Abstract: Systems and methods for guiding private artificial intelligence models with public solutions. A very large language model (VLLM) can be iteratively queried with an instruction code including public entities with associated public documents to generate public solutions. Rationale features can be extracted from the public solutions with the VLLM. The instruction code can be updated by combining an input query about public entities, the public solutions with the rationale features, text from reference chunks, and an input query about a single private entity, following a pre-determined template, to yield a private instruction code about a single private entity. The private instruction code can be answered with private large language models (PLLM) to obtain private answers for performing downstream tasks.

Abstract: Systems and methods for visual retrieval augmented generation for artificial intelligence models such as multimodal large language models. Associations between image and description pairs can be identified from an awareness dataset by finetuning a multi-modal large language model (MLLM) with the awareness dataset based on randomly chosen images added to each example from a relevant dataset. Visual distractions for image processing with the MLLM can be minimized by finetuning the MLLM with a focus dataset based on randomly chosen images added to each example from the relevant dataset. Visual hallucinations from the MLLM can be mitigated by finetuning the MLLM with a learning dataset based on related images having corresponding texts added to each example from the relevant dataset to utilize extracted information from associations between provided text from multiple images and a learning dataset.

Abstract: Methods and systems include generating a first question relating to supporting an input claim. A search is performed based on the first question to identify evidence relating to the input claim. An answer to the first question is generated based on the evidence. Additional questions are iteratively generated, with searches being performed based on the additional questions, and with answers to the additional questions being generated until a predetermined stop condition is reached. The input claim is classified by predicting a label based on evidence identified by the searches.

Abstract: A computer-implemented method for counting and extracting opinions in product reviews is provided. The method includes inputting a hypothesis opinion, a product name, and product reviews relating to a product, applying a decontextualization component to the product reviews by using the product name, applying the decontextualization component to the hypothesis opinion by using the product name, applying an entailment model to classify each sentence of the decontextualized product reviews against the decontextualized hypothesis opinion, and outputting one or more sentences classified as entailing the hypothesis opinion and a count of corresponding reviews.

Abstract: A method trains an inference model on two-hop NLI problems that include a first and second premise and a hypothesis, and further includes generating, by the model using hypothesis reduction, an explanation from an input premise and an input hypothesis, for an input single hop NLI problem. The learning step determines a distribution over extraction starting positions and lengths from within the first premise and hypothesis of a two-hop NLI problem. The learning step k extraction output slots with combinations of words from the first premise of the two-hop NLI problem and fills another extraction output slots with combinations of words from the hypothesis of the two-hop NLI problem. The learning step trains a sequence model by using the extraction output slots and the other extraction output slots together with the second premise as an input to a single-hop NLI classifier to output a label of the two-hop NLI problem.

Abstract: Methods and systems for generating text include sampling a plurality of sentences generated by a language model in response to a query. A sentence is selected from the plurality of sentences using a score that is based on token consistency between the plurality of sentences.

Abstract: Systems and methods for verifying complex sentences with artificial intelligence. Claim sentences can be filtered with source texts using a confirmation threshold, an unsupported threshold, and entailment probabilities computed by a natural language inference (NLI) classifier to obtain initial verification pairs. A trained imagination model can generate entailment outputs by employing initial verification pairs. A trained generalization model can generate generalized outputs by generalizing entailment outputs. Missing evidence generalizations can be chosen from sampled generalized outputs based on overlaps between sampled generalized outputs The NLI classifier can compute a final verification decision of the source texts against the missing evidence generalizations to obtain verified claim sentences. A corrective action for a monitored entity can be performed using the verified claim sentences.

Abstract: Methods and systems for answering a query include generating first tokens in response to an input query using a language model, the first tokens including a retrieval rule. A retrieval rule is used to search for information to generate dynamic tokens. The retrieval rule in the first tokens is replaced with the dynamic tokens to generate a dynamic partial response. Second tokens are generated in response to the input query. The second tokens are appended to the dynamic partial response to generate an output responsive to the input query.

Abstract: Methods and systems for language processing include augmenting an original training dataset to produce an augmented dataset that includes a first example that includes a first scrambled replacement for a first word and a definition of the first word, and a second example that includes a second scrambled replacement for the first word and a definition of an alternative to the first word. A neural network classifier is trained using the augmented dataset.

Abstract: A method for neural network training is provided. The method inputs a training set of textual claims, lists of evidence including gold evidence chains, and claim labels labelling the evidence with respect to the textual claims. The claim labels include refutes, supports, and not enough information (NEI). The method computes an initial set of document retrievals for each of the textual claims. The method also includes computing an initial set of page element retrievals including sentence retrievals from the initial set of document retrievals for each of the textual claims. The method creates, from the training set of textual claims, a Leave Out Training Set which includes input texts and target texts relating to the labels. The method trains a sequence-to-sequence neural network to generate new target texts from new input texts using the Leave Out Training Set.

Abstract: Systems and methods for autonomous generation of accurate healthcare summaries. Relevant healthcare questions can be predicted based on a preceding context by employing a fine-tuned transformer model. Answers to the relevant healthcare questions can be predicted by employing an extractive question answering model and utilizing extracted healthcare data from a healthcare data record to obtain predicted healthcare answers. Complete sentences can be synthesized, with artificial intelligence (AI), from the predicted healthcare answers and the relevant healthcare questions to obtain healthcare summary sentences. A healthcare technical report can be generated autonomously with AI from the healthcare summary sentences to assist with a decision making of a healthcare professional.

Abstract: Systems and methods are provided for analyzing and visualizing document corpuses based on user-defined semantic features, including initializing a Natural Language Inference (NLI) classification model pre-trained on a diverse linguistic dataset, analyzing a corpus of textual documents with semantic features described in natural language by a user. For each semantic feature, a classification process is executed using the NLI model to assess implication strength between sentences in the documents and the semantic feature, the classification process including a confidence scoring mechanism to quantify implication strength. Implication scores can be aggregated for each of the documents to form a composite semantic implication profile, and a dimensionality reduction technique, can be applied to the composite semantic implication profiles of each of the documents to generate a two-dimensional semantic space representation.

Abstract: Methods and systems for document summarization include splitting documents into sentences and sorting the sentences by a metric that promotes review opinion prevalence from the documents to generate a ranked list of sentences. Groups of sentences with similar embeddings are formed and a trained generalization encoder-decoder model is applied to output a common generalization of the sentences in each group. Sentences are added to a summary from the generalizations corresponding to the sentences in the ranked list, in rank-order, until a target summary length has been reached. An action is performed responsive to the summary.

Abstract: A computer-implemented method is provided for disentangled data generation. The method includes accessing, by a bidirectional Long Short-Term Memory (LSTM) with a multi-head attention mechanism, a dataset including a plurality of pairs each formed from a given one of a plurality of input text structures and given one of a plurality of style labels for the plurality of input text structures. The method further includes training the bidirectional LSTM as an encoder to disentangle a sequential text input into disentangled representations comprising a content embedding and a style embedding based on a subset of the dataset. The method also includes training a unidirectional LSTM as a decoder to generate a next text structure prediction for the sequential text input based on previously generated text structure information and a current word, from a disentangled representation with the content embedding and the style embedding.

Abstract: A computer-implemented method for counting and extracting opinions in product reviews is provided. The method includes inputting a hypothesis opinion, a product name, and product reviews relating to a product, applying a decontextualization component to the product reviews by using the product name, applying the decontextualization component to the hypothesis opinion by using the product name, applying an entailment model to classify each sentence of the decontextualized product reviews against the decontextualized hypothesis opinion, and outputting one or more sentences classified as entailing the hypothesis opinion and a count of corresponding reviews.

Abstract: Methods and systems for disentangled data generation include accessing a dataset including pairs, each formed from a given input text structure and a given style label for the input text structures. An encoder is trained to disentangle a sequential text input into disentangled representations, including a content embedding and a style embedding, based on a subset of the dataset, using an objective function that includes a regularization term that minimizes mutual information between the content embedding and the style embedding. A generator is trained to generate a text output that includes content from the style embedding, expressed in a style other than that represented by the style embedding of the text input.

Abstract: A method for neural network training is provided. The method inputs a training set of textual claims, lists of evidence including gold evidence chains, and claim labels labelling the evidence with respect to the textual claims. The claim labels include refutes, supports, and not enough information (NEI). The method computes an initial set of document retrievals for each of the textual claims. The method also includes computing an initial set of page element retrievals including sentence retrievals from the initial set of document retrievals for each of the textual claims. The method creates, from the training set of textual claims, a Leave Out Training Set which includes input texts and target texts relating to the labels. The method trains a sequence-to-sequence neural network to generate new target texts from new input texts using the Leave Out Training Set.

Abstract: Systems and methods for matching job descriptions with job applicants is provided. The method includes allocating each of one or more job applicants' curriculum vitae (CV) into sections; applying max pooled word embedding to each section of the job applicants' CVs; using concatenated max-pooling and average-pooling to compose the section embeddings into an applicant's CV representation; allocating each of one or more job position descriptions into specified sections; applying max pooled word embedding to each section of the job position descriptions; using concatenated max-pooling and average-pooling to compose the section embeddings into a job representation; calculating a cosine similarity between each of the job representations and each of the CV representations to perform job-to-applicant matching; and presenting an ordered list of the one or more job applicants or an ordered list of the one or more job position descriptions to a user.

Abstract: Systems and methods for opinion summarization are provided for extracting and counting frequent opinions. The method includes performing a frequency analysis on an inputted list of product reviews for a single item and an inputted corpus of reviews for a product category containing the single item to identify one or more frequent phrases; fine tuning a pretrained transformer model to produce a trained neural network claim generator model, and generating a trained neural network opposing claim generator model based on the trained neural network claim generator model. The method further includes generating a pair of opposing claims for each of the one or more frequent phrases, wherein a generated positive claim is entailed by the product reviews for the single item and a negative claim refutes the positive claim, and outputting a count of sentences entailing the positive claim and a count of sentences entailing the negative claim.

Abstract: Methods and systems for language processing include augmenting an original training dataset to produce an augmented dataset that includes a first example that includes a first scrambled replacement for a first word and a definition of the first word, and a second example that includes a second scrambled replacement for the first word and a definition of an alternative to the first word. A neural network classifier is trained using the augmented dataset.