Adaptive user interface using machine learning model

1. A computer implemented method comprising:

storing one or more interface rules;

receiving an indication of a metric value to be optimized;

presenting a user interface including a user interface feature to a group of users of a social networking system;

collecting response data from the users experiencing the user interface;

determining a plurality of attribute values for a user attribute of the group of users of the social networking system;

for each attribute value of the plurality of attribute values, calculating a value of the metric value for a subset of the group of users of the social networking system who have the user attribute of that attribute value, based on the response data;

analyzing using a machine learning model the values of the metric value for the subsets of the group of users of the social networking system;

identifying a list of frequently occurring user attributes by examining stored interface rules; and

generating at least one interface rule of applying the user interface feature, wherein the at least one interface rule comprises a type of the user interface feature and a feature value appropriately associated with the type of the user interface feature, wherein the at least one interface rule is generated based on a frequently occurring user attribute from the list of the frequently occurring user attributes and a user attribute closely correlated to the user interface feature, and wherein the type of the user interface feature and the feature value appropriately associated with the type of the user interface feature are generated based on the analysis of the machine learning model so that the metric value is optimized.

2. The computer implemented method of claim 1 , further comprising:

generating a correlation map between user interface features and user attributes by examining stored interface rules;

wherein the user attribute closely correlated to the user interface feature is determined according to the correlation map.

3. The computer implemented method of claim 2 , wherein the correlation map records frequencies of pairs of user attributes and interface features occurring in the stored interface rules, and wherein each frequency indicates a level of correlation between a user attribute and an interface feature of a corresponding pair.

4. The computer implemented method of claim 1 , wherein the metric value is a conversion rate, a registration success rate, a time user spending on user interface, an advertisement click rate, a newsletter subscription rate, or a download rate.

5. The computer implemented method of claim 1 , wherein the user interface features include at least one of user interface element, sequence of interface flow, or user interface text.

6. The computer implemented method of claim 1 , further comprising:

presenting an original user interface including an original user interface feature to a group of control users; and

collecting response data from the control users experiencing the original user interface.

7. The computer implemented method of claim 1 , wherein the response data includes data of user inputs.

8. The computer implemented method of claim 1 , wherein the response data includes time which users spend on the user interface.

9. The computer implemented method of claim 1 , further comprising:

randomly selecting a plurality of users as a group of test users.

10. The computer implemented method of claim 1 , wherein the interface rule of applying user interface features is a decision rule that is human comprehensible.

11. The computer implemented method of claim 1 , wherein the group of users of the social networking system has such a size that the interface rule is statistically significant for a population of users.

12. The computer implemented method of claim 1 , wherein the presenting comprises:

presenting a plurality of user interfaces to a plurality groups of test users, each user interface of the plurality of user interfaces including at least one new user interface feature.

13. The computer implemented method of claim 1 , further comprising:

storing the interface rule;

generating a user interface according to the interface rule; and

presenting the user interface to a user, wherein the user is associated with an attribute value that matches an attribute value associated with the interface rule.

14. One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

store one or more interface rules;

receive an indication of a metric value to be optimized;

present a user interface including a user interface feature to a group of users of a social networking system;

collect response data from the users experiencing the user interface;

determine a plurality of attribute values for a user attribute of the group of users of the social networking system;

for each attribute value of the plurality of attribute values, calculate a value of the metric value for a subset of the group of users of the social networking system who have the user attribute of that attribute value, based on the response data;

analyze using a machine learning model the values of the metric value for the subsets of the group of users of the social networking system;

identify a list of frequently occurring user attributes by examining stored interface rules; and

generate at least one interface rule of applying the user interface feature, wherein the at least one interface rule comprises a type of the user interface feature and a feature value appropriately associated with the type of the user interface feature, wherein the at least one interface rule is generated based on a frequently occurring user attribute from the list of the frequently occurring user attributes and a user attribute closely correlated to the user interface feature, and wherein the type of the user interface feature and the feature value appropriately associated with the type of the user interface feature are generated based on the analysis of the machine learning model so that the metric value is optimized.

15. A system comprising: one or more processors; and a memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:

store one or more interface rules;

receive an indication of a metric value to be optimized;

present a user interface including a user interface feature to a group of users of a social networking system;

collect response data from the users experiencing the user interface;

determine a plurality of attribute values for a user attribute of the group of users of the social networking system;

for each attribute value of the plurality of attribute values, calculate a value of the metric value for a subset of the group of users of the social networking system who have the user attribute of that attribute value, based on the response data;

analyze using a machine learning model the values of the metric value for the subsets of the group of users of the social networking system;

identify a list of frequently occurring user attributes by examining stored interface rules; and

generate at least one interface rule of applying the user interface feature, wherein the at least one interface rule comprises a type of the user interface feature and a feature value appropriately associated with the type of the user interface feature, wherein the at least one interface rule is generated based on a frequently occurring user attribute from the list of the frequently occurring user attributes and a user attribute closely correlated to the user interface feature, and wherein the type of the user interface feature and the feature value appropriately associated with the type of the user interface feature are generated based on the analysis of the machine learning model so that the metric value is optimized.