Automatic moderation of online content
Techniques are disclosed for automatically modeling and predicting moderator actions for online content. A model can be generated or updated based on the content received and the action or actions taken by the moderator in response to receiving the content. The model can be used to automatically predict which action, or combination of actions, are likely to be taken by the moderator when new content is received, and suggest those action(s) to the moderator. These suggestions can, among other things, simplify and speed up the decision-making process for the moderator.
1. A computer-implemented method comprising:
receiving first electronic content and second electronic content from a user computing device;
receiving a first input representing a first action taken with respect to the first electronic content;
receiving a second input representing a second action taken with respect to the second electronic content, the second action being different than the first action;
extracting a feature from each of the first electronic content and the second electronic content;
generating a first predictive model based on the feature and the first input, the first predictive model configured to model a dependence of the first action on a presence of the feature;
generating a second predictive model based on the feature and the second input, the second predictive model configured to model a dependence of the second action on the presence of the feature;
generating an inter-action agreement model representing a probability that the first action will be taken with respect to the feature extracted from the first and second electronic content, the inter-action agreement model further representing a probability that the second action will be taken with respect to the feature extracted from the first and second electronic content;
receiving third electronic content from the user computing device;
extracting the feature from the third electronic content;
determining, based on the first and second predictive models, the inter-action agreement model and the feature, a suggested action to be taken with respect to the third electronic content based on the relative probabilities of the first and second actions in the inter-action agreement model; and
presenting the suggested action via a user interface, the suggested action being selectable via the user interface,
wherein the suggested action is taken in response to receiving a third input representing a selection of the suggested action.
2. The method of claim 1 , wherein the feature includes at least one of a content-based feature, a user-based feature and a time-based feature.
3. The method of claim 1 , wherein the predictive model includes a machine learning discriminative model configured to model the dependence of the action on the feature.
4. The method of claim 3 , wherein the machine learning discriminative model includes a support vector machine classifier.
5. The method of claim 1 , wherein the input represents a plurality of actions taken with respect to the first electronic content, and wherein the machine learning discriminative model is further configured to model a dependence of at least one of the actions taken on at least one other of the actions taken.
6. The method of claim 5 , further comprising determining an additional suggested action to be taken based on the suggested action and the predictive model.
7. A system comprising:
a storage; and
a processor operatively coupled to the storage and configured to execute instruction stored in the storage that when executed cause the processor to carry out a process comprising:
receiving first electronic content and second electronic content from a user computing device, a first input representing a first action taken with respect to the first electronic content, and a second input representing a second action taken with respect to the second electronic content;
extracting a feature from each of the first electronic content and the second electronic content;
generating a first predictive model based on the feature and the first input, the first predictive model configured to model a dependence of the first action on a presence of the feature;
generating a second predictive model based on the feature and the second input, the second predictive model configured to model a dependence of the second action on the presence of the feature;
generating an inter-action agreement model representing a probability that the first action will be taken with respect to the feature extracted from the first and second electronic content, the inter-action agreement model further representing a probability that the second action will be taken with respect to the feature extracted from the first and second electronic content;
receiving third electronic content from the user computing device;
extracting the feature from the third electronic content;
determining, based on the first and second predictive models, the inter-action agreement model and the feature, a suggested action to be taken with respect to the third electronic content based on the relative probabilities of the first and second actions in the inter-action agreement model; and
presenting the suggested action via a user interface, the suggested action being selectable via the user interface,
wherein the suggested action is taken in response to receiving a third input representing a selection of the suggested action.
8. The system of claim 7 , wherein the feature includes at least one of a content-based feature, a user-based feature and a time-based feature.
9. The system of claim 7 , wherein the predictive model includes a machine learning discriminative model configured to model the dependence of the action on the feature.
10. The system of claim 9 , wherein the machine learning discriminative model includes a support vector machine classifier.
11. The system of claim 7 , wherein the input represents a plurality of actions taken with respect to the first electronic content, and wherein the machine learning discriminative model is further configured to model a dependence of at least one of the actions taken on at least one other of the actions taken.
12. The system of claim 11 , wherein the process further comprises determining an additional suggested action to be taken based on the suggested action and the predictive model.
13. A non-transitory computer readable medium having instructions encoded thereon that when executed by one or more processors cause a process to be carried out, the process comprising:
receiving first electronic content and second electronic content from a user computing device;
receiving a first input representing a first action taken with respect to the first electronic content;
receiving a second input representing a second action taken with respect to the second electronic content, the second action being different than the first action;
extracting a feature from each of the first electronic content and the second electronic content;
generating a first predictive model based on the feature and the first input, the first predictive model configured to model a dependence of the first action on a presence of the feature;
generating a second predictive model based on the feature and the second input, the second predictive model configured to model a dependence of the second action on the presence of the feature;
generating an inter-action agreement model representing a probability that the first action will be taken with respect to the feature extracted from the first and second electronic content, the inter-action agreement model further representing a probability that the second action will be taken with respect to the feature extracted from the first and second electronic content;
receiving third electronic content from the user computing device;
extracting the feature from the third electronic content;
determining, based on the first and second predictive models, the inter-action agreement model and the feature, a suggested action to be taken with respect to the third electronic content based on the relative probabilities of the first and second actions in the inter-action agreement model; and
presenting the suggested action via a user interface, the suggested action being selectable via the user interface,
wherein the suggested action is taken in response to receiving a third input representing a selection of the suggested action.
14. The non-transitory computer readable medium of claim 13 , wherein the feature includes at least one of a content-based feature, a user-based feature and a time-based feature.
15. The non-transitory computer readable medium of claim 13 , wherein the predictive model includes a machine learning discriminative model configured to model a dependence of the action on the feature.
16. The non-transitory computer readable medium of claim 15 , wherein the machine learning discriminative model includes a support vector machine classifier.
17. The non-transitory computer readable medium of claim 13 , wherein the input represents a plurality of actions taken with respect to the first electronic content, wherein the machine learning discriminative model is further configured to model a dependence of at least one of the actions taken on at least one other of the actions taken, and wherein the process further comprises determining an additional suggested action to be taken based on the suggested action and the predictive model.