Verizon Patent and Licensing Inc.

Miruna Jayakrishnasamy

Prakash Ranganathan

The present teaching relates to extracting insightful phrases from an input text. Independent contexts are first identified from the input text. With respect to each independent context, initial candidate phrases are generated with respect to linguistic features and are then filtered. Various features are then computed for each filtered candidate phrase and used to select top k candidate phrases for each independent context. A most insightful phrase is then selected from the k top candidate phrases using deep learned models. Such selected most insightful phrases for the independent contexts are then used for facilitating an understanding the input text.

Method and system for insightful phrase extraction from text

A device may receive text data, may extract phrases from the text data, and may calculate question scores for the phrases. The device may determine whether the phrases are associated with single answers or multiple answers, may calculate, single answer scores for the phrases associated with the single answers, and may calculate multiple answer scores for the phrases associated with the multiple answers. The device may extract, from the phrases, phrase data identifying grammar data, start tag data, and end tag data, and may process the phrase data, the single answer scores, and the multiple answer scores, with a machine learning model, to generate valid sentence scores for the text data. The device may identify, as a valid sentence, a group of phrases of the text data associated with a highest one of the valid sentence scores, and may add punctuation to the valid sentence to generate a final sentence.

Systems and methods for utilizing a machine learning model for sentence boundary detection

A device may generate first scores for sentences of text based on a cumulative frequency of words in each sentence, may generate second scores for the sentences based on a cumulative frequency of domain entities in each sentence, and may generate third scores for the sentences based on a sentiment analysis of each sentence. The device may generate a summary of the text, may filter the sentences to extract a first set of sentences, may filter the sentences to extract a second set of sentences, and may filter the sentences to extract a third set of sentences. The device may identify and assign weights to a first group of sentences, a second group of sentences, and a third group of sentences, may generate a ranked list of sentences based on the weighted first group, second group, and third group, and may perform actions based on the final summary.

Systems and methods for reducing input to and increasing processing speeds of natural language processing models

A device identifies, in multi-context text data, unrelated text and coreference text, and extracts coreference clusters, coreference sentences, and coreference sentiments based on the coreference text. The device extracts unrelated sentences from the unrelated text, and assigns tenses to the coreference sentences and the unrelated sentences. The device extracts phrases and entities from the coreference sentences and unrelated sentences, and assigns tense flags that exclude present tense sentences. The device selects past tense phrases and future tense phrases, and combines the past tense phrases and the future tense phrases to generate phrases. The device identifies invalid phrases in the phrases, and identifies similarities between the coreference sentences and the invalid phrases. The device processes the coreference text, the coreference tenses, the coreference sentiments, and the similarities, with a reinforcement learning model, to generate final context text.

Systems and methods for semantic separation of multiple intentions in text data using reinforcement learning

A device may receive dynamic customer data and static customer data, and may calculate additional customer data based on the dynamic customer data and the static customer data. The device may process the static customer data, the dynamic customer data, and the additional customer data, with a first machine learning model, to determine a next action prediction, and may process the static customer data, the dynamic customer data, and the additional customer data, with a second machine learning model, to determine a next sequence prediction. The device may concatenate the static customer data, the dynamic customer data, the additional customer data, the next action prediction, and the next sequence prediction to generate concatenated data, and may process the concatenated data, with a plurality of machine learning models, to calculate various outputs, and may generate a recommendation for the customer based on the various outputs.

Systems and methods for providing customer-behavior-based dynamic enhanced order conversion

The present teaching is directed to clustering with denoising capability and its use in network capacity planning. Data samples with attributes of network elements and respective key performance indicators are first clustered to obtain initial clusters. Each initial cluster is hierarchically clustered to generate subclusters, each of which is detected as a pure or an impure subcluster based on some criterion. Each impure subcluster is denoised based on a situation detected, with some samples merged with a corresponding pure subcluster, some bootstrapped using additional data samples with consistent properties, and some removed if additional data sample with consistent properties is not available. The denoising is iteratively performed until a denoising criterion is satisfied to obtain denoised clusters corresponding to clusters of network elements. Actions may be performed on the network elements according to their corresponding denoised clusters.

Growing community of inventors

Vellore, India

Miruna Jayakrishnasamy