in response to the first user click on the first snippet.

11. The computing system of claim 9 , wherein:

the user click on the first snippet is performed while the first snippet and the second snippet are displayed in a graphical user interface;

the first snippet is displayed at a first respective position in the graphical user interface;

the second snippet is displayed at a second respective position in the graphical user interface; and

the first respective position and the second respective position are such they visually indicate that the first corresponding document is more relevant to the query than the second corresponding document.

12. One or more non-transitory computer-readable media storing one or more programs for execution by one or more processors, the one or more programs comprising instructions configured for:

receiving a query and identifying a set of documents matching the query;

selecting, for a snippet representing a corresponding document in the set of documents, a set of a plurality of snippet field types from a plurality of predefined snippet field types, the selecting based at least in part on a snippet selection model;

causing display of the snippet in a graphical user interface, the snippet display including a separate value for each snippet field type in the set of the plurality of snippet field types selected for the snippet;

detecting a user click on the snippet;

detecting a user dwell time, associated with the user click, on the corresponding document; and

based at least in part on the user click and the user dwell time, adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types for future queries;

wherein, before the adjusting, the snippet selection model is configured to apply a first set of weights to a set of parameters of the snippet selection model;

wherein, after the adjusting, the snippet selection model is configured to apply a second set of weights to the set of parameters of the snippet selection model;

wherein the second set of weights are different than the first set of weights;

wherein the second set of weights is determined based at least in part on:

assigning a relevance label to the set of the plurality of snippet field types selected,

wherein the relevance label is assigned to the set of the plurality of snippet field types selected based at least in part on the user dwell time on the corresponding document,

determining a ground truth ranking of the plurality of predefined snippet field types;

wherein the ground truth ranking is determined based at least in part on the relevance label assigned, and

determining a set of weights that, when applied to the set of parameters of the snippet selection model, minimizes a loss function with respect to at least the ground truth ranking.

13. The one or more non-transitory computer-readable media of claim 12 , wherein:

the snippet is a first snippet;

the corresponding document is a first corresponding document;

the instructions are further configured for adjusting the snippet selection model to discourage selection of the set of a plurality of snippet field types based at least in part on all of: a user click on a second snippet that is not the first snippet representing a second corresponding document in the set of documents that is not the first corresponding document, a first ranking score for the first corresponding document, and a second ranking score for the second corresponding document;

the user click on the second snippet is performed while the first snippet and the second snippet are presented in the graphical user interface;

the user click on the second snippet is performed before any click on the first snippet;

the first snippet is presented in the graphical user interface at a first respective position based at least in part in the first ranking score;

the second snippet is presented in the graphical user interface at a second respective position based at least in part on the second ranking score; and

the first respective position and the second respective position are such they visually indicate that the first corresponding document is more relevant to the query than the second corresponding document.

14. The one or more non-transitory computer-readable media of claim 12 , wherein:

the instructions further configured for computing a weighted snippet field score for each of a plurality of snippet field types based at least in part on the corresponding document; and

wherein the adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types is based at least in part on adjusting weights applied to snippet field scores to reinforce or discourage selection of the set of the plurality of snippet field types.

15. The one or more non-transitory computer-readable media of claim 12 , wherein at least one snippet field type of the set of the plurality of snippet field types is:

a privacy snippet field,

a path snippet field,

a truncated path snippet field,

a path tail snippet field,

a shared with snippet field,

a last viewed snippet field,

an edited snippet field,

a commented snippet field,

a doc copy snippet field,

a presence snippet field, or

a logical organization snippet field.

16. The one or more non-transitory computer-readable media of claim 12 , wherein the adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types is based at least in part on detecting that the snippet did not receive a user click.

17. The one or more non-transitory computer-readable media of claim 12 , the instructions further configured for:

computing a ranking score for each document in the set of documents; and

wherein the adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types is based at least in part on detecting that the corresponding document received a best ranking score among the ranking scores computed for the set of documents.

18. The one or more non-transitory computer-readable media of claim 12 , the instructions further configured for:

computing a ranking score for the corresponding document; and

wherein the adjusting the snippet selection model to reinforce or discourage selection of the set of the plurality of snippet field types is based at least in part on the user click and the ranking score.

19. The method of claim 1 , wherein the separate value for each snippet field type pertains particularly to metadata about the corresponding document.