Abstract: Described are techniques for adverse drug detection using ontology-augmented Large Language Models (LLM). In one aspect, a method is provided that includes accessing medical text associated with a patient. The medical text is annotated for drug and event entities to create labelled medical text, and spans in the labelled medical text are linked to concepts in an ontology. Possible paths in the ontology are identified based on the labeling and linking, each of the possible paths are ranked, and a top number of paths are identified based on the ranking. A verbalized form of each path of the top number of paths are then concatenated with associated portion(s) of the medical text to create ontology augmented text instances and a prompt having the ontology augmented text instance(s) are generated. Thereafter, one or more adverse drug reaction relation predictions are generated, by an LLM, for the patient based on the prompt.

Abstract: Techniques for generating negative samples using a taxonomy of entities and relations and for using the negative samples for machine learning are disclosed herein. Taxonomic negative samples are generated by selecting entities and negated relations for the entities using a taxonomy of entity types and subject, object relations. The system defines taxonomic negative samples for same-sentence relations, cross-sentence relations, and/or header context relations. The system sieves taxonomic negative samples are sieved to eliminate some samples, such as false negative samples. The sieved taxonomic negative samples are included with positive samples in training data used to train and/or fine-tune a prediction engine or other machine learning model.

Abstract: Techniques include accessing a set of documents; generating a final unified representation for each document of the set of documents, wherein generating the final unified representation comprises performing an iterative process for each document, and wherein the iterative process comprises: encoding, using a semantic embedding vector, a document's core semantic features in a semantic encoding, mapping a time domain for at least the document into a dimensional vector space to encode temporal information into a temporal encoding, and aggregating the semantic encoding and temporal encoding to generate the final unified representation; generating a query embedding, where the query embedding comprises a time-aware embedding for a query; comparing the query embedding to the final unified representation for each document of the set of documents; identifying one or more documents of the set of documents based on the comparing; and providing the one or more identified documents for downstream use.

Abstract: Methods and systems for automatically determining correspondences between communication ports of a networked device and encoders and decoders connected to those communication ports. In some embodiments, the networked device and the encoders and decoders are connected to a video communications network provided by a switch. The networked device can query the video communications network for information related to the encoders and decoders to determine and save the port-to-device correspondences. In some embodiments, the networked device can extract device information from video signals received at its input ports to map the input ports to respectively connected decoders. In similar fashion, the networked device may transmit or embed port-specific information from its output ports to respectively connected encoders. Then, the networked device can query the video communications network for the port-specific information received at the encoders to map the output ports to respectively connected encoders.

Abstract: Various embodiments of the present disclosure provide summarization techniques for summarizing complex documents, such as long unstructured call transcripts. The summarization techniques include generating a plurality of interaction topics for an interaction transcript and iteratively summarizing each interaction topic based on a preceding partial summary for the interaction transcript that corresponds to a preceding interaction topic that precedes the interaction topic in the interaction transcript. An abstractive summary is generated using a recursive abstractive model that is trained using training data generated based on holistic similarity scores between interaction topics of a call transcript and summary sentences of a corresponding target summary.

Abstract: Methods and systems for automatically determining correspondences between communication ports of a networked device and encoders and decoders connected to those communication ports. In some embodiments, the networked device and the encoders and decoders are connected to a video communications network provided by a switch. The networked device can query the video communications network for information related to the encoders and decoders to determine and save the port-to-device correspondences. In some embodiments, the networked device can extract device information from video signals received at its input ports to map the input ports to respectively connected decoders. In similar fashion, the networked device may transmit or embed port-specific information from its output ports to respectively connected encoders. Then, the networked device can query the video communications network for the port-specific information received at the encoders to map the output ports to respectively connected encoders.

Abstract: A summarization system (SS) is described that uses novel techniques to generate entity-aware summaries, where the novel techniques employ directional stimulus prompting and a large language model (LLM) to generate the summaries. In some embodiments, the SS receives as input the content to be summarized and a set of one or more entity categories corresponding to entities to be included in the generated summary. Hint information is generated based upon the content to be summarized. A prompt comprising the generated hint information and the content to be summarized is provided as input to a black-box LLM to generate an entity-aware summary. Novel techniques, including supervised fine-tuning and reinforcement learning are also described for training a language model generating the hint information.

Abstract: Various embodiments of the present disclosure provide summarization techniques for summarizing complex documents, such as long unstructured call transcripts. The summarization techniques include generating a plurality of interaction topics for an interaction transcript and iteratively summarizing each interaction topic based on a preceding partial summary for the interaction transcript that corresponds to a preceding interaction topic that precedes the interaction topic in the interaction transcript. An abstractive summary is generated using a recursive abstractive model that is trained using training data generated based on holistic similarity scores between interaction topics of a call transcript and summary sentences of a corresponding target summary.

Abstract: Method includes: accessing text, where spans are identified within the text and include one or more pairs of target spans and one or more mid-context spans; generating embedding representations of tokens associated with each target span, tokens associated with the entity types of each target span, and tokens associated with each mid-context span; generating, for each target span, entity-focused span embedding representation based on embedding representations of tokens associated with each target span and embedding representations of tokens associated with entity type of target span; generating, for each mid-context span, mid-context embedding representation based on the embedding representations of tokens associated with each mid-context span; and generating probability distribution of each relation of set of relations based on entity-focused span embedding representations of subject span and object span that are included in each target pair and mid-context embedding representation for mid-context span appear

Abstract: Various embodiments provide methods, apparatus, systems, computing entities, and/or the like, for providing a summarization of a conversation, such as a telephonic conversation. In an embodiment, a method is provided. The method comprises receiving an input data object comprising textual data of a conversation, the textual data comprising sentence-level tokens. The method further comprises classifying some sentence-level tokens as interrogative sentence-level tokens, and identifying subtopic portions of the textual data, each interrogative sentence-level token located within one subtopic portion. The method further comprises determining whether an interrogative sentence-level token is substantially similar to one of a plurality of target queries, and for such interrogative sentence-level tokens, selecting sentence-level tokens from a subtopic portion corresponding to the such interrogative sentence-level tokens.

Abstract: Disclosed are machine learning techniques directed to training a machine learning model for the combined learning of multiple natural language processing (NLP) tasks. The NLP tasks may be named entity recognition (NER), relation extraction (RE), and assertion detection (AD) tasks. The machine learning model may be a multi-layer transformer model. Training the machine learning model may involve first training the NER module on the NER task, and thereafter training the RE module on the RE task while the AD module is simultaneously trained on the AD task. Training the machine learning model may alternatively involve training the NER module on the NER task concurrently with training the RE module on the RE task and training the AD module on the AD task. The trained machine learning model can predict entities and entity types in newly provided text, along with relations between the entities and assertions associated with the entities.

Abstract: Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing natural language processing operations using an attention-based text encoder machine learning model that is trained using a multi-task training routine that is associated with two or more training tasks (e.g., a multi-task training routine that is associated with two or more sequential training tasks, a multi-training routine that is associated with two or more concurrent training tasks, and/or the like).

Abstract: Machine learning techniques directed to span prediction for textual data are disclosed. As used herein, span prediction is the process of predicting the possible spans of text that can be assigned to a given entity type of a set of predefined entity types. To this end, a machine learning model can be trained to generate values that indicate the predicted probability that a given span of an identified set of spans within text of interest is appropriate for association with a given entity type of the set of predefined entity types. The predicted probability values may be used to determine whether a given span or spans is associated with a given entity type. The predicted spans can also be scored in some examples.

Abstract: Various embodiments provide for summarization of an interaction, conversation, encounter, and/or the like in at least an abstractive manner. In one example embodiment, a method is provided. The method includes generating, using an encoder-decoder machine learning model, a party-agnostic representation data object for each utterance data object. The method further includes generating an attention graph data object to represent semantic and party-wise relationships between a plurality of utterance data objects. The method further includes modifying, using the attention graph data object, the party-agnostic representation data object for each utterance data object to form a party-wise representation data object for each utterance data object. The method further includes selecting a subset of party-wise representation data objects for each of a plurality of parties.

Abstract: Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing natural language processing operations using a hybrid reason code prediction machine learning framework. Certain embodiments of the present invention utilize systems, methods, and computer program products that perform natural language processing using a hybrid reason code prediction machine learning framework that comprises one or more of the following: (i) a hierarchical transformer machine learning model, (ii) an utterance prediction machine learning model, (iii) an attention distribution generation machine learning model, (iv) an utterance-code pair prediction machine learning model, and (v) a hybrid prediction machine learning model.

Abstract: Methods and systems for automatically determining correspondences between communication ports of a networked device and encoders and decoders connected to those communication ports. In some embodiments, the networked device and the encoders and decoders are connected to a video communications network provided by a switch. The networked device can query the video communications network for information related to the encoders and decoders to determine and save the port-to-device correspondences. In some embodiments, the networked device can extract device information from video signals received at its input ports to map the input ports to respectively connected decoders. In similar fashion, the networked device may transmit or embed port-specific information from its output ports to respectively connected encoders. Then, the networked device can query the video communications network for the port-specific information received at the encoders to map the output ports to respectively connected encoders.

Abstract: Various embodiments of the present disclosure provide summarization techniques for summarizing complex documents, such as long unstructured call transcripts. The summarization techniques include generating a plurality of interaction topics for an interaction transcript and iteratively summarizing each interaction topic based on a preceding partial summary for the interaction transcript that corresponds to a preceding interaction topic that precedes the interaction topic in the interaction transcript. An abstractive summary is generated using a recursive abstractive model that is trained using training data generated based on holistic similarity scores between interaction topics of a call transcript and summary sentences of a corresponding target summary.

Abstract: Various embodiments of the present disclosure provide summarization techniques for summarizing complex documents, such as long unstructured call transcripts. The summarization techniques include generating a plurality of interaction topics for an interaction transcript and iteratively summarizing each interaction topic based on a preceding partial summary for the interaction transcript that corresponds to a preceding interaction topic that precedes the interaction topic in the interaction transcript. An abstractive summary is generated using a recursive abstractive model that is trained using training data generated based on holistic similarity scores between interaction topics of a call transcript and summary sentences of a corresponding target summary.

Abstract: Various embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing natural language processing operations using an attention-based text encoder machine learning model that is trained using a multi-task training routine that is associated with two or more training tasks (e.g., a multi-task training routine that is associated with two or more sequential training tasks, a multi-training routine that is associated with two or more concurrent training tasks, and/or the like).

Abstract: There is a need for more effective and efficient text categorization. This need can be addressed by, for example, techniques for semantic text categorization. In one example, a method includes determining an input vector-based representation of an input document; processing the input vector-based representation using a trained supervised machine learning model to generate the categorization based at least in part on the input vector-based representation, wherein: (i) the trained supervised machine learning model has been trained using automatically-generated training data, and (ii) the automatically generated training data is generated by determining an inferred semantic label for each unlabeled training document of one or more unlabeled training documents; and performing one or more categorization-based actions based at least in part on the categorization, and (iii) the labels are described by one or more short documents/short texts.