12results about How to "Reduce call costs" patented technology

This invention relates to an active learning-based data-free black-box attack method and system based on multidimensional value assessment, belonging to the field of artificial intelligence security technology. This method constructs a pre-emptive "sample screening funnel," utilizing a local substitution model to perform multidimensional assessments of sample boundary approximation, information uncertainty, and geometric diversity before sending images to a commercial cloud API. Only high-value samples are selected for querying, thereby achieving low-cost, high-efficiency model theft and adversarial attacks. This invention ensures the diversity and training stability of data-free generated samples, significantly improves the transfer success rate of adversarial examples, and achieves "low-cost, low-risk" economical attacks. It has strong versatility and can be seamlessly integrated into various existing data-free attack frameworks, facilitating deployment and implementation in practical security assessment systems.

The embodiment of the present application relates to the technical field of computer information security and artificial intelligence, and discloses a sensitive information automatic identification method combining a rule engine and a natural language processing technology and related devices, which comprises the following steps: receiving original data to be identified; performing rule matching on the original data to be identified to obtain a matching result; calculating a confidence score corresponding to the matching result; when the confidence score is lower than a first confidence threshold and higher than or equal to a second confidence threshold, inputting the matching result into a lightweight model for verification to obtain a first verification result; when the first verification result is consistent with the matching result, taking the matching result as a text information identification result; when the confidence score is lower than the second confidence threshold or the verification result of the lightweight model on the matching result is inconsistent, inputting the matching result into a large model for analysis and identification to obtain a large model identification result, and taking the large model identification result as the text information identification result. The embodiment of the present application realizes the balance between the text information identification efficiency and accuracy.

The invention discloses an environment survey report structured information extraction method based on a large language model, and the method comprises the steps: obtaining a PDF file of an environment survey report, constructing an environment survey report chapter tree, and carrying out the splitting processing, and obtaining a lightweight environment survey report; introducing a field layer, a fragment layer and a dialogue layer, constructing a hierarchical Prompt large language model, and extracting the lightweight environment survey report to obtain a plurality of candidate results; and performing confidence comprehensive score calculation on the plurality of candidate results, performing screening, and performing adaptive optimization on the screening result to realize extraction of the structured information of the environmental survey report. According to the method, invalid Token calculation can be reduced, the model calling cost is reduced, and meanwhile, the extraction precision of the key field is remarkably improved. The environmental survey report structured information extraction method based on the large language model can be widely applied to the technical field of artificial intelligence.

The application relates to the technical field of data optimization, and discloses a SQL optimization method based on dynamic statistical feature perception, equipment and a medium. The method comprises the following steps: identifying and analyzing a to-be-optimized SQL, and extracting table fields in filtering, association, grouping or sorting conditions as predicate columns; calculating the selectivity of each predicate column, and determining a predicate column with selectivity lower than a threshold value or a single value proportion exceeding a preset proportion as a key column; extracting column-level statistical information including histogram boundary values, null value proportions, different value quantities and the frequency of the most common values for the key column, and converting the column-level statistical information into a natural language statistical summary through a statistical feature descriptor; constructing a prompt word context based on an original SQL statement, table structure information of the key column and the summary, and inputting a large language model to generate a candidate SQL. According to the application, key columns are dynamically screened, and a natural language description of data distribution perception is generated, so that the consumption of Token input of the large language model is reduced, irrelevant information interference is avoided, and the SQL optimization precision is improved.

The application discloses a query condition conversion method and device for adaptive AI output correction, and relates to the technical field of artificial intelligence and natural language processing. The method comprises the following steps: receiving a user natural language input, dynamically constructing a system prompt word based on target table field metadata, sending the system prompt word to a large language model together with the user natural language input, obtaining a JSON format query condition array and transmitting and storing the JSON format query condition array. JSON strings in the array are extracted and converted into a query condition object array, logical operator position correction processing is performed on the query condition object array, the first query condition logical relationship is set to be empty, and the logical operator value of the previous query condition is read as the logical relationship of the previous query condition. The corrected array is converted into a system internal data structure, a database query is performed by using the system internal data structure, and data records meeting the conditions are obtained and returned. The problem that a large language model sets a logical operator on a previous expression, resulting in incorrect query results, is solved.

The invention provides a mass corpus knowledge point automatic labeling method and system, and belongs to the technical field of computers. According to the method, a semantic clustering mechanism is introduced, massive corpora are compressed into at least a few semantic clusters, only representative samples of each cluster are labeled by calling a large language model, and related knowledge point labels are determined step by step according to levels by combining a tree-shaped knowledge point structure and adopting a top-down dynamic layer-by-layer screening type labeling mechanism. And a lightweight increment mapping mechanism is constructed, so that low-delay and high-consistency automatic labeling is realized. According to the method, the problems of low efficiency, high cost, poor structural consistency, insufficient incremental adaptive capacity and the like existing in a mass corpus and ten-thousand-level knowledge point system in a traditional method are solved, automatic, high-precision and interpretable knowledge point mapping of the mass corpus is realized through combination of clustering compression and tree-shaped dynamic labeling, and the knowledge point mapping efficiency is improved. And low-cost increment processing in a continuous corpus input scene is supported.

The application discloses a model calling method, a medical consultation task processing method and device, and relates to the technical field of internet medical treatment and artificial intelligence. A specific embodiment of the model calling method comprises: translating a prompt word of an original language type input by a user into a prompt word of a target language type; the token number of the prompt word of the original language type is greater than the token number of the prompt word of the target language type; performing data compression processing on the prompt word of the target language type; performing intention and complexity recognition on the processed prompt word, and calling a target task model based on the recognition result to process a task corresponding to the prompt word; the token number input to the model can be reduced, and the model calling cost can be reduced. A specific embodiment of the medical consultation task processing method comprises: determining medical consultation information corresponding to a task; calling a medical health large model based on the medical consultation information to generate response information, and taking the response information as a processing result; the task processing cost can be reduced, and the efficiency can be improved.

The invention discloses an interaction method of a model context protocol system and the model context protocol system. The interaction method comprises the steps that in response to a big language model, according to a client request, a service classification list obtaining request is sent to a target virtualization instance, and the target virtualization instance returns a service classification list to the big language model; in response to a target business classification in the business classification list by the large language model, sending a target tool list acquisition request to the target virtualization instance, and returning a target tool list to the large language model by the target virtualization instance; and in response to a target tool in the target tool list, the large language model sends a target tool calling request to the target virtualization instance, and the target virtualization instance executes the target tool calling request to obtain an execution result and sends the execution result to the client through the large language model. Through two-time layered interaction of classifying lists and then tool lists, Token consumption is reduced, and the calling cost of a large language model is reduced.

The embodiment of the invention provides a dialogue processing method, a server, terminal equipment, a storage medium and a program product. In the dialogue processing method, before a large language model is called to answer a target inquiry message, a plurality of knowledge retrieval components corresponding to different knowledge structures can be utilized to perform pre-recall of heterogeneous knowledge and unified screening of the heterogeneous knowledge; therefore, it is ensured that the knowledge data provided for the large language model is reliable data recalled from the heterogeneous knowledge data, and the large language model does not need to execute the structure classification operation of the knowledge data. On one hand, the risk of inaccurate answering caused by the fact that the big language model classifies the knowledge structures wrongly is reduced, and the accuracy of the answering result output by the big language model can be improved. And on the other hand, the knowledge pre-recall operation is executed before the large language model is called, the large language model does not need to be hierarchically called for multiple times, the dialogue response efficiency is improved, and the calling cost of the large language model and the end-to-end response delay are reduced.

The application discloses a model routing method and system, electronic equipment, storage medium and program product, relates to the technical field of artificial intelligence, and is used for guaranteeing uninterrupted user service and automatically switching models in the routing middleware layer, driving the automatic switching of the model by counting token consumption, and better realizing the control of the model calling cost. The method comprises the following steps: receiving a request message initiated by a user corresponding to a node of a tree structure, wherein the tree structure comprises n levels, each level has m nodes, and n is an integer greater than or equal to 2; determining a model gear according to token consumption parameters of each level on a path where the node initiating the request message is located; routing the request message to a large model corresponding to the model gear for processing in the routing middleware layer; the switching of the model gear occurs in the routing middleware layer; and when the model gear is switched, a client API interface protocol remains unchanged and an established connection remains valid and continues.

The application provides an agent memory management method and device based on asynchronous processing and a storage medium. The method comprises the following steps: after an interaction event is written into a first storage area, a processing event associated with an event identifier is published to an event queue, and the corresponding interaction event is read from the first storage area according to the event identifier; when the interaction event passes an importance determination, the interaction event is converted into structured memory data; a second storage area is determined based on a type identifier field of the structured memory data, and the structured memory data is written into at least one second storage area; a query request is parsed to obtain a query condition, at least one target second storage area is determined based on the query condition, the structured memory data related to the query request is read from the target second storage area and is subjected to aggregation processing, and the aggregated memory data is output to an agent to generate a response result. The application can reduce interaction response delay, reduce model calling cost and improve memory processing auditability.

This invention discloses an industry dictionary for the data analysis field, its construction method, and a semantic element recognition system. The industry dictionary for data analysis includes: an entity lexicon for storing entity words related to the data analysis field; a semantic element lexicon for storing the semantic elements corresponding to each entity word; and a mapping relationship lexicon for storing the mapping relationships between entity words and their corresponding semantic elements. By constructing a structured, dynamically updatable industry dictionary, this invention persistently encapsulates and reuses the powerful semantic understanding capabilities of large language models, successfully transforming the semantic element recognition task from "requiring complex model calculations each time" to "requiring only simple dictionary lookups in most cases," thus achieving breakthrough technical effects in terms of efficiency, cost, and intelligent evolution.