Abstract: A system for judging input mode of form data is configured to extract a study-use field information and a study-use volume of time stamp from each form data having a ground truth of being manually input or automatically input to accordingly execute a learning calculation to generate a judgment calculation model, further to extract an under-judged field information and an under-judged volume of time stamp from each under-judged form data without the ground truth to accordingly generate a judgment result for predicting that the under-judged form data is manually input or automatically input, and further to define judgment-abnormal form data and a trace back ground truth to re-execute the learning calculation for revising the judgment calculation model when the judgment result does not comply with a feedback ground truth.

Abstract: Systems and methods are provided for detecting input via a user interface on a computing device, determining that the input triggers a recommended action related to the input and analyzing historical data to extract relevant data for the recommended action. The systems and methods further provide for generating the recommended action based on the extracted relevant data and causing display of the recommended action on the user interface of the computing device.

Abstract: Embodiments of the invention are directed to a system, method, or computer program product for convenient, secure digital data archiving and utilization across multiple communication channels. The method includes extracting one or more data input fields from a form and generating one or more user prompts to be presented to a user in order to complete at least one of the one or more entries of the one or more data input fields. The method further includes causing the transmission of at least one of the one or more user prompts to the user and receiving a prompt response from the user. The method further includes displaying to the user a completed generated form in an appropriate channel format, and provides a process for secure e-signature and cross-stream utilization of user input.

Abstract: Disclosed in some examples are methods, systems, and machine-readable mediums for utilizing context information to create decoding feedback information to improve decoder accuracy and/or performance. In some examples, the context information is from layers of a network stack above the layers in which the decoders are present. The context information may be or be based upon information about previously received and decoded data and/or information about the sender to provide decoding feedback information to the decoder that is used either to correct a previous decoding error or to inform the decoder on which of a plurality of decoding choices is more likely to be correct. This may increase decoding performance by decreasing errors and in some examples, reducing the complexity of choices by eliminating certain decoding possibilities and thus increasing decoder efficiency.

Abstract: A method and apparatus for determining a quality of a vehicle using text mining may perform operations including performing, by an electronic apparatus, pre-processing on input data including text, performing, by the electronic apparatus, text mining on the pre-processed input data, labeling, by the electronic apparatus, at least one onomatopoeia or at least one morpheme extracted from the pre-processed input data according to the performing of the text mining, and determining, by the electronic apparatus, a quality of the vehicle based on the labeling.

Abstract: The present application is suitable for the technical field of data processing, and provides a data error correction method and apparatus and an electronic device. The method includes: decoding a base sequence to be subjected to error correction into a first text, the base sequence to be subjected to error correction being composed of a plurality of bases; performing word segmentation on the first text to obtain a plurality of text units; performing error detection on the plurality of text units to obtain a text unit having an error; and performing error correction on the base sequence to be subjected to error correction according to the text unit having the error. By means of the above method, error correction for data can be achieved, and the storage cost of DNA can also be reduced.

Abstract: A training data augmentation device (10) includes: an augmented sentence generator (11) configured to generate a plurality of augmented sentences by processing a sentence for training included in training data given in advance according to a plurality of degrees of word replacement; and an augmentation data generator (12) configured to derive a degree of association in a word pair having a dependency relationship in each augmented sentence, determine whether or not to add the augmented sentence to augmentation data for augmenting the training data for each stage based on a comparison result between an obtained degree of association and thresholds of a plurality of stages determined in advance, and generate augmentation data of a plurality of stages by an augmented sentence determined to be added.

Abstract: Systems and methods for processing input streams of calendar applications. An example method, performed by a computer system, may comprise receiving an input stream; processing the input stream to produce a sequence of characters; responsive to determining that the sequence of characters comprises a time reference, storing, in a memory, a calendar entry in a first data structure of a first type, the calendar entry comprising an identifier of a time referenced by the time reference; responsive to determining that the sequence of characters does not comprise a time reference, storing, in the memory, a memorandum in a second data structure of a second type, the memorandum comprising at least part of the sequence of characters.

Abstract: Debiasing pre-trained sentence encoders with probabilistic dropouts may be performed by various systems, services, or applications. A sentence may be received, where the words of the sentence may be provided as tokens to an encoder of a machine learning model. A token-wise correlation using semantic orientation may be determined to determine a bias score for the tokens in the input sentence. A probability of dropout that for tokens in the input sentence may be determined from the bias scores. The machine learning model may be trained or tuned based on the probabilities of dropout for the tokens in the input sentence.

Abstract: A device may receive text data associated with a chatbot, a live chat, or an interactive voice response system, and may preprocess the text data with one or more preprocessing techniques to generate preprocessed data and key intents. The device may convert the preprocessed data and the key intents into embeddings, and may combine the embeddings into an input vector. The device may process the input vector, with a language model, to identify relationships between words and phrases of the text data, and may process the input vector and the relationships, with a summary generation model, to generate a summary of the text data. The device may perform one or more actions based on the summary of the text data.

Abstract: Approaches presented herein provide for the identification of differences between local map data, for a region of a physical environment, and observation or perception data generated by one or more machines or other such sources. In at least one embodiment, sensors on an ego machine can capture sensor data for a region in which the ego machine is located, and a language model on the ego machine can compare this sensor data, or perception data generated using the sensor data, against the local map data. The language model can generate a tokenized description of identified differences, in a domain-specific language. The tokenized description can be transmitted to a map management service that can compare these differences against differences identified by other machines, for example, to determine whether to update and redistribute at least a portion of the map data.

Abstract: Approaches presented herein provide for the automated, end-to-end generation of map data based at least in part on sensor data captured for an environment. At least one language model can be used to generate a text-based, tokenized description of the environment that includes semantics, topology, geometry, and/or other information for the environment. A generation pipeline can use one or more language models in one or more stages, and the data passed between stages can be in a determined tokenized representation format, as may correspond to a tokenized text string in a specific structured language.

Abstract: Approaches presented herein provide for the performance of quality assurance-related tasks with respect to a representation of an environment, such as a set of generated map data. In particular, a language model can be used to generate a tokenized representation of a map using information such as the semantics, topology, and/or geometry determinable from the map data. A language model-generated representation can comply with real-world rules and constructs, and can account for omissions or errors in the input data based upon known relationships and semantics for various objects in the environment. One or more language models can be used to not only identify potential issues in the map data, but also to make recommendations for modifications and/or to generate plaintext descriptions of the issues, modifications, or recommendations to assist a human reviewer in addressing the issues.

Abstract: Approaches presented herein provide for the training of a language model to provide human-style reasoning or “train-of-thought” support for generated inferences. In at least one embodiment, a language model can be used to assist in the generation and/or annotation of content for a specific domain or type of data. This can include, for example, tasks such as performing quality checks for high definition (HD), standard definition (SD), and/or navigational maps. In the mapping context, a language model can be trained using a large set of rules relevant to the mapping domain, in order to become a domain expert. In addition to training the language model on domain-specific rules and data, the language model can be further trained based at least in part on human feedback, such as corrections made to map data by an authorized human.

Abstract: Approaches presented herein provide for the generation of a realistic, higher resolution representation of an environment using a trained language model. In at least one embodiment, map data representative of at least a portion of the environment can be obtained. This map data can be processed using a language model to generate a first tokenized description of the environment based on the input map data. This first tokenized description, which may be in a domain-specific language, can be passed as input to a language model, such as the same language model, which can generate a second tokenized description of the environment that is also in the domain-specific language, but includes additional detail and thus provides a higher resolution representation. This additional detail may include filling in of gaps or accounting for omissions, but may also include inferring aspects such as continuous lanes or complex intersection topography not identified in the input map data.

Abstract: Approaches presented herein provide for the ability to process, store, index, and search geospatial information such as maps with flexible granularity. A set of observations, such as may include sensor data captured for a region of an environment, can be fed as input to a language model. The language model can generate a tokenized description of the region, as may include a text string of tokens encapsulating semantics, topology, geometry, and/or other aspects of the region. A feature vector or embeddings for the region can be generated based on the tokenized description, and a similarity search performed against a vector database, for example, to identify similar feature vectors corresponding to similar regions or domains. Labels or other information associated with these similar feature vectors can be automatically applied to the example region. Clustering of feature vectors or other embeddings can also be performed based in part on the similarity.

Abstract: In systems and methods for processing regulatory questions, textual data representing regulatory questions is obtained by one or more processors. The systems and methods also use one or more natural language processing models to classify the regulatory questions, generate answers to the regulatory questions, generate summaries of the regulatory questions, and/or identify documents that are similar to the regulatory questions. The systems and methods also store, transmit, and/or display data indicative of the classifications, answers, summaries, and/or similar documents.

Abstract: Semantic frame identification involves associating identified target words in the sentential context of their natural language source with semantic frames from a frame lexical database. The disclosed invention utilizes a transformer for semantic frame identification of a target word in a natural language input. Through the transformer's multi-headed attention mechanism, the model learns contextual relationships between words in an input text. The disclosed invention generates a list of potential substitute words for a target word and then verifies the context and meaning of the words in the list to identify a frame from a frame lexical database.

Abstract: A method includes receiving an indication of a first coverage value corresponding to a desired overlap between a dataset of natural language phrases and a training dataset for training a machine learning model; determining a second coverage value corresponding to a measured overlap between the dataset of natural language phrases and the training dataset; determining a coverage delta value based on a comparison between the first coverage value and the second coverage value; modifying, based on the coverage delta value, the dataset of natural language phrases; and processing, utilizing a machine learning model including the modified dataset of natural language phrases, an input dataset including a set of input features. The machine learning model processes the input dataset based at least in part on the dataset of natural language phrases to generate an output dataset.

Abstract: At least some embodiments are directed to an entity classification system that receives informational data associated with an entity. The informational data includes sentences associated with the entity. The entity classification system utilizes a first machine learning model to determine a first contextual meaning among words of a sentence associated with the entity based on a first word embedding technique, and determines at least one category associated with the entity based at least in part on the first contextual meaning. The entity classification system utilizes a second machine learning model to determine a second contextual meaning shared by a set of sentences based on a second embedding technique, and determines a subcategory of the category associated with the entity based at least in part on the second contextual meaning. The entity classification system generates an output including the category and subcategory associated with the entity.

Abstract: Various embodiments discussed herein are directed to improving existing technologies by detecting a likelihood of hallucination arising from one-shot, few-shot, or outside knowledge contexts. For example, regarding the one-shot or few-shot contexts, some embodiments determine a set of tokens in a language model output that are not found in target content but are found in at least one example. When such phrases are not very common words, this is highly indicative that the model is hallucinating because these phrases should be located in the target content but are not, but are instead located in the examples.

Abstract: Provided are processes of balancing between exploration and optimization with knowledge discovery processes applied to unstructured data with tight interrogation budgets. A process may include determining a relevance probability distribution of responses and scores as an explanatory diagnostic. A distribution curve may be determined based on a probabilistic graphical network and a result may be audited relative to the distribution curve to determine noise measurements. The distribution curve may be determined based on a distribution of posterior predictions of entities to score ranking entity bias and noisiness of ranking entity feedback.

Abstract: A method for quantitatively measuring a psychological complexity of a response to visual stimuli is described. The method includes generating, by a trained machine learning model, different dataset types of reference text descriptions and/or representations of the visual stimuli. The method also includes collecting input response text descriptions from an individual after perceiving the visual stimuli. The method further includes scoring a difference between the input response text descriptions and the different dataset types of reference text descriptions and/or representations of the visual stimuli, in which a score is generated based on the different dataset types of reference text descriptions. The method also includes generating a complexity score representative of a cognitive complexity of the input response text description from the individual based on the score generated based on the different dataset types of reference text descriptions.

Abstract: Disclosed are various embodiments for compliance detection using natural language processing. Various embodiments include a computing device that can transcribe a sound recording of a transcript, where the sound recording can be representative of a telephonic call occurring between an agent and a client. The computing device can determine that the telephonic call included a discussion related to a regulated subject based on an analysis by a natural language processor of the transcript. The computing device can obtain a compliance rule based at least in part on the regulated subject of the telephonic call. The computing device can determine that the compliance rule has been violated using a natural language processor of the transcript. The computing device can also store various information related to the compliance violation.

Abstract: Systems and methods are provided for facilitating automated administration of email messages received by a customer service representative including a service email administrator system that reviews incoming email messages, analyzes the content, assesses emotional sentiment, urgency, and determines the best course of action. The conversation administrator is configured to use natural language processing and machine learning algorithms to generate automated responses, provide email response drafts for customer service representative (CSR) review, or route emails to CSRs based on the sentiment or urgency.

Abstract: A customized prompt generation service automates prompts to a large language model to perform a specified software engineering task. The service stores the custom data of a client that includes code diff hunks, source code segments, code reviews, repaired code, and unit tests from a code base or repository of the client. Prompt templates are associated with each software engineering task that include the requisite information needed for the large language model to perform the target task. A prompt to a large language model includes examples of the software engineering task from the custom data of the client of the service.

Abstract: Systems and methods for generating output data based on a data graph are provided. An output request for the output data based on the data graph is received. The output request comprises one of a natural language request from a target user or an application request from an application of the target user. The data graph has nodes and edges between the nodes. The nodes represent entities associated with an enterprise organization and the edges represent relationships among the entities. A graph data query is generated with a large language model (LLM) using the output request as a first input to the LLM. The graph data query is performed against the data graph to obtain a graph data output that represents a sub-portion of the data graph. The output data is generated with the LLM using the graph data output as a second input to the LLM.

Abstract: The technology provides rewriting suggestions for communication styles in different pre-defined styles. Rewriting may convert text in different ways, including visually enhancing the message. A method provides input to a trained large language model, the input including curated examples associated with one or more writing style choices. The set of curated examples has a first size. The method also includes generating, using the model, a rewriting corpus according to one or more writing style choices. The rewriting corpus has a size two or more orders of magnitude larger than a size of curated examples. The writing style choices include at least one of tone, conversion, application context, or conversation type. A text rewriting model is trained using at least a subset of the rewriting corpus. The model is configured to generate vivid textual information in response to user input in an interactive domain, according to specific writing style choices.

Abstract: The techniques disclosed herein manage computing environments associated with radio access networks using a natural language interface. This is achieved through utilizing natural language processing to analyze user generated inputs and generate robust large language model queries. In various examples, the queries can include radio access network documentation, diagnostic data, and past interactions to provide custom context to the large language model. Accordingly, the query can cause the large language model to generate an operation sequence comprising a plurality of commands to interface with a resource management tool and control computing resources and supporting components. In this way, the present techniques can alleviate the technical burden on end users and minimize the risk of errors.

Abstract: This disclosure describes one or more implementations of systems, non-transitory computer-readable media, and methods that extract viewpoints from content for syntopical reading using an efficient claim-relation graph construction approach. For example, the disclosed systems utilize sentence transformers with claims from content to embed the claims within a metric space (as claim nodes). Furthermore, in some embodiments, the disclosed systems generate a claim relation graph for the claims by utilizing approximate nearest neighbor searches to determine relational edges between a claim node and the claim node's approximate nearest neighbors. Moreover, in some implementations, the disclosed systems utilize the claim relation graph with an edge weighted graph neural network to determine stance labels during extraction of viewpoints (e.g., stance, aspect, and topic) for the claims. Additionally, in one or more instances, the disclosed systems utilize the extracted viewpoints in content retrieval applications (e.g.

Abstract: Systems and methods for providing an artificial intelligence (AI)-based interface for an application include receiving a prompt from a user interface (UI) component of an interface client that defines at least one task to be performed in the application. The prompt is supplied to at least one language model as input. The at least one language model is trained to process the prompt to identify the at least one task to be performed, generate new content if required by the at least one task, and domain-specific instructions for causing the tasks to be performed in the notes application. Notes domain-specific language (NDSL) instructions are provided as output to the notes application where they are executed in the notes application to perform the at least one task.

Abstract: One example method includes selecting an entry from a validation set based on similarity to a first communication record, wherein the entry from the validation set includes a second communication record and a corresponding summary of the second communication record; generating a prompt that includes the first communication record, the second communication record, and the corresponding summary of the second communication record; inputting the prompt to a first language model to obtain a generated summary of the first communication record; and training a second language model based on the first communication record and the generated summary.

Abstract: A computer implemented method for translating a document. A number of processor units separate the document into elements having media types. The number of processor units determine attributes for the elements. The number of processor units create a virtual map identifying relationships between the elements using the attributes. The number of processor units translate the elements into a target language based on media types for the elements. The number of processor units adjust translations for the elements based on the relationships between the elements using the virtual map to create adjusted translations for the elements. The number of processor units generate the translated document using the adjusted translations for the elements and the virtual map.

Abstract: Technology is disclosed herein for an application for generating and auto-tuning reading passages. A method of operating an application includes generating a prompt which elicits a reply from a foundation model service that includes a reading passage tailored to the ability of a reader. The prompt includes instructions for generating the reading passage using one or more trouble words and according to a desired readability. The desired readability is based on one or more indicators of a reader's reading ability. The application assesses the readability of the reading passage, resulting in an assessed readability. The application generates a follow-on prompts with which to elicit follow-on replies from the foundation model service that includes a request to adjust the reading passage based on the assessed reliability. The application enables display of the adjusted reading passage in a user interface of the application.

Abstract: A voice processing device for generating translation results for voices of speakers is disclosed. The voice processing device comprises: a microphone for generating voice signals associated with voices of speakers in response to the voices of the speakers; a memory for storing location-language information indicating languages corresponding to sound source locations of the voices of the speakers; and a processor which uses the voice signals and the location-language information so as to generate translation results obtained by translating the languages of the voices of each speaker, and which uses the translation results so as to generate translation conference minutes including the voice contents of each speaker expressed in different languages.

Abstract: Systems and methods are disclosed for contrastive learning with adversarial data for robust speech translation. For example, a method may include inputting a speech signal to an automatic speech recognition model to obtain a transcript hypothesis including a first sequence of tokens, wherein the speech signal is associated with a golden transcript including a second sequence of tokens; inputting the first sequence of tokens to an encoder of a neural machine translation model to obtain a first sentence representation; inputting the second sequence of tokens to the encoder of the neural machine translation model to obtain a second sentence representation; determining a contrastive loss function based on the first sentence representation and the second sentence representation; and training the encoder of the neural machine translation model based on the contrastive loss function.

Abstract: Real time location systems are provided including one or more ultra-wideband (UWB) sensors positioned in an environment; one or more image capture sensors positioned in the environment; and at least one UWB tag associated with an object in the environment to provide a tagged item in the environment. The one or more UWB sensors and the one or more image capture sensors are integrated into at least one location device. The at one location device includes a UWB location device, a combination UWB/camera location device and/or a camera location device. A location of the tagged item is tracked using the at least one location device. and wherein a location of the tagged item is tracked using the at least one location device.

Abstract: A wireless networking system is disclosed. The wireless networking system includes an application layer associated with one or more applications having a wireless bandwidth requirement. A first wireless transceiver resource associated with an actual MAC layer and PHY layer is employed. The first wireless transceiver resource has a first bandwidth availability up to a first actual bandwidth. A second wireless transceiver resource associated with the actual MAC layer and the PHY layer is employed. The second wireless transceiver resource has a second bandwidth availability up to a second actual bandwidth. A processing layer evaluates the wireless bandwidth requirement and the first and second bandwidth availabilities of the wireless transceiver resources. The processing layer includes a bandwidth allocator to allocate at least a portion of each of the first and second actual bandwidths to virtual MAC and virtual PHY layers, and to satisfy the application layer wireless bandwidth requirement.

Abstract: An image forming apparatus comprises a first obtaining unit configured to obtain first color values of a base color of a print medium by reading the print medium; a second obtaining unit configured to obtain second color values different from the first color values based on the first color values; and a print unit configured to print a mark of the second color values on the print medium.

Abstract: Embodiments of a system and a method for detecting moisture within a structure can use a cementitious panel including a moisture sensor. The moisture sensor can comprise a moisture-sensing, passive RFID tag. A reader can be used to periodically interrogate the RFID moisture-sensing tag to determine whether the RFID moisture-sensing tag has been subjected to moisture. Systems and methods for detecting moisture within a structure can use at least one such cementitious panel in a structural assembly, such as a roof assembly, for example, to help detect a leak in the roof assembly through periodic monitoring of the installed cementitious panel(s).

Abstract: A storage device includes a storage controller and a printed circuit board. The printed circuit board includes a host interface connector including first pins coupled to an external host device, a controller socket in which the storage controller is mounted, and a first slot that may receive a first secure digital (SD) Express device, the first slot including second pins to be coupled to the first SD Express device. A first receive pin among the first pins is connected to a first receive pin among the second pins, a second receive pin among the first pins is connected to a second receive pin among the second pins, a first transmit pin among the first pins is connected to a first transmit pin among the second pins, and a second transmit pin among the first pins is connected to a second transmit pin among the second pins.

Abstract: A contactless electronic module for a data carrier comprises a substrate, at least one electronic chip, and at least one electrical connector. The electronic chip is arranged on the substrate, and the electrical connector is in connection with the electronic chip. The substrate is electrically non-conductive. Other embodiments disclosed.

Abstract: Disclosed are pre-cure RFID-enabled bead labels based on an RFID inlay construction consisting of an aluminum antenna etched on to a high temperature resistant polyimide film that is connected to an integrated memory circuit positioned on the surface of the polyimide film. This RFID inlay being further inserted into an overall label construction having a generally arcuate or semi-arcuate shape and a plurality of layers that include, for example, a plurality of polyester layers and a plurality of high temperature resistant adhesive layers that bond/adhere layers together, the plurality of layers further protecting and insulating the RFID inlay while the label is bonded to the external bead (or sidewall) of a tire. The compositions/devices disclosed herein can be used for electronic identification when applied on rubber-based articles (e.g., tires) prior to being subjected to stress related to the vulcanization process and normal use of this article during the manufacturing process.

Abstract: A patch radiofrequency identification (RFID) device for tires comprises: a substrate; a first insulating layer covering part of the substrate; an RFID chip and first antenna connected to each other and arranged on the first insulating layer; a second antenna electromagnetically coupled with the first antenna and extending, at least partially, on the first insulating layer. Part of the second antenna extends around the first antenna and includes two ends located at opposite sides thereof, and another part includes one or more arms extending in opposite directions from the first antenna, the arms being respectively formed by meander-line-shaped and/or spiral-line-shaped wire. The RFID chip, the first antenna, and the second antenna lie on a same plane. The second antenna is a parasitic radiator. A second insulating layer covers the first insulating layer, the RFID chip, the first antenna and, at least partially, the second antenna.

Abstract: A system and method for evaluating the performance and usage of a cognitive computing tool which answers questions from users. A log file for these interactions includes the questions, the answers and a confidence rating assigned by the tool to each answer. Questions and answers are analyzed to determine validity, accuracy, and categories by subject matter experts or text analytics tools, and the results are added to the log file. Comments and sentiments from users may be analyzed and added to the log file. Additional data about the users, such as identities, demographics, and locations, may be added. Data from the log file may be presented in a dashboard display as metrics, such as trends and comparisons, describing the usage and performance of the cognitive computing tool. Answers may be displayed as they were presented to the users. Selectable filters may be provided to control the data displayed.

Abstract: According to an embodiment of the present disclosure, a method for providing an artificial intelligence-based model, performed by a computing device, is disclosed. The method includes providing a candidate model list comprising a plurality of candidate models which are artificial intelligence-based models. The method includes determining a target model to be converted or benchmarked, based on a first user input on the candidate model list. The method includes providing a candidate node list comprising a plurality of candidate nodes. The method includes determining a target node to be converted or benchmarked, based on a second user input on the candidate node list. The method includes converting the target model into a model supportable by the target node, based on information of the determined target model and information preset on the target node.

Abstract: An allocation system includes a memory storing instructions, and a processor. The processor is configured to access the memory and execute the instructions to: obtain data that includes profiles of doctors and clinical initial information of a patient that arrives at a hospital; preprocess the obtained data, wherein preprocessing includes a text preprocessing pipeline and word embedding; allocate a doctor to the patient that arrives at the hospital by using a neural network, wherein the neural network is machine-learned by using graph data of clinical narratives composed by doctors about patients, historical profiles of the patients and the profiles of doctors; and output information indicating the allocated doctor.

Abstract: Systems, methods, and computer program products for determining long-range dependencies using a non-local graph neural network (GNN): receive a dataset comprising historical data; generate at least one layer of a graph neural network by generating graph convolutions to compute node embeddings for a plurality of nodes of the dataset, the graph convolutions generated by aggregating node data from a first node of the dataset and node data from at least one second node comprising a neighbor node of the first node; cluster the node embeddings to form a plurality of centroids; determine an attention operator for at least one node-centroid pairing, the at least one node-centroid pairing comprising the first node and a first centroid; and generate relational data corresponding to a relation between the first node and at least one third node comprising a non-neighbor node of the first node using the attention operator.

Abstract: Provided are a multi-criteria recommender apparatus and method. The multi-criteria recommender apparatus obtains authorization for multi-criteria evaluation data provided by a user evaluating each item according to a plurality of different evaluation criteria to acquire a multi-criteria extended graph including a user node and an item node, wherein the item node is expanded into a plurality of sub-nodes according to the plurality of evaluation criteria, and selects a recommended item in consideration of the user's preferences for the plurality of evaluation criteria on the basis of embedding data that is obtained by performing a neural network operation on the multi-criteria extended graph.

Abstract: Embodiments relate to an automatic detection of fraudulent behavior for a transaction at an online system. The online system requests a large language model (LLM) to determine, based on a prompt input into the LLM, information about a refund event for a first order placed by a user of the online system. The online system accesses a computer model trained to detect a fraudulent behavior associated with an order placed with the online system. The online system applies the computer model to determine a score associated with the refund event, based on the information about the refund event received from the LLM. The online system determines, based on the score, whether the refund event was due to a fraudulent behavior of the user. The online system performs at least one action associated with the online system, based on the determination whether the refund event was due to the fraudulent behavior.