Abstract: For various content campaigns (or content), an online system generates a score indicating a likelihood of the content item having deceptive information, such as including a picture or name of a celebrity to promote something that the celebrity has not actually endorsed. The online system receives a request to determine whether a content item comprises deceptive information. The online system extracts features from the content item, and provides the extracted features to a machine learning based model configured to generate score indicating whether a content item comprises deceptive information. The online system executes the machine learning based model to generate the score for the content item. Responsive to the generated score indicating that content item comprises deceptive information, the online system verifies whether the content item conforms to content policies.

Abstract: An online system receives a content item including a link to a landing page and determines a likelihood the landing page violates an online system policy based on a structural similarity between the landing page and a web page violating the policy. To determine the likelihood, the online system determines a hierarchical structure associated with the web page violating the policy and an additional hierarchical structure associated with the landing page. The hierarchical structure represents a structure of at least a portion of the web page and the additional hierarchical structure represents a structure of a corresponding portion of the landing page. The online system compares the hierarchical structure and additional hierarchical structure. Based on the comparison, the online system computes a measure of dissimilarity between the hierarchical structure and additional hierarchical structure and determines a likelihood the landing page violates the policy based on the measure of dissimilarity.

Abstract: A content review system for an online system automatically determines if received content items to be displayed to users violate any policies of the online system. The content review system generates a semantic vector representing the semantic features of a content item, for example, using a neural network. By comparing the semantic vector for the content item with semantic vectors of content items previously determined to violate one or more policies, the content review system determines whether the content item also violates one or more policies. The content review system may also maintain templates corresponding to portions of semantic vectors shared by multiple content items. An analysis of historical content items that conform to the template is performed to determine a probability that received content items that conform to the template violate a policy.

Abstract: A content review system for an online system automatically determines if received content items to be displayed to users violate any policies of the online system. The content review system generates a semantic vector representing the semantic features of a content item, for example, using a neural network. By comparing the semantic vector for the content item with semantic vectors of content items previously determined to violate one or more policies, the content review system determines whether the content item also violates one or more policies. The content review system may also maintain templates corresponding to portions of semantic vectors shared by multiple content items. An analysis of historical content items that conform to the template is performed to determine a probability that received content items that conform to the template violate a policy.

Abstract: An online system receives a content item including a link to a landing page and determines a likelihood the landing page violates an online system policy based on a structural similarity between the landing page and a web page violating the policy. To determine the likelihood, the online system determines a hierarchical structure associated with the web page violating the policy and an additional hierarchical structure associated with the landing page. The hierarchical structure represents a structure of at least a portion of the web page and the additional hierarchical structure represents a structure of a corresponding portion of the landing page. The online system compares the hierarchical structure and additional hierarchical structure. Based on the comparison, the online system computes a measure of dissimilarity between the hierarchical structure and additional hierarchical structure and determines a likelihood the landing page violates the policy based on the measure of dissimilarity.

Abstract: For various content campaigns (or content), an online system predicts a likelihood score of context violations (e.g., account term violations) of a content campaign. The online system derives a plurality of feature vectors of the content campaign. The online system predicts a likelihood score of context violation of the content campaign using a memorization model based on the plurality of feature vectors. The memorization model comprises a plurality of categories and a plurality of items of each category. Each of the plurality of categories has a category weight, and each of the plurality of items of each category has an item weight. The predicted likelihood score is based on a combination of a plurality of category weights and a plurality of item weights associated with the plurality of feature vectors. The online system performs an action affecting the content campaign based in part on the predicted likelihood score.

Abstract: An online system determines a quality of content provided by third party systems for distribution to users. The online system analyzes URL's posted within the online system by content providers to determine the quality of content of the webpages obtained by accessing the URLs. For each URL, the online system receives an original markup language document and a copy of the markup document obtained by applying a content filter. The online system extracts features from both markup language documents. The online system provides the extracted features to a machine learning based model to generate a content quality score. The online system categorizes the URL as having high quality content or low quality content. The online system restricts distribution of content items including URLs to websites with low quality content.

Abstract: An online system distributes content items provided by content providers. The online system determines a likelihood of a content item having deceptive information. The online system stores images showing faces of people in an image database. The online system extracts features from the content item, and provides the extracted features to a machine learning based model configured to generate score indicating whether a content item comprises deceptive information. The machine learning based model uses at least a feature based on matching of faces of users shown in the content item with faces of users shown in the images of the image database. If the online system determines that a content item is deceptive, the online system adds images comprising faces extracted from the content item to the image database to grow the image database.

Abstract: An online system reviews various user profiles for compliance with policies enforced by the online system. However, users may attempt to subvert action by the online system by creating additional user profiles for presenting content. Accordingly, the online system generates a graph identifying connections user profiles, content items associated with the user profiles, and objects identified by the content items. User profiles, content items, or objects previously identified to have violated one or more policies enforced by the online system are identified via the graph. The online system computes a profile score for various user profiles based on a probability of reaching an object, user profile, or content item identified as violating a policy through a random walk in the graph. Based on the profile scores, the online system trains to identify user profiles for review against one or more enforced policies.

Abstract: A content review system for an online system automatically determines if received content items to be displayed to users violate any policies of the online system. The content review system generates a semantic vector representing the semantic features of a content item, for example, using a neural network. By comparing the semantic vector for the content item with semantic vectors of content items previously determined to violate one or more policies, the content review system determines whether the content item also violates one or more policies. The content review system may also maintain templates corresponding to portions of semantic vectors shared by multiple content items. An analysis of historical content items that conform to the template is performed to determine a probability that received content items that conform to the template violate a policy.

Abstract: A content review system for an online system automatically determines if received content items to be displayed to users violate any policies of the online system. The content review system generates a semantic vector representing the semantic features of a content item, for example, using a neural network. By comparing the semantic vector for the content item with semantic vectors of content items previously determined to violate one or more policies, the content review system determines whether the content item also violates one or more policies. The content review system may also maintain templates corresponding to portions of semantic vectors shared by multiple content items. An analysis of historical content items that conform to the template is performed to determine a probability that received content items that conform to the template violate a policy.

Abstract: An online system receives a content item including a link to a landing page and determines a likelihood the landing page violates an online system policy based on a structural similarity between the landing page and a web page violating the policy. To determine the likelihood, the online system determines a hierarchical structure associated with the web page violating the policy and an additional hierarchical structure associated with the landing page. The hierarchical structure represents a structure of at least a portion of the web page and the additional hierarchical structure represents a structure of a corresponding portion of the landing page. The online system compares the hierarchical structure and additional hierarchical structure. Based on the comparison, the online system computes a measure of dissimilarity between the hierarchical structure and additional hierarchical structure and determines a likelihood the landing page violates the policy based on the measure of dissimilarity.

Abstract: An online system reviews various user profiles for compliance with policies enforced by the online system. However, users may attempt to subvert action by the online system by creating additional user profiles for presenting content. Accordingly, the online system generates a graph identifying connections user profiles, content items associated with the user profiles, and objects identified by the content items. User profiles, content items, or objects previously identified to have violated one or more policies enforced by the online system are identified via the graph. The online system computes a profile score for various user profiles based on a probability of reaching an object, user profile, or content item identified as violating a policy through a random walk in the graph. Based on the profile scores, the online system trains to identify user profiles for review against one or more enforced policies.

Abstract: For various content campaigns (or content), an online system predicts a likelihood score of context violations (e.g., account term violations) of a content campaign. The online system derives a plurality of feature vectors of the content campaign. The online system predicts a likelihood score of context violation of the content campaign using a memorization model based on the plurality of feature vectors. The memorization model comprises a plurality of categories and a plurality of items of each category. Each of the plurality of categories has a category weight, and each of the plurality of items of each category has an item weight. The predicted likelihood score is based on a combination of a plurality of category weights and a plurality of item weights associated with the plurality of feature vectors. The online system performs an action affecting the content campaign based in part on the predicted likelihood score.

Abstract: An online system obtains risk scores determined by a machine learning model for a content item provided by a user of an online system for display to users of the online system, where the risk scores indicate the likelihood of content items violating a content policy. The online system uses the risk scores to determine sampling weights used to select content items for inclusion in a sampled subset of content items. The sampling weights are determined from risk score counts indicating the relative frequency of the obtained risk scores and impression counts indicating the number of times content items have been presented to the users of the online system. The online system presents the selected content items for evaluation by a human reviewer using a quality review interface. Using the results of the quality review, the online system determines quality performance metrics of the machine learning model.

Abstract: An online system obtains risk scores determined by a machine learning model for a content item provided by a user of an online system for display to users of the online system, where the risk scores indicate the likelihood of content items violating a content policy. The online system uses the risk scores to determine sampling weights used to select content items for inclusion in a sampled subset of content items. The sampling weights are determined from risk score counts indicating the relative frequency of the obtained risk scores and impression counts indicating the number of times content items have been presented to the users of the online system. The online system presents the selected content items for evaluation by a human reviewer using a quality review interface. Using the results of the quality review, the online system determines quality performance metrics of the machine learning model.

Abstract: An online system receives advertisements from advertisers and reviews the advertisement for compliance with policies enforced by the online system. The online system computes scores for each advertisement based on an expected revenue from presenting various advertisement and/or interactions with various advertisements and orders advertisements for review based on their scores. If a predicted time for the online system to review an advertisement is greater than a threshold amount of time, the online system allows the online system to be evaluated for presentation to users. As the online system receives interactions with the advertisement, the online system may modify the score for the advertisement and modify the order of the advertisement for review based on the modified score.