Abstract: In one embodiment, indexing content in streamed data includes receiving streams of audio data encoding a recording of a live ongoing group communication, where each stream of audio data encodes a different one of multiple voices. Each of the streams of audio data is provided to a recognizer to cause separate recognition of words in each of the streams. The recognized words are indexed to corresponding locations in each of the streams, and the streams are combined into a combined stream of audio data by synchronizing at least one common location in the streams. Embodiments allow accurate recognition of speech in group communications in which multiple speakers have simultaneously spoken, and accurate search of content encoded and processed from such speech.

Abstract: A computer-implemented technique includes determining, at a computing device including one or more processors, one or more scripts in which a user is capable of inputting text. The technique includes determining, at the computing device, whether at least one of the one or more scripts is a script having a right-to-left (RTL) writing directionality. The technique also includes automatically outputting, at the computing device: (i) a first user interface when at least one of the one or more scripts is a script having an RTL writing directionality, wherein the first user interface is configured to allow the user to adjust the writing directionality at the computing device, or (ii) a second user interface when none of the one or more scripts is a script having an RTL writing directionality, wherein the second user interface is not configured to allow the user to adjust the writing directionality at the computing device.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium and a method for automatically providing support solutions in response to user feedback items. The method comprises receiving user feedback items and corresponding support solutions. The method further comprises identifying, using clustering techniques, associations between the user feedback items and the corresponding support solutions. The method further comprises storing the identified associations as an items-solutions model that correlates the user feedback items with the corresponding support solutions. The method further comprises receiving a new user feedback item. The method further comprises automatically determining, using the items-solutions model, at least one support solution that corresponds to the new user feedback item. The method further comprises providing the at least one support solution in response to the received new user feedback item.

Abstract: Methods, systems, and apparatuses for determining a font substitution hierarchy using unsupervised clustering techniques are provided. A font substitution hierarchy may be determined by identifying a plurality of font request files associated with web-based application documents. Determining an association between a plurality of font requests within the plurality of font request files based on a clustering analysis, and determining a font substitution hierarchy for an individual font request based on the association.

Abstract: Social networking applications may be improved by incorporating a user's language proficiency to make content suggestions to the user. A language preference of a user, which may represent one of a plurality of signals, may be received. A signal may be, for example, an online activity of the user, a text generated or received by the user, or content requested by the user. At least one of the plurality of signals may be analyzed using a machine learning program. A machine learning program may be trained on data for a test group of users with a known language proficiency. A user-assigned language proficiency may be incorporated as a signal in training a machine learning program. The language proficiency of the user may be determined based upon the analysis of at least one of the plurality of signals. Content may be presented to the user based upon the language proficiency of the user.

Abstract: Social networking applications may provide an enhanced user experience by considering information regarding the language preference of a user as well as other signals. According to the implementations disclosed herein, one or more signals may be utilized to determine the language in which content may be displayed to a user of a social networking application or website. A first language preference for a user may be determined based on at least one signal. A second language preference may be determined for a social graph component of the social networking application based on at least one signal. The first language preference may be compared to the second language preference and a weighted value may be assigned to the social graph component. The social graph component of the social networking application may be presented based upon the weighted value assigned to the social graph component.

Abstract: Techniques for determining one or more preferred languages for a user are provided. The preferred languages may be determined based upon a set of language indicators. The language indicators are analyzed using, for example, rules-based techniques, clustering, language classifiers, and the like, or combinations thereof. Language indicators can include or be derived from information about the user's behavior, location, preferences, social connections, or other data related to the user.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium and a method for automatically providing support solutions in response to user feedback items. The method comprises receiving user feedback items and corresponding support solutions. The method further comprises identifying, using clustering techniques, associations between the user feedback items and the corresponding support solutions. The method further comprises storing the identified associations as an items-solutions model that correlates the user feedback items with the corresponding support solutions. The method further comprises receiving a new user feedback item. The method further comprises automatically determining, using the items-solutions model, at least one support solution that corresponds to the new user feedback item. The method further comprises providing the at least one support solution in response to the received new user feedback item.

Abstract: Techniques for determining one or more preferred languages for a user are provided. The preferred languages may be determined based upon a set of language indicators. The language indicators are analyzed using, for example, rules-based techniques, clustering, language classifiers, and the like, or combinations thereof. Language indicators can include or be derived from information about the user's behavior, location, preferences, social connections, or other data related to the user.

Abstract: A process for a language selection using one or more language preferences is disclosed. Information for presenting a language preference selection control may be sent to a user. A language preference may be received from the user and stored to a server. A cookie, which includes an indication of the language preference, may be caused to be stored to a device of a user. The stored language preference may be accessed from the cookie responsive to a communication from the user. Responsive to whether the user has logged into an account, the language preference may be retrieved from either the server or from the cookie. The retrieved language preference may be compared with languages available in an application and a most preferred language for the application may be determined based upon the comparison. Information from the application may be provided to the user in the most preferred language.

Abstract: Embodiments of the method include receiving, from a client device, a request for products similar to a target product. The request specifies a similarity channel for the target product that indicates one or more themes. Thereafter, a set of products similar to the target product is identified, where each of the products in the set shares in common with the target product at least one of the themes indicated by the similarity channel. After identifying the set of similar products, the set is ranked according to a similarity of each product in the set to the target product. Subsequently, the set of ranked products is provided to the client device.

Abstract: Applications can be localized by a localization expert to allow them to be used by a broader customer base. The localization can be conducted by a localization expert who is provided with a resource file containing localizable components. The resource file can be generated by providing a developer with one or more text strings based on the development code of the application and receiving the localizable components which are selected from with the text strings based on the development code.

Abstract: A technique can include identifying a collection of documents to be clustered. The collection of documents can include foreign language documents and base language documents. The foreign language documents can be translated into the base language at a base language translation module. Keywords in the base language documents and keywords in the translated foreign language documents can be determined at a document indexing module. The base language documents can be clustered with the foreign language documents in a common set of document clusters based on the determined keywords in the base language documents and the determined keywords in the translated foreign language documents. In response to a search query in a first language, a listing of search results can be provided that includes documents in the first language and another language from the a common document cluster.

Abstract: Performance of the machine learning technique is assessed using Bayesian analysis where previously grouped documents belonging to a machine-assigned class or cluster are presented to a human rater and the rater's assessment is fed to the Bayesian analysis processor that updates a Beta bionomial model with each document. The model represents the precision probability associated with the class or cluster under test. Monitoring the precision probability, the technique enforces a set of stopping rules corresponding to an acceptance/rejection assessment of the machine learning apparatus. A Markov Chain Monte Carlo process operates on the model to infuse the processing of each subsequent class or cluster with knowledge from those previously processed.

Abstract: A technique can include identifying a collection of documents to be clustered. The collection of documents can include foreign language documents and base language documents. The foreign language documents can be translated into the base language at a base language translation module. Keywords in the base language documents and keywords in the translated foreign language documents can be determined at a document indexing module. The base language documents can be clustered with the foreign language documents in a common set of document clusters based on the determined keywords in the base language documents and the determined keywords in the translated foreign language documents. In response to a search query in a first language, a listing of search results can be provided that includes documents in the first language and another language from the a common document cluster.

Abstract: Among other disclosed subject matter, a computer-implemented method includes receiving a plurality of electronic documents associated with a domain at a server. Each of the plurality of electronic documents includes meta-data and textual content. The method includes identifying one or more text strings in the textual content that are to be processed differently than an identical or similar text string in other electronic documents, and associating, with the electronic document, data indicating that each of the identified text strings is to be processed differently than an identical or similar text string in other electronic documents. The method also includes performing an analysis of the electronic documents to identify one or more subsets of the electronic documents that include related subject matter. A plurality of degrees of relatedness can be associated with text strings associated with data indicating that each of the text strings is to be processed differently.

Abstract: An ordered list of language indicators may be obtained, wherein each of the language indicators is either user defined or system defined. Each language indicator may be ranked according to a priority of the language indicator for determining a most-preferred language for a user. A list of languages may be received that indicates the languages available in an application. The ordered list of language indicators may then be compared to the received languages, and a highest-ranked language indicator determined. The most-preferred language for the user may be selected based upon the highest-ranked language indicator. Information from the application may then be provided to the user in the most preferred language.

Abstract: Information about named entities referenced in an electronic book (ebook) is provided to a client device. An ebook identifier identifying the ebook is received from the client device. A set of layers available for use with the ebook is determined. The layers in the set provide information associated with the ebook and a layer in the set provides information associated with named entities referenced in content of the ebook. A content range identifying a range of content of the ebook for which layer information is requested and an identification of one or more of the layers in the set for which layer information is requested is received from the client device. Layer information associated with the ebook content identified by the content range for the identified layers is transmitted to the client device. The transmitted layer information includes information associated with named entities referenced by ebook content.

Abstract: A computer-implemented method includes receiving, by one or more computer systems, first information from a first channel and second information from a second channel; merging the first information with the second information; applying an unsupervised clustering model to the merged information; and generating, based on results of the applying, a cross-channel cluster, the cross-channel cluster including (i) a portion of the first information associated with a subject matter, and (ii) a portion of the second information associated with the subject matter.

Abstract: Among other disclosed subject matter, a computer-implemented method that includes receiving a set of clusters of documents and calculating a similarity score for each cluster wherein the similarity score is based at least in part on features included in the documents in the cluster and indicates a measure of similarity of the documents in the cluster. For each cluster associated with a respective similarity score greater than a first threshold, identifying the cluster as satisfying a quality assurance requirement. For each cluster associated with a respective similarity score less than a second threshold, identifying the cluster as failing the quality assurance requirement. For each cluster associated with a similarity score less than or equal to the first threshold value and greater than or equal to the second threshold value, reviewing at least a subset of documents in the cluster to determine whether the cluster satisfies the quality assurance requirement.