Abstract: Systems and methods are used to detect underlying themes from a collection of documents at an aggregated level. A representative set of documents may be selected from a cluster of documents, with the representative set of documents corresponding to a general theme of the cluster. Candidate theme phrases may then be extracted from the documents and used to generate document embeddings and candidate phrase embeddings, which may be ranked, such as with a diversity-based ranking approach. Certain candidates may be selected from the ranking. Each of the documents forming the representative set may then be concatenated and a query embedding may be generated and ranked against the candidate phrases. In this manner, a collection of phrases associated with both the general underlying theme of the cluster, along with granular topics associated with that theme, may be identified.

Abstract: Clusters of users of networked services are defined based on tasks performed by such users during such networked services. Activities of the users during sessions of the networked services are tracked, and representations of such users or such activities are used to train a model to predict activities of users in the future, including but not limited to services utilized by such users, or pages visited by such users. Subsequently, when a user accesses a networked service during a session, activities of the user may be determined, and a representation of the session is provided as an input to the model, along with contextual information such as an identifier of the persona of the user. A next action, e.g., a service or a page utilized by the user, may be predicted based on outputs received from the model.

Abstract: The present disclosure relates to systems and methods for providing a network-based service infrastructure configuration for a plurality of network-based services. A configuration service may identify one or more network-based services and actions required for the services based on analyzing customer input. After processing the customer input, the configuration service may automatically configure the infrastructure configuration based on analyzing the customer input. The configuration service may identify and verify attributes required by each identified service and its associated property values. The configuration service may configure the infrastructure configuration by selecting a template from the plurality of templates stored in a datastore.

Abstract: In response to determining that a particular sequence of natural language input has been generated by a first entity participating in a multi-interaction dialog, a first representation of accumulated dialog state associated with the sequence is obtained from a machine learning model at an artificial intelligence service. Based on the first representation, a state response entry is selected from a collection of state response entries. The state response entry indicates a mapping between a second representation of accumulated dialog state, and a response recorded in a training example of the model. The recorded response is implemented.

Abstract: The present disclosure generally relates to a feedback processing service that can receive customer input, as customer feedback corresponds to a service context. The feedback processing service aggregates semantically similar feedback as a cluster. Then, the feedback processing service can prioritize each of the clusters by ranking each of the clusters.

Abstract: Techniques are described for a contextual natural language understanding (cNLU) framework that is able to incorporate contextual signals of variable history length to perform joint intent classification (IC) and slot labeling (SL) tasks. A user utterance provided by a user within a multi-turn chat dialog between the user and a conversational agent is received. The user utterance and contextual information associated with one or more previous turns of the multi-turn chat dialog is provided to a machine learning (ML) model. An intent classification and one or more slot labels for the user utterance are then obtained from the ML model. The cNLU framework described herein thus uses, in addition to a current utterance itself, various contextual signals as input to a model to generate IC and SL predictions for each utterance of a multi-turn chat dialog.

Abstract: Techniques for generating conversational training data are described. In some instances, a request to generate conversational training data for a goal-oriented conversation model is received, a transitional graph of intents is traversed to generate a conversation template for each intent of the transitional graph, each intent being a task to fulfill a request and comprising one or more slot to be filled by a user of the bot machine learning model, the conversation template including a path including at least one placeholder for an utterance or a slot level utterance, and at least utterances from one or more dictionaries are sampled to fill in the placeholders for the utterances of the path to generate conversational training data.

Abstract: A method, computer system, and a computer program product for determining the reliability of a claim is provided. The present invention may include receiving an input data from a user. The present invention may also include analyzing the claim associated with the received input data to determine a reliability score associated with the input data, wherein the claim is semantically similar to the received input data. The present invention may further include generating, from a prediction model, the reliability score for the claim associated with the received input data. The present invention may also include presenting the reliability score for the claim associated with the received input data to the user.

Abstract: A goal-oriented dialog system interacts with a user over one or more turns of dialog to determine a goal expressed by the user; the dialog system may then act to fulfill the goal by, for example, calling an application-programming interface. The user may supply dialog via text, speech, or other communication. The dialog system includes a first trained model, such as a translation model, to encode the dialog from the user into a context vector; a second trained model, such as another translation model, determines a plurality of candidate probabilities of items in a vocabulary. A language model determines responses to the user based on the input from the user, the context vector, and the plurality of candidate probabilities.

Abstract: A method of identifying location data in a data set comprises generating a data sample from the data set, training a plurality of models with the data sample to identify the location data in the data set, and applying the data set to the trained models to determine the location data within the data set. The plurality of models includes one or more first models to identify primary attributes of the location data indicating a geographical area and one or more second models to identify secondary attributes of the location data used to determine corresponding primary attributes.

Abstract: A method of identifying location data in a data set comprises generating a data sample from the data set, training a plurality of models with the data sample to identify the location data in the data set, and applying the data set to the trained models to determine the location data within the data set. The plurality of models includes one or more first models to identify primary attributes of the location data indicating a geographical area and one or more second models to identify secondary attributes of the location data used to determine corresponding primary attributes.

Abstract: Mechanisms are provided to implement a sentiment analysis mechanism for performing sentiment analysis of a medical event and a drug name within a medical document based on a medical context. The sentiment analysis mechanism analyzes a medical document to identify an occurrence of a medical event associated with a drug name and analyzes contextual content associated with the occurrence of the medical event and the drug name to identify one or more sentiment terms present in the contextual content. The sentiment analysis mechanism determines a sentiment associated with the medical event and drug name. The sentiment analysis mechanism generates medical clue metadata linking the sentiment with the medical event and the drug corresponding to the drug name and applies the medical clue metadata to analysis of other medical documents to identify sentiments associated with instances of the drug name or medical event in the other medical documents.

Abstract: Techniques for intelligent task-oriented multi-turn dialog system automation are described. A seq2seq ML model can be trained using a corpus of training data and a loss function that is based at least in part on a distance to a goal. The seq2seq ML model can be provided a user utterance as an input, and a vector of a plurality of values output by a plurality of hidden units of a decoder of the seq2seq ML model can be used to select one or more candidate responses to the user utterance via a nearest neighbor algorithm. In some embodiments, the specially adapted seq2seq ML model can be trained using unsupervised learning, and can be adapted to select intelligent, coherent agent responses that move a task-oriented dialog toward its completion.

Abstract: A mechanism is provided for implementing a medication interaction detection engine for automatically detecting interactions of medications from social media posts. Responsive to receiving an identification of a medication under consideration, a set of social media posts are searched to identify discussion forums. Responsive to identifying a medication, a medication probability for each topic in the discussion forums directed to the medication is generated. Responsive to identifying an adverse event, an adverse event probability for each topic in the discussion forums identified by the medication probability for each topic is generated. The adverse event probability for each topic is compared to the medication probability for each topic to identify an adverse event probability of occurrence for each medication.

Abstract: Merging synonymous entities from multiple structured sources into a dataset includes receiving a first set of paired terms from a first authoritative source for a domain and a second set of paired terms from a second authoritative source for the domain. The first set of paired terms is compared to the second set of paired terms with a similarity assessment based on a clustering statistical algorithm to identify paired terms from the first set of paired terms that share a synonymous term with one or more paired terms from the second set of paired terms. The paired terms associated with the synonymous term are merged and a dataset is generated that associates a normalized version of the synonymous term with any terms included in the merged paired terms.

Abstract: Relationship extraction between descriptors in one or more lists of weather condition descriptors, and adverse event descriptors within unstructured data sources using natural language processing. Medical condition descriptor may be a descriptor that may be used to further extract relationships between weather condition descriptors and adverse event descriptors. A data object is generated, according to a data model, based on the extracted relationships between the descriptors. A set of candidate unstructured documents containing the extracted relationship between the descriptors is retrieved and filtered by selecting unstructured documents that include a precautionary measure descriptor. The filtered precautionary measure descriptors are presented to a user in a summarized message to a user device.

Abstract: Relationship extraction between descriptors in one or more lists of weather condition descriptors, and adverse event descriptors within unstructured data sources using natural language processing. Medical condition descriptor may be a descriptor that may be used to further extract relationships between weather condition descriptors and adverse event descriptors. A data object is generated, according to a data model, based on the extracted relationships between the descriptors. A set of candidate unstructured documents containing the extracted relationship between the descriptors is retrieved and filtered by selecting unstructured documents that include a precautionary measure descriptor. The filtered precautionary measure descriptors are presented to a user in a summarized message to a user device.

Abstract: Relationship extraction between descriptors in one or more lists of weather condition descriptors, and adverse event descriptors within unstructured data sources using natural language processing. Medical condition descriptor may be a descriptor that may be used to further extract relationships between weather condition descriptors and adverse event descriptors. A data object is generated, according to a data model, based on the extracted relationships between the descriptors. A set of candidate unstructured documents containing the extracted relationship between the descriptors is retrieved and filtered by selecting unstructured documents that include a precautionary measure descriptor. The filtered precautionary measure descriptors are presented to a user in a summarized message to a user device.

Abstract: A mechanism is provided for automatically detecting and cleansing erroneous concepts in an aggregated knowledge base. A graph data structure representing the concept present in a portion of the natural language content is generated. The graph data structure is analyzed to determine whether or not the graph data structure comprises one or more concept conflicts in association with a set of nodes in the graph data structure, the one or more concept conflicts are associated with the set of nodes if two or more nodes represent separate and distinct concepts. Responsive to determining that there are one or more concept conflicts due to there being two or more nodes representing separate and distinct concepts, the two or more nodes are split into separate distinct concepts within the knowledge base.

Abstract: A method, computer system, and a computer program product for determining the reliability of a claim is provided. The present invention may include receiving an input data from a user. The present invention may also include analyzing the claim associated with the received input data to determine a reliability score associated with the input data, wherein the claim is semantically similar to the received input data. The present invention may further include generating, from a prediction model, the reliability score for the claim associated with the received input data. The present invention may also include presenting the reliability score for the claim associated with the received input data to the user.