15 results about "Word structure" patented technology

The invention discloses a power plant operation and maintenance knowledge intelligent query method based on a large language model and an RAG technology. The method comprises the following steps: constructing a power plant operation and maintenance knowledge vector library covering structured, semi-structured and unstructured data; receiving a natural language question of a user, inputting an improved instruction to align a preprocessor, and generating a question semantic vector and an intention tag; relevant knowledge fragments are retrieved and sorted through a semantic matching retriever in combination with the intention labels; constructing a large language model cue word structure based on the retrieval result and the original question, generating candidate answers and recording a reference path; and finally, performing term specification and consistency verification according to the expert rule base, and outputting a structured and traceable final answer. According to the invention, the improved RAG technology is fused to realize intelligent query of the operation and maintenance knowledge of the power plant.

The invention discloses an emergency pre-examination grading system and method based on collaborative decision of a large language model and a tree model. The method comprises the following steps: constructing a data set according to patient information and screening data; training an initial random forest model based on the data set, generating a basic decision rule, mining and extracting high-frequency features based on association rules, and combining to obtain a candidate rule pool; constructing a cue word structure adaptive to the field, driving LLM to complete rule correction, and forming a correction rule set; performing multiple rounds of rule random division and rule combination generation based on the correction rule set, and then screening out an optimal rule; and on the basis of a specific scene, the expert rule and the corrected optimal rule are fused, matching is carried out on patients, and emergency pre-examination grading is realized. According to the method, the interpretability is improved while the grading accuracy is improved, and the problem of poor cross-courtyard generalization of a machine learning model is effectively solved. The generalization ability of the rule is remarkably improved, and the method is adaptive to a multi-center combined diagnosis and treatment scene.

The invention discloses a system for interconversion between a Word file and an HTML file. The system comprises a double-engine analysis module, a format mapping module and a document output module, the double-engine analysis module comprises a Word analysis engine and an HTML analysis engine, and the format mapping module comprises a Word-to-HTML mapping sub-module and an HTML-to-Word mapping sub-module; the Word analysis engine is used for processing Word file input; the HTML analysis engine is used for processing HTML file input; the Word-to-HTML mapping sub-module is used for mapping Word information obtained through analysis of the Word analysis engine into an HTML equivalent structure; the HTML-to-Word mapping sub-module is used for mapping the HTML information obtained by analysis of the HTML analysis engine into a Word equivalent structure; and the document output module is used for generating an HTML document or a Word document according to the conversion result and outputting the HTML document or the Word document. According to the system, bidirectional lossless conversion can be realized, complete closed-loop migration of styles, structures and functions is realized, and platform adaptability of interactive elements and cross-format deep semantic feature correspondence are kept.

The invention relates to a domain tool calling conflict-oriented large model cue word optimization method and system, and belongs to the technical field of artificial intelligence and natural language processing. The method comprises the following steps: performing semantic analysis on a natural language instruction of a user to generate a structured semantic representation, and performing conflict diagnosis from different dimensions based on the structured semantic representation to obtain structured conflict diagnosis information; determining a processing priority and a to-be-executed coordination strategy according to the conflict diagnosis information, and generating an optimized cue word structure set according to the coordination strategy; and finally, inputting the optimized cue word structure set into a tool calling interface, constructing and executing a standardized calling request containing a target tool identifier, parameter filling content and an operation sequence, and carrying out tool calling according to the standardized calling request. According to the method, the problem of prompt word conflict in multi-tool calling is effectively relieved, and the calling accuracy and the scheduling stability of the system in a complex task scene are improved.

The invention relates to the technical field of data processing, in particular to a data information storage method and system for an online Chinese language learning platform. The method comprises the following steps: obtaining platform user interaction data, and carrying out pinyin input track analysis and Chinese character component structure coding to obtain a pinyin alignment input frame; extracting a pinyin-Chinese character mapping relation in the pinyin alignment input frame, and performing Chinese character structure path reconstruction to obtain a user input word structure path set; learning content data are called, multi-modal content node embedding is carried out, and a content linkage fragment cluster is obtained; constructing a single session learning topology track according to the content linkage fragment cluster, and performing bidirectional topology sorting to obtain a learning track topological graph; and establishing a multi-dimensional primary key index and behavior triggering mechanism for the learning track topological graph to obtain a dynamic index distribution block, and transmitting the dynamic index distribution block to an online Chinese language learning platform. According to the invention, the stability and expansibility of the learning platform are obviously improved.

The utility model relates to carton stacking technology field, concretely relates to a carton stacking device, including the inner chamber is the mouth word structure's frame, the upper portion of frame is provided with the inner chamber is the mouth word structure's support, the top of frame is provided with a plurality of fixed rods, the bottom of a plurality of fixed rods is installed at the edge of frame respectively, all be provided with fixed frame on a plurality of fixed rods, the bottom of fixed frame is installed at the top of fixed rod, the top of fixed frame is movably sleeved with fixed leg, and fixed leg is installed on the support, two fixed shafts that rotate are provided on the frame, fixed plate and baffle are all provided on the fixed shaft respectively, one end of baffle and one end of fixed plate are all installed on the fixed shaft, first telescopic cylinder is arranged between fixed plate and frame, one end of first telescopic cylinder is hinged on the other end of fixed plate, and the other end of first telescopic cylinder is hinged on the frame. The utility model provides a kind of carton stacking device for conveniently taking out carton, support multiple paths to carton and stack.

This invention discloses a method and system for fine-tuning secure instructions for a multimodal large model based on virtual adversarial techniques, belonging to the field of large models. This invention performs two-stage fine-tuning of the multimodal large model: the first stage uses gradient-based virtual adversarial perturbation technology to securely align the large language model and cross-modal connection layers, ensuring the harmless nature of the model's output content and improving its resistance to subsequent malicious data; the second stage, with the parameters fixed after secure alignment, introduces a trainable implicit prompt word structure, achieving effective adaptation to user-personalized data only by adjusting its key-value pairs, preventing the reintroduction of harmful information during fine-tuning. This invention reduces dependence on preference data while achieving synergistic optimization of security and personalization needs, providing a reliable guarantee for the practical deployment of multimodal large models.

The application provides a model calling request generation method and device and electronic equipment, and is applied to the field of artificial intelligence technology. Through determination of a classification scene label and matching of a prompt word structure template, dynamic slot adaptation based on a classification scene is realized. After user input is converted into prompt word configuration data, unassigned slots and weak constraint slots are determined, and active guidance information is generated, so that field mismatch caused by random completion or general expansion of a model is avoided. After receiving a user response and updating slot values, a prompt word package is generated and compiled into a calling request of a target generation model, so that each slot is accurately mapped to a model support field, unstructured sentences are converted into structured control information executable by the model, the number of repeated generation times and generation quota consumption of the user are reduced, and generation quality and efficiency are significantly improved.

The present application relates to the technical field of data processing, and more particularly to a data information storage method and system for an online Chinese language learning platform. The method comprises the following steps: obtaining platform user interaction data, and performing pinyin input track analysis and Chinese character component structure coding to obtain pinyin aligned input frames; extracting pinyin-Chinese character mapping relationships in the pinyin aligned input frames, and performing Chinese character structure path reconstruction to obtain a user input word structure path set; calling learning content data, and performing multi-modal content node embedding to obtain content linkage segment clusters; constructing a single session learning topology track according to the content linkage segment clusters, and performing bidirectional topology sorting to obtain a learning track topology graph; establishing a multi-dimensional primary key index and a behavior triggering mechanism for the learning track topology graph to obtain a dynamic index distribution block, and transmitting the dynamic index distribution block to the online Chinese language learning platform. The present application significantly improves the stability and expandability of the learning platform.

The application relates to the technical field of intelligent voice interaction, in particular to a multi-scene intelligent assistant system based on dialect voice wake-up and an interaction method, which comprises a voice segmentation module, a scene screening module, a terminal access module, an intra-domain wake-up module and an entry determination module. The application establishes a continuous processing relationship by surrounding the voice segment boundary, the scene label residence state, the terminal release state and the dialect wake-up word structure, narrows the scene range of the current voice segment participating in identification according to the scene label residence state, checks the tone fluctuation, the syllable sequence and the tailing convergence in the limited range, judges the interaction entry attribution in combination with the terminal access state and the activation confirmation information, keeps the scene constraint and the dialect wake-up discrimination synchronous, keeps the terminal access and the entry confirmation connected, controls the cross-scene repeated matching, the near-sound trigger confusion and the terminal access conflict, and makes the dialect voice wake-up and the subsequent interaction more coherent.

The invention provides a live broadcast content personalized recommendation method and device based on a user portrait. The method comprises the following steps: collecting real-time hot word data of each live broadcast content on a current live broadcast platform; constructing a live broadcast hot word structure tree based on the real-time hot word data of each live broadcast content; performing user portrait analysis based on the user portrait data to obtain user interest hot words; performing association mapping based on the live broadcast hot word structure tree of each live broadcast content and the user interest hot word to obtain a hot word association mapping result; and determining a matching degree value between each live broadcast content and the user interest hot word based on the hot word association mapping result of each live broadcast content, and generating a live broadcast content personalized recommendation list based on the matching degree value of each live broadcast content. According to the invention, the accuracy and diversity of live broadcast content recommendation are improved, and poor homogenization and timeliness are avoided.

The application provides a multi-track music generation method and device, the method comprising: modifying a compound word structure; generating a compound word sequence using MIDI file data and the modified compound word structure; the compound word sequence comprising a word element sequence of instrument attributes; inputting the compound word sequence into a trained improved Transform neural network model to obtain multi-track music. The application adds instrument attributes to the compound word sequence by modifying the compound word structure, which is conducive to the generation of multi-track music, and the improved Transform neural network model is used to generate multi-track music with better effects.

The application discloses an event detection method and device based on semantic network word representation and attention graph, comprising the following steps: splicing the word content vector, the word structure vector and the position feature vector of each word to generate the feature map of each sentence; combining the POS vector of each word to calculate the attention mechanism and generate the new feature map of each sentence; generating the sentence-level feature vector based on the new feature map; and obtaining the event detection result by using the splicing result of the sentence-level feature vector and the word content vector. The application comprehensively utilizes the external corpus, the semantic network, the part of speech and the attention graph, optimizes the features, more accurately extracts the trigger word, introduces more information, solves the polysemy problem, expresses the associated information between the synonymous words, and obtains more accurate event detection result.

The invention discloses a multi-mode large model security instruction fine adjustment method and system based on virtual confrontation, and belongs to the field of large models. According to the method, two-stage fine tuning is carried out on a multi-modal large model: in the first stage, through a gradient-based virtual confrontation disturbance technology, safe alignment is carried out on a large language model and a cross-modal connection layer, it is ensured that the output content of the model is harmless, and the resistance to subsequent toxic data is improved; in the second stage, on the premise that the parameters after safety alignment are fixed, a trainable implicit cue word structure is introduced, effective adaptation on personalized data of the user is achieved only by adjusting key value pair parameters of the implicit cue word structure, and harmful information is prevented from being introduced again in the fine adjustment process. According to the method, dependence on preference data is reduced, collaborative optimization of security and personalized requirements is realized, and a reliable guarantee is provided for actual deployment of the multi-modal large model.