Abstract: A text span forming either a single word or a series of two or more words that a user intended to select is predicted. A document and a location pointer that indicates a particular location in the document are received and input to different candidate text span generation methods. A ranked list of one or more scored candidate text spans is received from each of the different candidate text span generation methods. A machine-learned ensemble model is used to re-score each of the scored candidate text spans that is received from each of the different candidate text span generation methods. The ensemble model is trained using a machine learning method and features from a dataset of true intended user text span selections. A ranked list of re-scored candidate text spans is received from the ensemble model.

Abstract: Architecture that detects and corrects writing errors in a human language based on the utilization of three different stages: error detection, correction candidate generation, and correction candidate ranking. The architecture is a generic framework for generating fluent alternatives to non-grammatical word sequences in a written sample. Error detection is addressed by a suite of language model related scores and other scores such as parse scores that can identify a particularly unlikely sequence of words. Correction candidate generation is addressed by a lookup in a very large corpus of “correct” English that looks for alternative arrangements of the same or similar words or subsequences of these words in the same context. Correction candidate ranking is addressed by a language model ranker.

Abstract: A set of representations of item-page pairs of items and respective web pages that include the respective items is obtained, each representation including feature function values indicating weights associated with features of associated web pages, the features including page classification features. An annotated set of labeled training data that is annotated with salience annotation values of items for respective web pages that include the items is obtained. The salience annotation values are determined based on a soft function, by determining a first count of a total number of user queries associated with corresponding visits to the respective web pages, and determining a ratio of a second count to the first count, the second count determined as a cardinality of a subset of the corresponding visits that are associated with user queries that include the item, the subset included in the corresponding visits. Models are trained using the annotated set.

Abstract: An “Interestingness Modeler” uses deep neural networks to learn deep semantic models (DSM) of “interestingness.” The DSM, consisting of two branches of deep neural networks or their convolutional versions, identifies and predicts target documents that would interest users reading source documents. The learned model observes, identifies, and detects naturally occurring signals of interestingness in click transitions between source and target documents derived from web browser logs. Interestingness is modeled with deep neural networks that map source-target document pairs to feature vectors in a latent space, trained on document transitions in view of a “context” and optional “focus” of source and target documents. Network parameters are learned to minimize distances between source documents and their corresponding “interesting” targets in that space.

Abstract: Automatically summarizing electronic communication conversation threads is provided. Electronic mails, text messages, tasks, questions and answers, meeting requests, calendar items, and the like are processed via a combination of natural language processing and heuristics. For a given conversation thread, for example, an electronic mail thread associated with a given task, a text summary of the thread is generated to highlight the most important text in the thread. The text summary is presented to a user in a visual user interface to allow the user to quickly understand the significance or relevance of the thread.

Abstract: An “Engagement Predictor” provides various techniques for predicting whether things and concepts (i.e., “nuggets”) in content will be engaging or interesting to a user in arbitrary content being consumed by the user. More specifically, the Engagement Predictor provides a notion of interestingness, i.e., an interestingness score, of a nugget on a page that is grounded in observable behavior during content consumption. This interestingness score is determined by evaluating arbitrary documents using a learned transition model. Training of the transition model combines web browsing log data and latent semantic features in training data (i.e., source and destination documents) automatically derived by a Joint Topic Transition (JTT) Model. The interestingness scores are then used for highlighting one or more nuggets, inserting one or more hyperlinks relating to one or more nuggets, importing content relating to one or more nuggets, predicting user clicks, etc.

Abstract: A text span forming either a single word or a series of two or more words that a user intended to select is predicted. A document and a location pointer that indicates a particular location in the document are received and input to different candidate text span generation methods. A ranked list of one or more scored candidate text spans is received from each of the different candidate text span generation methods. A machine-learned ensemble model is used to re-score each of the scored candidate text spans that is received from each of the different candidate text span generation methods. The ensemble model is trained using a machine learning method and features from a dataset of true intended user text span selections. A ranked list of re-scored candidate text spans is received from the ensemble model.

Abstract: Affective state classification embodiments are described which train and use a classifier to identify an affect exhibited by a segment of text. The affect being identified is chosen from a group of affects, each of which corresponds to a different emotion or sentiment being expressed by a person authoring the segment of text. In addition, each affect in the group of affects relates more than the valence of the emotion or sentiment being expressed. In other word, the identified affect is more than just an indication of the positive or negative nature of the text segment. Rather, in one embodiment, the classifier is trained to identify whether a segment of text exhibits one of the following affects: fear, sadness, guilt, hostility, joviality, self-assurance, attentiveness, shyness, fatigue, surprise, and serenity.

Abstract: Computing functionality converts an input linguistic item into a normalized linguistic item, representing a normalized counterpart of the input linguistic item. In one environment, the input linguistic item corresponds to a complaint by a person receiving medical care, and the normalized linguistic item corresponds to a definitive and error-free version of that complaint. In operation, the computing functionality uses plural reference resources to expand the input linguistic item, creating an expanded linguistic item. The computing functionality then forms a graph based on candidate tokens that appear in the expanded linguistic item, and then finds a shortest path through the graph; that path corresponds to the normalized linguistic item. The computing functionality may use a statistical language model to assign weights to edges in the graph, and to determine whether the normalized linguistic incorporates two or more component linguistic items.

Abstract: A set of representations of item-page pairs of items and respective web pages that include the respective items is obtained, each representation including feature function values indicating weights associated with features of associated web pages, the features including page classification features. An annotated set of labeled training data that is annotated with salience annotation values of items for respective web pages that include the items is obtained. The salience annotation values are determined based on a soft function, by determining a first count of a total number of user queries associated with corresponding visits to the respective web pages, and determining a ratio of a second count to the first count, the second count determined as a cardinality of a subset of the corresponding visits that are associated with user queries that include the item, the subset included in the corresponding visits. Models are trained using the annotated set.

Abstract: Described is estimating whether an online search query is a news-related query, and if so, outputting news-related results in association with other search results returned in response to the query. The query is processed into features, including by accessing corpora that corresponds to relatively current events, e.g., recently crawled from news and blog articles. A corpus of static reference data, such as an online encyclopedia, may be used to help determine whether the query is less likely to be about current events. Features include frequency-related data and context-related data corresponding to frequency and context information maintained in the corpora. Additional features may be obtained by processing text of the query itself, e.g., “query-only” features.

Abstract: An overwhelming number of articles are available everyday via the internet. Unfortunately, it is impossible to peruse more than a handful, and it is difficult to ascertain an article's social context. The techniques disclosed herein address this problem by harnessing implicit and explicit contextual information from social media. By extracting text surrounding a hyperlink to an article in a post and assessing the article as a function of content surrounding the hyperlink, an article's social context is determined and presented. Additionally, articles that are sufficiently similar in content may be grouped to establish a many-to-one relationship between posts and an article, creating a more accurate assessment.

Abstract: A search system that automatically generates questions to refine an underspecified query. The system may generate questions even for queries against a database that contains unstructured textual descriptions of items, allowing the system to operate on a database of items that can be constructed inexpensively. The system extracts from the unstructured text combinations of words that may serve as a set of attribute values. The system uses a classifier to filter out attribute values from the set that would generate unanswerable questions. The remaining attribute values are ranked on their ability to narrow the search results and the highest ranking attribute value is used to generate a question to the user who submitted the query. The response to the question narrows the search results, and the process can be repeated iteratively until the search results are sufficiently narrow.

Abstract: Different advantageous embodiments provide for response prediction. A social element is received by a prediction mechanism. A feature set is generated for the social element. A prediction is generated using the feature set and a prediction model.

Abstract: In one embodiment, a web service engine server 104 may predict a successive action by a user based on an entity reference 302. The web service engine server 104 may identify an entity reference 302 in a data transmission caused by a user. The web service engine server 104 may determine from the data transmission a user intention towards the entity reference 302 using an intention model based on a transmission log. The web service engine server 104 may predict a related successive web action option 522 for the entity reference 302 based on the user intention.

Abstract: Functionality is described herein for converting an input linguistic item into a normalized linguistic item, representing a normalized counterpart of the input linguistic item. In one environment, the input linguistic item corresponds to a complaint by a person receiving medical care, and the normalized linguistic item corresponds to a definitive and error-free version of that complaint. In operation, the functionality uses plural reference resources to expand the input linguistic item, creating an expanded linguistic item. The functionality then forms a graph based on candidate tokens that appear in the expanded linguistic item, and then finds a shortest path through the graph; that path corresponds to the normalized linguistic item. The functionality may use a statistical language model to assign weights to edges in the graph, and to determine whether the normalized linguistic incorporates two or more component linguistic items.

Abstract: Automatically detected and identified tasks and calendar items from electronic communications may be populated into one or more tasks applications and calendaring applications. Text content retrieved from one or more electronic communications may be extracted and parsed for determining whether keywords or terms contained in the parsed text may lead to a classification of the text content or part of the text content as a task. Identified tasks may be automatically populated into a tasks application. Similarly, text content from such sources may be parsed for keywords and terms that may be identified as indicating calendar items, for example, meeting requests. Identified calendar items may be automatically populated into a calendar application as a calendar entry.

Abstract: Automatically summarizing electronic communication conversation threads is provided. Electronic mails, text messages, tasks, questions and answers, meeting requests, calendar items, and the like are processed via a combination of natural language processing and heuristics. For a given conversation thread, for example, an electronic mail thread associated with a given task, a text summary of the thread is generated to highlight the most important text in the thread. The text summary is presented to a user in a visual user interface to allow the user to quickly understand the significance or relevance of the thread.

Abstract: A summarization system and method. The summarization method includes utilizing a first body of information to obtain a second body of information, which is identified (by a hyperlink, an attachment identifier, a reference, etc.) in the first body of information. A summary of the obtained second body of information is then computed. The computed summary can be displayed to a user and/or stored for later use.

Abstract: A social newsfeed being delivered to a user is triaged. A personalized model is established which predicts the importance to the user of data elements within a current social newsfeed being delivered to the user. The personalized model is established based on implicit actions the user takes in response to receiving previous social newsfeeds. The personalized model is then used to triage the data elements within the current social newsfeed.