565results about "Text database querying" patented technology

The invention discloses a method for analyzing a matching degree between a demand and an output result based on text semantics. The method comprises the following steps: step 1, labeling a data set; step 2, technical document preprocessing; 3, training and predicting a single-parameter model; 4, integrating prediction results of the multi-parameter model; the method has the beneficial effects thatthe method is simple; deep learning and the NLP technology are applied to the field of project association degree calculation of enterprise project management for the first time. Calculating an association matching degree between the two projects according to project requirements and result description; the associated project positioning difficulty is effectively reduced; meanwhile, the demand side can be helped to quickly and efficiently locate high-quality projects adapting to the demand of the demand side; time and resource investment for achievement screening and matching are greatly reduced, the association matching degree between projects is calculated by means of text data of existing project achievement technical documents and project declaration guidelines, and then large enterprises are assisted in screening high-quality projects with the high matching degree in the project bidding and tendering link.

A system and method for executing computer readable code instructions for an on-the-box (OTB) artificial intelligence (AI) productivity tool comprising a hardware processor accessing capabilities associated with each of a plurality of AI productivity tool-enablable software applications, a natural language capabilities database memory to store natural language descriptions of the capabilities and capability intent values generated from the natural language descriptions in a capabilities decision tree with each capability node grouped under a branch of the capabilities decision tree according to logical topics in hierarchical parent-child relationships, the hardware processor generating a query input intent value from a user query input and performing a cosine semantic similarity search comparing the capability intent values of the capability nodes along the branch of the capabilities decision tree for identifying a best match capability node having a highest cosine semantic similarity search score, and the hardware processor executing the best match capability.

The invention discloses an enterprise knowledge base retrieval and intelligent answering method and system based on a large language model. The method comprises the following steps: performing clause-level segmentation on an enterprise knowledge base document, associating document metadata to form structured knowledge entries, and establishing a keyword reverse index and a semantic vector index for the structured knowledge entries; analyzing the natural language query of the user, and performing multi-strategy expansion to generate an enhanced query expression and a query semantic vector; performing dual-channel mixed retrieval, performing duplicate removal, version filtering and weighted fusion sorting on a result, and generating a final candidate knowledge item list; and based on the candidate list and a predefined instruction, calling a large language model to generate a structured answer with complete traceability information. The method is compatible with an existing retrieval framework, precise understanding, knowledge point-level positioning, cross-document content integration and version consistency control of natural language problems are achieved, and the retrieval accuracy, answer availability and service intelligence level of an enterprise knowledge base are remarkably improved.

Provided are system, apparatus, device, method and / or computer-program product embodiments, combinations and / or sub-combinations thereof for using an AI model to facilitate natural language interactions with databases. An example method can include receiving a natural language prompt associated with a user and identifying tables in a database based on the natural language prompt. The method can further include determining a table schema(s) of each of the tables identified, generating, using a large language model, a query to the tables in the database based on the natural language prompt and the table schema(s), and obtaining, using the query, data from at least one table of the tables in the database. The method can include generating, using the large language model or another large language model, a response to the natural language prompt based on the data obtained from the at least one table of the tables in the database.

Techniques for determining whether retrieved content is sufficient for a language model to generate a response to an input are described. In some embodiments, a system may determine a retrieval complexity (RC) metric based on a question, a reference answer and document results retrieved from a retrieval component (e.g., a search engine, a knowledge search, etc.). The RC metric may be based on whether one of the document results corresponds to the reference answer and whether the reference answer can be determined from the set of the document results (e.g., using two or more document results). The RC metric can be used to determine when retrieved content is to be used by a language model for responding to an input corresponding to the question.

Service request generation using generative artificial intelligence is disclosed. For instance, a user query comprising character strings is received; the character strings are converted to initial vector values associated with a first continuous vector space. Groups of data representing knowledge base articles that are identified. The knowledge base articles are retrieved from first database equipment. The text content of the knowledge base articles is converted to further vector values associated with a second continuous vector space. A listing of support agent identities is determined based on the further vector values being proximate to additional vector values associated with the listing of support agent identities. The additional vector values are used to retrieve the listing of support agent identities from second database equipment and generate and transmit a service ticket to resolve the user query.

A composite AI system and method for advanced reasoning and automation that integrates symbolic knowledge graphs and algorithms with non-symbolic, or connectionist, models such as neural embeddings. A hierarchical architecture enables dynamically distributed, cooperative reasoning through layperson and expert-led challenge-based verification, model blending, model fitness and retraining and selection, comprehensive feedback loops at individual model or model blend or process flow with or without supervision, and specialized routing of processing to account for various operational risk, regulatory, legal, privacy, or economic considerations. Models, datasets, knowledge bases, simulations and simulation components, and embeddings are iteratively refined using knowledge graph elements and model, process, simulation or flow / process optimal hyperparameters which are recorded and tracked. Extraction of symbolic representations from connectionist models links them to curated ontologies of facts and principles.

A system and method for matching multimodal user-query input at an information handling system includes storing capabilities associated with a plurality of AI productivity tool-enablable software applications in a hierarchical capabilities decision tree with each node including natural language textual and non-textual modality descriptions of a capability and multimodal capability intent values generated from the same. Executing code instruction to receive a multimodal user-query input in any of text, audio, or image and generate a multimodal query input intent value for matching to a best match capability for a responsive action to be taken by one of the plurality of AI productivity tool-enablable software applications executing on the information handling system via a semantic similarity search comparing the multimodal query input intent value to the multimodal capability intent values in the hierarchical capabilities decision tree based on a highest cosine semantic similarity search score.

Various embodiments of the present disclosure provide prompt engineering and text quality assessment techniques for improving generative text outputs. The techniques include identifying a training cluster for an input document, generating a candidate prompt for a generative machine learning model based on the training cluster and a prompt template, providing the candidate prompt to the generative machine learning model to receive at least a portion of a candidate document, generating a plurality of quality metrics for the candidate prompt based on the candidate document, and selecting the candidate prompt from a plurality of candidate prompts based on the plurality of quality metrics.

A method includes obtaining data source tokens that identify corresponding data sources, and adding the data source tokens to a vocabulary list of a response tokenizer. An internal mapping of the response tokenizer is updated with the data source tokens, mapping each data source token to a corresponding token identifier (ID). Vector representations corresponding to the data source tokens are added to a response embedding model. The response tokenizer and the response embedding model are included in a response foundation model. An annotated dataset including multiple instances of an input utterances, retrieved data, and a data source token is created. The retrieved data is retrieved from a data source corresponding to the data source token. The response foundation model is trained with the annotated dataset to output data source tokens for a new input utterance based on a context of the new input utterance.

In some implementations, the techniques described herein relate to a method including: receiving, by a processor, a natural language input; retrieving, by the processor, a plurality of semantically relevant results based on the natural language input; classifying, by the processor, an intent of the natural language input using a first machine learning model; selecting, by the processor, a second machine learning model based on the intent; generating, by the processor, a prompt based on a type of the second machine learning model using the natural language input and the plurality of semantically relevant results. inputting, by the processor, to prompt into the second machine learning model; obtaining, by the processor, a result responsive to the prompt from the second machine learning model; and providing, by the processor, the result to the user.

The invention discloses a dynamic updating law and regulation knowledge graph construction and recall method. The method comprises the following steps: collecting law and regulation text data in real time; generating a structured law article object sequence through regular fragmentation processing; performing full-text structure recognition and semantic processing to generate segmentation-level and full-text-level semantic data; carrying out composite vectorization on the content of the law article, the outline of the segmented text and the abstract to generate a context enhanced semantic vector; fusing the full-text and segmented triads and enhancing relation attributes; respectively storing the vector and the original data to a vector database and an original index; importing the fused triple into a graph database to construct a knowledge graph; and based on the map response query, outputting a structured answer through entity recognition, semantic extension, multi-hop query and result fusion. According to the method, efficient, accurate and real-time retrieval and dynamic updating of law and regulation knowledge are achieved, and the recall rate and accuracy of complex legal questions and answers are effectively improved.

System, method, and various embodiments for a functional code generation system leveraging LLM capabilities, are described herein. An embodiment operates by receiving a user instruction to generate a report. An information prompt is generated, and an large language model (LLM) derives a plurality of vectors from the instruction, including a first vector. A first question is identified based on comparing the first vector to a plurality of questions; the first question is associated with a first query template. A query result from a collection database based on executing a query corresponding to the first query template against the collection database. The LLM is instructed to generate an answer comprising a natural language interpretation of the query result in view of the first question. The LLM is instructed to generate a report based on the answer in view of the user instruction.

Certain aspects of the disclosure provide for information retrieval that exploits context derived from document structure. Source documents can be preprocessed to identify fields and determine context attributes related to each field based on the structural layout of a source document. Resource documents can also be preprocessed to segment a resource document into passages and determine context related to the passages based on structural layout. Queries pertaining to a field can be enhanced by adding context metadata associated with the field. A query embedding can be generated and compared with previously generated passage embeddings to locate candidate matches based on similarity. A machine learning model can be provided with the top-ranked passages and tasked with re-ranking the passages based on relevancy to the original query. The highest re-ranked passage or set of passages can be output in response to the query.

Methods and systems for automatic detection and filling of information gaps in a reference database are described. Responsive to a user query in an ongoing chat session, a query embedding associated with the user query is obtained. A synthetic question embedding is identified from a vector database, based on a similarity to the query embedding. Responsive to determining that the similarity between the synthetic question embedding and the query embedding does not meet a similarity threshold, the ongoing chat session is monitored to detect an answer to the user query. A prompt is provided to a large language model (LLM) to generate and display a textual content corresponding to the user query, based on the detected answer, for automatically updating the reference database. The disclosed methods and systems effectively incorporate new or undocumented information that is not currently captured within the reference database, as gaps are identified.

In one embodiment, a method for managing embeddings and text for enhanced natural language understanding includes dividing, by a process, a corpus of one or more documents into a first plurality of fragments based on a first threshold size for a first large language model and computing, by the process, a first set of embeddings using the first large language model to analyze the first plurality of fragments. The method further includes dividing, by the process, the corpus of one or more documents into a second plurality of fragments based on a second threshold size for a second large language model and computing, by the process, a second set of embeddings using the second large language model to analyze the second plurality of fragments.

Techniques and mechanisms for fine-tuning a language model to be optimized for a network device to which the language model is deployed. A controller for a network may maintain an inventory of network devices in a network, and obtain device information for the network devices. The controller may analyze the device information to determine a device type or role for the network devices. The controller may then select a pre-trained model that is optimal or well-suited for a device type of a particular network device, and perform a distillation function of the language model. Once the language model has been distilled, the controller may augment the language model with locally relevant information such that the language model is contextually relevant for the network device. After fine-tuning the language model, the controller pre-positions the language model on the device so network administrators and other users can access it when necessary.

Disclosed herein are system, apparatus, device, method and / or computer program product aspects, and / or combinations and sub-combinations thereof, for using large language models and text embedding models to generate trivia questions. An example aspect operates by a computer-implemented method. The method includes receiving, by at least one computer processor, a category for a plurality of trivia questions and receiving a difficulty level for the plurality of trivia questions. The method further includes using the category to retrieve additional information for the plurality of trivia questions and generating the plurality of trivia questions based at least on the additional information and the difficulty level. The method further includes displaying the plurality of trivia questions on a display device.

A system and method for improving generative artificial intelligence (AI) software application response is provided. The method includes: receiving a query directed to a generative AI software application; receiving a response to the query, the response generated by the generative AI software application; generating a first contextual value based on the received query; generating a second contextual value based on the received response; generating a verification score based on the first contextual value and the second contextual value; and initiating a mitigation action in response to detecting that the verification score is below a predetermined threshold

A secure query method for external LLM service is provided. The secure query method comprises receiving an AI query. The secure query method also comprises processing private data stored in a memory of the edge device to obtain a private prompt according to the AI query. The secure query method also comprises transmitting the AI query to a second language generation model. The secure query method also comprises receiving an initial response generated by the second language generation model according to the AI query. The secure query method also comprises inputting the initial response and the private prompt to a first language generation model of the edge device, for obtaining a final response generated by the first language generation model of the edge device.

The present disclosure of the various embodiments relates to using a large language model to assistant with the creation of secure code and / or the completion of threat modeling tasks in software development. In one example, a system comprises a computing device configure to identify a prompt that requests generating secure source code for source code with a security vulnerability. A security data source is queried for a security threat embedding. The security threat embedding is received from the security data source and an augmented prompt is generated. The augmented prompt is transmitted to the large language model. A secure source code is received from the large language model and imported into application source code in a software development environment.

A system and method are provided for performing keyword-assisted semantic searching.

A computer-implemented system is disclosed which enhances legal research by utilizing a first user interface (UI) to receive search criteria as text prompts from users, which are then processed by an application program interface (API) across an electronic connection. The system includes an instruction base for expanding the prompts into snippets for a machine learning network, which defines the legal database searches. The process further refines the prompts through a transformation module, breaking down and tagging the text for compatibility with the machine learning network. Coupled with non-transitory computer memory, the system efficiently outputs results in response to the applied search criteria. A second interactive UI displays these results and allows for user interaction to delve into detailed information, optimizing the legal information retrieval process. Also disclosed is a method for processing legal research queries enhanced by a machine learning network.

Systems and methods are provided for creation of project meshes by defining agentic tool instances that can be activated to access agentic tools. Examples include providing an interface for defining a mesh of agentic tools and receiving, via the interface, queries descriptive of a project. The queries can be provided to a reasoning engine that generates first parameters defining tasks and second parameters defining agentic tool instances. The agentic tool instances can be created in accordance with the first parameters and the second parameters and activated for accessing agentic tools located in a data store. The agentic tools can correspond to a parameter of the first parameters associated an agentic tool instance. Examples can configure the mesh comprising the agentic tool instances, and the mesh can be executed to cause the agentic tool instances to access and run one or more of the agentic tools.

The embodiment of the invention provides a low-rank cache enhancement generation method and device based on a large language model, and relates to the technical field of large model enhancement. The method comprises the following steps: obtaining knowledge embedding and user query; according to the knowledge embedding, constructing a low-rank key value cache matrix; determining a query projection parameter and an output projection parameter according to the low-rank key value cache matrix; inputting the user query into a preset large language model, splicing context caches corresponding to the low-rank key value cache matrix, and forming and generating cue words; and based on the query projection parameter, the output projection parameter and the low-rank key value cache matrix, performing reasoning generation on the generated cue word, and determining an output result. According to the scheme, the problems that in an existing large language model knowledge enhancement scheme, the process is complex, the retrieval precision is insufficient, and the cache context storage capacity is limited are solved.

One or more computing devices and / or methods for retrieval augmented multiple choice question and answer generation on search queries are provided. A pregeneration prompt may be generated based upon a user input query, one or more chunks of content, and / or instructions for a model. The pregeneration prompt is input into the model to generate an initial question. Another prompt is generated based upon the initial question, the one or more chunks of content, and / or the instructions. The prompt is input into the model to generate question and answer content in a multiple choice format that is provided through a user interface for user engagement.

A method includes receiving, by one or more processors, a natural language query, executing, by the one or more processors, a first large language model (LLM) using as input the natural language query to generate a preliminary response to the natural language query, executing, by the one or more processors, a machine learning model using as input the preliminary response to generate a preliminary response embedding, querying, by the one or more processors, a vector database using the preliminary response embedding to retrieve contextual data for the natural language query, and executing, by the one or more processors, a second LLM using as input the natural language query and the contextual data to generate a response to the natural language query.

Mechanisms are provided to compress a tool prompt. An original tool prompt is segmented into text chunks. At least one semantic vector representation of the text chunks is generated and a first semantic distribution of the original tool prompt is generated based on the at least one semantic vector representation. A perturbed semantic vector representation is generated by eliminating at least one text chunk from the text chunks, and a second semantic distribution is generated based on the perturbed semantic vector representation. A comparison of the first and second semantic distributions is performed to generate at least one similarity metric. A compressed tool prompt is generated based on the at least one similarity metric by eliminating one or more text chunks that have a similarity metric that is above a threshold similarity value.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for using hidden states of a large language model to generate responses to queries via speculative decoding. In particular, the disclosed systems determine, for a large language model (“LLM”), a prompt comprising a query and an input guide text related to the query. The disclosed systems determine, at a time step of the LLM, candidate draft predictions from token sequences in the input guide text that correspond to a most recent token generated by the LLM. The disclosed systems generate, at the time step of the LLM, a draft prediction by comparing tokens associated with the candidate draft predictions to a hidden state of a previous token prior to the most recent token. Furthermore, the disclosed systems generate, for display via a client device, a response comprising the draft prediction for the prompt to the LLM.

Embodiments are directed to security analysis agents. Events associated with a computing environment may be provided. Prompt fragments may be determined based on the events. A prompt may be generated for a large language model (LLM) based on a prompt template and the prompt fragments such that the prompt fragments may be included in the prompt and provided to the LLM. Actions for evaluating the events may be determined based on the LLM response. These actions may be executed to evaluate the events. Portions of the response that correspond to the prompt fragments may be determined. A performance score may be determined for each prompt fragment based on its corresponding portion of the response such that the prompt may be modified to exclude a portion of the prompt fragments that have a performance score less than a threshold value.