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: A composer interface allows a user to provide input, such as text data, to the social networking system. To reduce the number of inputs for a user to provide to the social networking system, one or more phrase objects are presented to the user when providing input. A phrase object includes a group of objects that users have previously provided to the social networking system. The social networking system presents phrase objects to the user that match at least a portion of the received input. Context information associated with the input and associated with various phrase objects may be used to select the presented phrase objects. By selecting a presented phrase object, the user includes the selected phrase object in the input.

Abstract: In one embodiment, a method includes deriving input topics based on a content item, generating a matrix of scores for the input topics according to a first set of cross-indexed topics, where each of the scores indicates a degree of similarity between a corresponding one of the input topics and a corresponding one of the first cross-indexed topics, calculating a total score for each of the first cross-indexed topics based on the scores for the first cross-indexed topic across all of the input topics, and selecting one or more of the first cross-indexed topics based on the total scores of the first cross-indexed topics. Deriving the input topics may include using a topic tagger to identify the topics based on the content item. The first set of cross-indexed topics may be generated from a database of topics, such as an online encyclopedia.

Abstract: A social networking system provides access to personas comprising information, for example, web pages describing users or entities. The information may be suggested by the social networking system or requested by the user, for example, via search. The social networking system authenticates the personas so that only authentic personas are suggested to users or returned when a user is searching for information. The authenticity of a persona is determined based on the connections and/or likes coming from other personas, user accounts, or other entities represented within the social networking system that have been previously authenticated. The authenticity of the person is also determined based on external links to the persona, for example, external websites referring to the persona or the rate at which external systems such as search engines direct web traffic to the persona.

Abstract: A social networking system determines the meaning of an anchor term used in a communication received from a communicating user. Candidate nodes are identified in the dictionary based on the anchor term, where each candidate node represents a possible meaning of the anchor term. The context of the anchor term is determined, and a score is determined for each candidate node based on the determined context. A candidate node is selected that most likely represents the meaning of the anchor term based on the determined candidate node scores. The context of the anchor term may be a social context derived from users connected to the communicating user that use the anchor term in communications. A communicating user may be prompted to identify the meaning of the anchor term explicitly based on the use of the term in communications from other users connected to the communicating user.

Abstract: A composer interface allows a user to provide input, such as text data, to the social networking system. To reduce the number of inputs for a user to provide to the social networking system, one or more phrase objects are presented to the user when providing input. A phrase object includes a group of objects that users have previously provided to the social networking system. The social networking system presents phrase objects to the user that match at least a portion of the received input. Context information associated with the input and associated with various phrase objects may be used to select the presented phrase objects. By selecting a presented phrase object, the user includes the selected phrase object in the input.

Abstract: Identifying whether an image includes a search image can be accomplished using region analysis and/or parts model analysis. Region analysis can include dividing a received image into regions and analyzing each region by: enlarging the region, applying a gradient histogram transformation to the enlarged region, and comparing the transformed region, at various angles, to a set of search patterns. Parts model analysis can include identifying key features of an image such as prominent lines or areas, objects or object edges, consecutive colors; generating one or more parts models from the key features by representing the key features as geometric objects; and comparing each parts model, at various angles, to a set of search patterns. The comparison employed in region analysis or parts model analysis can employ classifiers trained on the search images.

Abstract: In one embodiment, a method includes deriving input topics based on a content item, generating a matrix of scores for the input topics according to a first set of cross-indexed topics, where each of the scores indicates a degree of similarity between a corresponding one of the input topics and a corresponding one of the first cross-indexed topics, calculating a total score for each of the first cross-indexed topics based on the scores for the first cross-indexed topic across all of the input topics, and selecting one or more of the first cross-indexed topics based on the total scores of the first cross-indexed topics. Deriving the input topics may include using a topic tagger to identify the topics based on the content item. The first set of cross-indexed topics may be generated from a database of topics, such as an online encyclopedia.

Abstract: A arbitrated image classifier can be trained to identify whether an image contains specified features, such as sexual, violent, or other potentially objectionable content. An arbitrated image classifier can include a configuration of classifiers and an arbitrator that determines a final image classification based on classification results from the classifiers. An arbitrated image classifier can be trained to identify image features by dividing images labeled as including or not including a specified feature into portions, which are provided to the classifiers of the arbitrated image classifier. The arbitrator of the arbitrated image classifier can determine a result for whether or not the image includes the specified feature. If the final result does not match the image label, parameter values for various of the classifiers or the arbitrator combining procedure can be adjusted. A trained arbitrated image classifier can then be used to determine whether new images include the particular feature.

Abstract: A composer interface allows a user to provide input, such as text data, to the social networking system. To reduce the number of inputs for a user to provide to the social networking system, one or more phrase objects are presented to the user when providing input. A phrase object includes a group of objects that users have previously provided to the social networking system. The social networking system presents phrase objects to the user that match at least a portion of the received input. Context information associated with the input and associated with various phrase objects may be used to select the presented phrase objects. By selecting a presented phrase object, the user includes the selected phrase object in the input.

Abstract: A arbitrated image classifier can be trained to identify whether an image contains specified features, such as sexual, violent, or other potentially objectionable content. An arbitrated image classifier can include a configuration of classifiers and an arbitrator that determines a final image classification based on classification results from the classifiers. An arbitrated image classifier can be trained to identify image features by dividing images labeled as including or not including a specified feature into portions, which are provided to the classifiers of the arbitrated image classifier. The arbitrator of the arbitrated image classifier can determine a result for whether or not the image includes the specified feature. If the final result does not match the image label, parameter values for various of the classifiers or the arbitrator combining procedure can be adjusted. A trained arbitrated image classifier can then be used to determine whether new images include the particular feature.

Abstract: A arbitrated image classifier can be trained to identify whether an image contains specified features, such as sexual, violent, or other potentially objectionable content. An arbitrated image classifier can include a configuration of classifiers and an arbitrator that determines a final image classification based on classification results from the classifiers. An arbitrated image classifier can be trained to identify image features by dividing images labeled as including or not including a specified feature into portions, which are provided to the classifiers of the arbitrated image classifier. The arbitrator of the arbitrated image classifier can determine a result for whether or not the image includes the specified feature. If the final result does not match the image label, parameter values for various of the classifiers or the arbitrator combining procedure can be adjusted. A trained arbitrated image classifier can then be used to determine whether new images include the particular feature.

Abstract: A social networking system provides access to personas comprising information, for example, web pages describing users or entities. The information may be suggested by the social networking system or requested by the user, for example, via search. The social networking system authenticates the personas so that only authentic personas are suggested to users or returned when a user is searching for information. The authenticity of a persona is determined based on the connections and/or likes coming from other personas, user accounts, or other entities represented within the social networking system that have been previously authenticated. The authenticity of the person is also determined based on external links to the persona, for example, external websites referring to the persona or the rate at which external systems such as search engines direct web traffic to the persona.

Abstract: Identifying whether an image includes a search image can be accomplished using region analysis and/or parts model analysis. Region analysis can include dividing a received image into regions and analyzing each region by: enlarging the region, applying a gradient histogram transformation to the enlarged region, and comparing the transformed region, at various angles, to a set of search patterns. Parts model analysis can include identifying key features of an image such as prominent lines or areas, objects or object edges, consecutive colors; generating one or more parts models from the key features by representing the key features as geometric objects; and comparing each parts model, at various angles, to a set of search patterns. The comparison employed in region analysis or parts model analysis can employ classifiers trained on the search images.

Abstract: A arbitrated image classifier can be trained to identify whether an image contains specified features, such as sexual, violent, or other potentially objectionable content. An arbitrated image classifier can include a configuration of classifiers and an arbitrator that determines a final image classification based on classification results from the classifiers. An arbitrated image classifier can be trained to identify image features by dividing images labeled as including or not including a specified feature into portions, which are provided to the classifiers of the arbitrated image classifier. The arbitrator of the arbitrated image classifier can determine a result for whether or not the image includes the specified feature. If the final result does not match the image label, parameter values for various of the classifiers or the arbitrator combining procedure can be adjusted. A trained arbitrated image classifier can then be used to determine whether new images include the particular feature.

Abstract: A social networking system provides access to personas comprising information, for example, web pages describing users or entities. The information may be suggested by the social networking system or requested by the user, for example, via search. The social networking system authenticates the personas so that only authentic personas are suggested to users or returned when a user is searching for information. The authenticity of a persona is determined based on the connections and/or likes coming from other personas, user accounts, or other entities represented within the social networking system that have been previously authenticated. The authenticity of the persona is also determined based on external links to the persona, for example, external websites referring to the persona or the rate at which external systems such as search engines direct web traffic to the persona.

Abstract: A social networking system provides access to personas comprising information, for example, web pages describing users or entities. The information may be suggested by the social networking system or requested by the user, for example, via search. The social networking system authenticates the personas so that only authentic personas are suggested to users or returned when a user is searching for information. The authenticity of a persona is determined based on the connections and/or likes coming from other personas, user accounts, or other entities represented within the social networking system that have been previously authenticated. The authenticity of the persona is also determined based on external links to the persona, for example, external websites referring to the persona or the rate at which external systems such as search engines direct web traffic to the persona.

Abstract: An online system allows content items to be targeted based on interests associated with users. When the online system receives a request to specify targeting criteria associated with a content item, the online system provides an interface to specify targeting criteria. As the online system receives input specifying an interest for inclusion in targeting criteria, the online system retrieves stored interests associated with online system users. Each interest stored by the online system is associated with a type. For example, a type associated with a stored interest indicates whether the interest is from a set of user-generated keywords, from a set of semantic topics mapped from the keywords, or from a set of manually curated broad categories. To avoid confusion from overlap in the types of interests, the online system applies rules to stored interests matching at least a portion of the input to select a set of interests.

Abstract: A composer interface allows a user to provide input, such as text data, to the social networking system. To reduce the number of inputs for a user to provide to the social networking system, one or more phrase objects are presented to the user when providing input. A phrase object includes a group of objects that users have previously provided to the social networking system. The social networking system presents phrase objects to the user that match at least a portion of the received input. Context information associated with the input and associated with various phrase objects may be used to select the presented phrase objects. By selecting a presented phrase object, the user includes the selected phrase object in the input.

Abstract: A social networking system receives advertisement requests from advertisers describing information about advertisements and determines one or more ad topics associated with the advertisements. When an advertisement is to be presented to a user, the social networking system determines one or more topics associated with the user from actions performed by the user and identifies candidate advertisements having ad topics matching, or similar to, the topics associated with the user. The topics associated with the user may be determined based on the user's most recent actions. One or more of the candidate advertisements are selected for presentation to the user.