7results about How to "Avoid logical confusion" patented technology

This application relates to the field of code generation technology, and discloses a software code generation method and its apparatus, device, and medium. The method includes: responding to a code generation driving event and determining the multimodal information corresponding to the event, wherein the multimodal information corresponds to the original software development requirements; semantically aligning the different modal information in the multimodal information and fusing them to generate software function planning information; extracting each function planning unit as a code generation subtask of the corresponding code generation type according to the code generation type corresponding to the function planning unit described in the software function planning information; calling a large language model to execute each code generation subtask and synchronously triggering the corresponding coding constraint instructions to generate target code. This application, through the semantic alignment and fusion of multimodal information and the synchronous triggering mechanism of coding constraints in the code generation process, highly restores the development requirements of the software code and reduces the subsequent security audit costs and software maintenance risks of the software code.

This specification provides a method for constructing a character background knowledge base and a method for training a dialogue model. The method for constructing the character background knowledge base includes: acquiring character background data of a virtual character; extracting entities and relationships between entities from the character background data; constructing a background knowledge graph of the virtual character using entities as nodes and relationships as edges; assigning corresponding weight information to each layer of the knowledge graph to obtain a background knowledge graph containing weight information; and constructing a background knowledge base of the virtual character based on the background knowledge graph containing weight information. The background knowledge base is used as the context for retrieval enhancement generation of the dialogue model for training the virtual character. This can solve the problems of the virtual character generating dialogue content that is inconsistent with its role positioning, logically contradictory, and outputting responses that deviate from the user's expression focus due to a lack of background knowledge support, thereby improving the immersion of the user in the dialogue process with the virtual character.

The invention discloses a full-process virtual simulation training method and device based on multi-agent collaboration, and relates to the technical field of virtual simulation. Comprising the following steps: acquiring a training target, dynamically generating a task sequence and an event triggering condition by a plot agent, and constructing a unified world state including an environment state, an object state and a task state by a world agent; in the training process, operation of a student is received, an event engine is combined with a world state to trigger a corresponding event, each agent generates a decision packet based on an updated state, a target decision is obtained through conflict detection and weighted arbitration, and world state updating is completed based on a transaction consistency mechanism; meanwhile, an updating result is fed back to the simulation client side, and the multiple intelligent agents are driven to make circular collaborative decisions; ability vectors are constructed in combination with student behavior data, and dynamic evaluation and training content self-adaptive adjustment are achieved.

This application provides a method, device, storage medium, and program product for managing multi-turn dialogues, relating to the field of artificial intelligence technology. The method utilizes a drift detection model to detect drift in the current dialogue content and effectively stores and tracks historical intents through an intent management mechanism. When intent drift is detected, a drift type recognition model can be used to identify the drift type, and differentiated dialogue processing can be performed based on the drift type. This endows the customer service robot with long-term memory and multi-tasking capabilities, allowing it to continue or connect previously unfinished dialogue tasks in any turn without requiring the user to repeat already mentioned questions. Even after multiple turns of intent switching, the customer service robot can still accurately generate responses that meet the user's true needs, achieving a more coherent, intelligent, and humanized multi-turn interactive experience, significantly improving dialogue efficiency and user satisfaction.

The application relates to an ethoxylation reaction stage temperature optimization control method, device, equipment and readable storage medium. The method comprises the following steps: obtaining an ethylene oxide feed flow cumulative amount as a reaction progress judgment benchmark; dividing multiple reaction stages according to the cumulative amount, and presetting a control parameter group for each stage; calculating the cumulative amount in real time, automatically switching the stage when a preset threshold is reached and a precondition is met, and the parameter group of each stage is executed only once; performing cascade control of the kettle bottom temperature and the feed flow according to a feed flow benchmark value to actively adjust the reaction heat release; and automatically switching a heat conduction oil control strategy without disturbance according to a heat conduction oil control mode. The method realizes self-adaptive control of ethoxylation reaction in stages, actively controls temperature from the source, overcomes the hysteresis of traditional passive heat removal, and improves temperature control precision and production safety. Moreover, the method is also suitable for strong exothermic and staged batch reaction processes of propoxylation or polyesterification.

The present application relates to the technical field of artificial intelligence, and provides a data synthesis method and device, electronic equipment and storage medium, wherein the method comprises: generating a heterogeneous attribute graph based on an initial person setting template, the heterogeneous attribute graph containing entity nodes representing person setting attributes and edges representing logical constraint relationships between the person setting attributes; generating question texts and corresponding answer texts based on the heterogeneous attribute graph; and performing person setting reverse deduction based on the question texts and corresponding answer texts, and optimizing the question texts and corresponding answer texts based on the reverse-deduced person setting, thereby solving the problems of high artificial writing cost, poor simple synthesis quality, high quality inspection cost and difficult data quality guarantee, and greatly improving the logical consistency, attribute coverage and style fitting degree of the synthesized data, and providing a data basis for training a large language model with clear self-awareness and stable person setting.

The application relates to the technical field of personalized recommendation, in particular to a dynamic knowledge graph and large model personalized recommendation system, which comprises an intention representation module, a graph calling module, a graph screening module and a large model reasoning generation module. The intention representation module receives user historical behavior logs and real-time interaction signals, generates user current intention representation through semantic coding; the graph calling module calls a dynamic knowledge graph containing entity nodes, relationship edges and time limit weight matrices; the graph screening module performs graph traversal based on the current intention representation, screens out a candidate entity set with real-time correlation scores; and the large model reasoning generation module inputs the two into a large language model, combines entity causal relationships and time limit weights in the graph for reasoning, and generates a candidate recommendation list containing recommendation reasons. The system can improve recommendation accuracy and interpretability and adapt to real-time user needs.