18 results about "Recall rate" patented technology

The invention provides a method and system for improving RAG recall rate and accuracy based on block global relevance, and the method comprises the steps: detecting a newly-added document through timing scheduling, carrying out block processing, and fusing an explicit structure relation and an implicit semantic relation to construct a weighted adjacency matrix; block global correlation scores are calculated through an iterative algorithm; the query similarity and the global correlation score are dynamically fused, and Top-K blocks are returned to generate answers; the system comprises a knowledge block graph construction server, a global correlation calculation server, a fusion retrieval server and a data storage server. According to the method, the limitation that traditional RAG only depends on local similarity is broken through, the core knowledge recall rate is remarkably increased, redundant noise is restrained, and the accuracy and logicality of generated answers are improved by quantifying the global importance of the blocks in the knowledge network.

The application discloses a strategy determination method and device and related equipment, and relates to the technical field of computers. The method comprises the following steps: determining a first recall rate and a first injury rate corresponding to a first strategy based on a classification result of the first strategy; the first strategy comprises at least one sub-strategy; different sub-strategies in the at least one sub-strategy correspond to different preset strategy values, and the preset strategy values are used for determining a classification label; the recall rate is the proportion of correct classification labels determined by a strategy, and the injury rate is the proportion of incorrect classification labels determined by a strategy; at least one iteration is performed on at least part of the sub-strategies in the first strategy, and the iteration is used for adjusting preset strategy values corresponding to the at least part of the sub-strategies; and a second strategy after iteration is obtained under the condition that a preset condition is met. The method improves the classification accuracy of the second strategy after iteration optimization.

The application discloses a kind of lightweight explainable aviation equipment abnormal state perception method and system, belong to aircraft health management and deep learning technical field.The application constructs 1+1 branch's LiteInception light weight feature extraction network, designs and aligns two-stage cascaded diagnosis process with aviation maintenance process: first stage adopts LiteInception and hybrid architecture of Transformer to execute fault detection binary classification, and priority is guaranteed recall rate;Second stage adopts pure LiteInception architecture to execute fault identification multi-classification to abnormal sample, and priority is guaranteed accuracy and explainability.The application significantly compresses model size, adapts edge device real-time inference needs, can output the diagnosis evidence chain that can be traced to specific sensor, specific time period, considers diagnosis efficiency, precision and compliance, and provides complete solution for general aircraft fault diagnosis.

The application discloses a Chinese human phenotype ontology recognition method and related equipment, which can be applied to the technical field of text information processing. The application can effectively improve the recall rate by expanding the human phenotype ontology concept synonym set through a large language model, generating an expanded dictionary with the initial dictionary, significantly enhancing the discrimination ability of the recognition model for semantically ambiguous concepts through two-stage comparative learning, balancing the precision and recall rate by constructing an initial dictionary based on the human phenotype ontology official database and the Chinese human phenotype ontology official data, merging and disambiguating the current input text after accurate matching and semantic similarity matching to obtain a fused candidate result list, and finally obtaining the structured recognition result corresponding to each current input text by post-processing each candidate fragment in the fused candidate result list, which can provide effective data support for disease diagnosis.

The embodiment of the invention provides an auxiliary proof-giving method, an auxiliary proof-giving system, electronic equipment, a storage medium and a computer program product. The auxiliary proof-proof method comprises the following steps: acquiring right protection request content corresponding to a to-be-proof right protection event; performing case analysis on the right protection request content to obtain a right protection category positioning result matched with the right protection event; retrieving a preset checking template based on a right protection category positioning result to obtain a proof item list for the right protection event; and executing a verification task corresponding to the proof item list based on a mode of iteratively retrieving a preset knowledge base, and obtaining a proof suggestion report corresponding to the right protection event for guiding the user to provide a proof. According to the method, an iterative autonomous retrieval mechanism is driven by a task, an analysis process of manual gradual evidence collection is simulated through multiple rounds of query, and compared with a manual dominant evidence collection process, the evidence collection efficiency is improved, and the recall rate and the accuracy of key evidence are effectively improved.

The present application relates to a credit report generation method and device based on a multi-index context bus, applied to the field of artificial intelligence technology, comprising: generating micro-context through "entity-spatiotemporal-semantic" three-dimensional filtering, physically isolating noise data, effectively solving the "lost in the middle" effect of large model long text input, significantly improving the recall rate and reasoning accuracy of key risk information; using a "configurable expert skill body" to replace full-parameter fine-tuning, decoupling business logic and model, and realizing instant effect by updating the configuration only for credit policy adjustment, avoiding catastrophic forgetting and reducing model maintenance cost; using pointerized context transmission and life cycle management to greatly reduce memory occupation and IO overhead, effectively supporting high-concurrency credit reasoning scenarios.

This disclosure provides a category-level recall control method, device, electronic device, and storage medium, involving natural language processing and deep learning technologies in the field of artificial intelligence. It can be applied to multi-category classification scenarios in high-risk businesses such as content security review, financial risk control, and public opinion monitoring. The specific implementation scheme is as follows: Input samples are classified based on a pre-set multi-category classification model to obtain calibration scores for each category on the input samples; based on pre-configured target recall rates for each category, and combined with a pre-built mapping table of recall rates and thresholds, a standard threshold corresponding to the target recall rate of each category is determined; the category with the highest calibration score among the multiple categories is selected as the candidate predicted category for the input sample; based on the candidate predicted category, a standard threshold for the candidate predicted category is determined from the standard thresholds corresponding to the target recall rates of each category; based on the calibration score of the candidate predicted category and the corresponding standard threshold, it is determined whether to recall the input sample.

The application provides an abnormal waybill detection method and device. The abnormal waybill detection method comprises the following steps: obtaining a plurality of to-be-predicted waybill information; determining a plurality of target matching conditions met by each to-be-predicted waybill information based on the plurality of to-be-predicted waybill information and a plurality of preset matching conditions; determining an abnormal probability value of each to-be-predicted waybill information belonging to an abnormal waybill based on each to-be-predicted waybill information and the plurality of target matching conditions met by each to-be-predicted waybill information; and performing abnormal detection on the plurality of abnormal probability values to obtain an abnormal value in the plurality of abnormal probability values. Compared with the prior art, the application improves the prediction accuracy by only relying on waybill information. In addition, after obtaining the abnormal probability value of each to-be-predicted waybill information, the application performs abnormal detection on the plurality of abnormal probability values, thereby identifying abnormal waybills, improving the recall rate of prediction, and thus improving the accuracy and recall rate of abnormal waybill detection.

Some embodiments include systems and methods for recall estimation. An exemplary method comprises determining reservoir sampling and size of labeling from each strata; determining recall and variance for each strata; determining sum of sample size for a time period; and determining reservoir sampling for each strata and label the sample items. Other embodiments are described.

The invention discloses a semantic enhanced vulnerability information retrieval method and related equipment. The method comprises the following steps: acquiring a vulnerability retrieval request of a user; the vulnerability retrieval request of the user is input into the trained Sensor Transform model to be subjected to vectorization processing, and a vulnerability retrieval request vector of the user is obtained; and performing semantic similarity calculation on the vulnerability retrieval request vector of the user and a vulnerability vector set pre-stored in the vulnerability retrieval database by utilizing an ElasticSearch engine to obtain a vulnerability retrieval result. According to the method, semantic related results can be accurately understood and retrieved, and the retrieval accuracy and recall rate are remarkably improved.

This application relates to the field of data processing technology, and provides a method and related apparatus for selecting recall strategies, which selects suitable recall strategies for specific users of an application to improve the recovery of users with low activity. The method mainly includes: determining the number X of recall strategies in the recall strategy set; dividing the sample user group into X sub-sample user groups, where X is a positive integer greater than 0; for each of the X sub-sample user groups, pushing the same recall strategy from the recall strategy set to sample users within the same sub-sample user group; calculating the recall rate for each sub-sample user group based on the number of users responding to the recall strategies in the recall strategy set within each sub-sample user group; and selecting target recall strategies that meet a preset standard for a target recall rate, where the target recall rate is one or more of the recall rates, and the target recall strategy is one or more of the recall strategies.

The embodiment of the invention provides a text partition length determination method and device, equipment and a medium. The method comprises the steps that text data and at least one sliding step length are obtained; performing sliding window sampling on the text data based on the at least one sliding step length, and determining at least one first candidate interval; determining a first overall evaluation score of each first candidate interval in the at least one first candidate interval, wherein the first overall evaluation score is obtained based on weighted comprehensive evaluation of recall rates and accuracy rates of a plurality of partition lengths in the first candidate interval; determining a target interval in the at least one first candidate interval according to the first overall evaluation score; and the golden section method is applied to the target interval to determine the length of the target partition, so that the recall rate and the accuracy rate are effectively improved, and the application of the large model in actual production is ensured.

The invention relates to the technical field of network security, and provides a line protection device network attack detection method and system based on deep learning, which combines mean square error loss function optimization with a recall rate index, ensures the high-precision reconstruction capability of normal data through a loss function in a model training stage, and improves the network attack detection accuracy. And the weight corresponding to the highest recall rate is used as a final model parameter, so that the reconstruction precision and the attack sample detection sensitivity are considered, the risk of missing report is remarkably reduced, and the generalization robustness is improved.

The application relates to the field of vector databases, and aims to solve the problems of high delay and low recall rate caused by different distribution user queries, and provides a vector database system based on query distribution perception, which comprises an index construction module, a vector retrieval module, an interval measurement module, a log recording module, a distribution perception module, a distribution detection module, a distribution mapping module, a storage module, an interface agent module and a graphical management interface; the application provides an efficient solution for storage and retrieval of vector data, simultaneously, the distribution of user queries is monitored in real time, the vector distribution is effectively measured, the vector index structure is optimized through historical user queries, and the stability and reliability of vector database query quality are guaranteed.

The invention is applied to the technical field of digital pathology and artificial intelligence crossing, and particularly discloses a multi-instance learning ovarian cancer classification method capable of resisting single-side tag noise. Comprising the following steps: S1, collecting and preprocessing tissue pathological images of an ovarian cancer patient; S2, generating an initial weak label training set; S3, constructing a dual-weighted anti-noise objective function; S4, training a model by adopting an alternating optimization strategy; according to the multi-instance learning ovarian cancer classification method resistant to unilateral tag noise, through an innovative instance-packet dual weighting mechanism and an alternative optimization strategy, the problem of unilateral tag noise generally existing in pathology image weak supervised learning is effectively solved, high accuracy, high recall rate and high robustness can still be kept under noise interference, and the classification method is applicable to classification of ovarian cancer. And the method has good interpretability.

The invention discloses a gene mutation detection method, device and equipment, a medium and a product. The method comprises the steps that suspected mutation sites of a nucleic acid sample to be detected are obtained, and the suspected mutation sites are determined through first mutation characteristic data obtained by conducting mutation detection on sequencing data of the nucleic acid sample to be detected through a first mutation detection module; the recall rate of a gene mutation site identified by the first mutation detection module is greater than or equal to a preset recall rate; acquiring second mutation feature data and third mutation feature data of the suspected mutation sites, inputting the second mutation feature data and the third mutation feature data into a pre-trained target mutation detection model, and outputting mutation detection results of the suspected mutation sites.

The embodiment of the invention provides a model training method and device, electronic equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: retrieving from a preset literature library based on a keyword set to obtain a literature set; clustering the literature set to obtain a plurality of literature subsets; extracting at least one document from each document subset as a first document; positive samples and negative samples are determined from the first literature through the multiple large language models, the initial model is trained, and a target model is obtained. The method comprises the following steps: training a baseline model on a few-sample annotation data set of multi-model voting, enhancing the baseline model by adopting an active learning strategy, evaluating a randomly selected literature set by using the baseline model, iteratively annotating a sample with the lowest predictability, adopting a large-model debate thought, and carrying out iterative annotation on a sample with the lowest predictability in a simulation debate mode. And when indexes such as the accuracy rate and the recall rate of the model are no longer remarkably improved, iteration is stopped, so that the trained model has relatively high target domain identification capability.

The invention discloses an RAG recall rate improving method, and belongs to the technical field of natural language processing. The method specifically comprises the following steps: S1, knowledge base document multi-level processing: performing classification screening, semantic paragraph segmentation, entity labeling and multi-granularity vector coding on a target domain knowledge base to obtain a standardized semantic paragraph containing classification ID-paragraph ID-entity list-multi-granularity vector; and S2, generating and filtering Q2Q semantic association: generating a multi-dimensional potential problem for each standardized semantic paragraph by adopting a pre-training language model of field fine tuning, and screening 3-5 high-quality potential problems through rule filtering and model scoring. Through'field fine tuning of a T5-XXL model + multi-dimensional variant generation ', three variants of deep semantic association, coverage of expression styles, scene supplementation and field term adaptation of user intentions are accurately captured, and the problems of long tails and the recall rate bottleneck of oral query are solved.