142 results about "Preference vector" patented technology

The invention discloses an AI-based intelligent marketing advertisement putting method and system, and the method comprises the steps: collecting and fusing heterogeneous user behavior data streams from a plurality of independent platforms in real time, and generating a user behavior feature map with a unified space-time mark; on the basis of the user behavior characteristic spectrum, a deep auto-encoder and a space-time diagram convolutional network are used for joint modeling, and fine-grained scene-decoupled user interest preference vectors are generated; inputting the user interest preference vector into a pre-trained generative adversarial network, and dynamically generating a personalized advertisement material highly matched with the current user interest preference and scene; and taking the personalized advertisement material as a candidate arm, and dynamically generating an advertisement putting strategy including an advertisement display form, a display opportunity and a display channel in combination with a current user state and historical putting feedback data. By utilizing the embodiment of the invention, the accuracy and the real-time performance of advertisement putting can be improved, and the advertisement conversion rate and the user experience are improved.

The embodiment of the invention relates to the technical field of artificial intelligence, and provides a personalized learning resource recommendation method and system based on deep learning. The method comprises the following steps: acquiring a user learning scene and a real-time operation behavior to construct a heterogeneous interaction graph; in combination with a Transform meta-coding model of an MAML architecture, pre-training a meta-model based on a real-time operation behavior to adapt to fine tuning of scene parameters; embedding a domain knowledge graph entity to obtain a semantic association vector, and capturing an entity pre-repair relationship; performing multi-hop reasoning on the heterogeneous interaction map through a GAT map attention network, and iteratively generating a user preference vector and a learning resource feature vector; for interactive sparse users, real-time operation behaviors are input into the pre-training meta-model to generate exclusive recommendation parameters, new resource feature vectors are generated in combination with the knowledge graph, and user preference vectors are matched based on the exclusive parameters; and predicting the interaction probability according to the matching vector, and outputting the target recommendation information according to the interaction probability so as to improve the recommendation efficiency and accuracy of the learning resources and guarantee the adaptation degree of the learning resources and the user.

The invention discloses a personalized costume design recommendation system and method based on AI and big data, and relates to the technical field of personalized recommendation data processing, the system comprises an information acquisition module, a feature analysis module, a preference analysis module and a design recommendation module; the information acquisition module is used for acquiring user multi-modal information; the feature analysis module is used for extracting and fusing user preference features from the user multi-modal information and calculating a user multi-modal interest expression vector; the preference analysis module is used for carrying out joint modeling by combining a Transform network and a GRU structure based on the user multi-modal interest expression vector, and generating a user multi-dimensional dynamic preference vector through fusion of a gating mechanism; and the design recommendation module is used for outputting a personalized costume design recommendation result through a reinforcement learning model in combination with the multi-dimensional dynamic preference vector of the user.

The invention relates to the technical field of data processing, and discloses an intelligent agent architecture-based education large model autonomous teaching planning method and system. The method comprises the following steps: inputting a cognitive ability matrix, a knowledge mastery degree vector and a learning preference vector into an improved Bayesian knowledge tracking algorithm, and adding an agent parameter consistency constraint term and a teaching planning coordination factor to obtain a learning track state vector; based on the learning track state vector, multiple agents cooperatively generate a recommendation content identification sequence, a difficulty level sequence and a time distribution sequence; and monitoring the knowledge mastery degree vector variation, and reordering the recommended content sequence when the knowledge mastery degree vector variation exceeds a preset threshold value to generate a teaching sequence. According to the invention, the Bayesian knowledge tracking algorithm is improved, the agent parameter consistency constraint term and the teaching planning coordination factor are fused, and coordination and unification of multi-agent teaching decisions in a communication-free environment are realized.

The invention relates to an underwater wireless sensor network path sensing routing method based on deep reinforcement learning, which comprises the following steps that: firstly, a node constructs and periodically updates a transmission preference model based on local and neighbor node interaction information; secondly, deploying a deep reinforcement learning model at each underwater sensor node to perform distributed routing strategy learning; and finally, generating a global guide vector by the sink node according to the routing path information of the received data packet, reversely spreading the global guide vector to the source node, fusing the global guide vector with a local transmission preference vector of the node to generate a guide reward, optimizing the deep reinforcement learning model, and updating a routing strategy. According to the method, the problems of difference and complexity of underwater transmission tasks can be solved, and the network data transmission efficiency and the overall service quality are improved in combination with local preference and global guidance while the node online learning is kept to adapt to the dynamic underwater environment.

The invention discloses a personalized travel route planning method and system based on a large language model, and a terminal. The method comprises the steps: collecting multi-source data information, carrying out the extraction and standardization of the multi-source data information, obtaining a target multi-source data set, and constructing a travel knowledge graph according to the target multi-source data set; the method comprises the following steps: constructing a big language model, performing instruction type fine tuning on the big language model according to a tourism knowledge graph and a preset fine tuning task, enhancing the big language model after instruction type fine tuning through a knowledge distillation technology to obtain a tourism route planning model, and outputting a user preference vector and a candidate scenic spot set according to the tourism route planning model. And according to a user preference vector and the candidate scenic spot set, generating a personalized travel route plan through a traditional route optimization algorithm. According to the method, entity extraction and knowledge graph dynamic modeling are carried out on the multi-source heterogeneous tourism data, deep semantic understanding and preference analysis are carried out in combination with a large language model, personalized tourism routes can be provided, and the travel efficiency of users is improved.

The invention relates to the technical field of tourism resource information processing, in particular to a tourism hotel dynamic scoring and recommending method and system based on multi-factor fusion, and the method comprises the steps: constructing a dynamic game portrait containing a user space-time behavior mark and hotel multi-modal attributes, and carrying out the dynamic scoring and recommending of a hotel based on the dynamic game portrait; the method comprises the following steps: analyzing a recessive preference vector of a user and an adaptive strategy vector of a hotel by simulating an interactive game process of the user and hotel attributes, performing dynamic equalization matching on the recessive preference vector and the adaptive strategy vector to generate a hotel situation score, generating an anti-consensus recommendation sequence according to the hotel situation score, and performing recommendation on the hotel situation according to the anti-consensus recommendation sequence. The anti-consensus recommendation sequence is subjected to interpretable packaging, and a final recommendation result with a decision guide clue is output to the user, so that the user can obtain a personalized hotel recommendation list and can also understand logic and diversity values behind recommendation, and the user experience is improved. Therefore, the transparency, the user trust degree and the exploration satisfaction degree of the recommendation system are improved.

The invention discloses a big language model dialogue recommendation method based on multi-modal geographic information fusion and context modulation, which comprises the following steps of: firstly, coding and fusing a text address, geographic coordinates and numerical attributes to obtain a geographic context matrix; then intention modeling is carried out to generate user intention representation, then context feature modulation is utilized, dynamic modulation is carried out on the user intention representation by utilizing an asymmetric fusion mechanism, a context guiding signal containing geographical constraints is generated, and finally, the context guiding signal is spliced before the user intention representation. The big language model input to the parameter freezing obtains user preference vectors and calculates recommendation scores with the candidate positions, the candidate positions are sorted according to the recommendation scores, and the candidate positions with high scores are recommended to the user. According to the method, on the premise that internal parameters of the large language model are not changed, the geospatial reasoning ability is effectively given to the large language model, and the accuracy and interpretability of dialogue recommendation are remarkably improved.

The invention relates to the technical field of intelligent logistics and path optimization, and particularly discloses a multi-target vehicle path optimization method based on an enhanced self-attention mechanism. The method comprises the following steps: S1, constructing an enhanced self-attention encoder, and carrying out deep fusion encoding on input node features and preference vectors through a self-attention module integrating channel attention, SwiGLU activation and adaptive scaling; s2, constructing a path generation network based on a problem scale perception decoder, embedding problem scale information injection nodes, and generating a path sequence in an autoregression mode; and S3, based on a reinforcement learning framework, carrying out model training in combination with a random preference strategy, and generating a Pareto frontier approximate solution of a multi-target vehicle path problem. Through the collaborative design, the model expression ability, the generalization ability and the solving precision are improved, and an efficient and reliable solution is provided for the multi-target vehicle path problem.

The invention belongs to the field of man-machine interaction, and particularly discloses a multi-scene switching method, system, equipment and medium based on a song ordering table, and the method comprises the steps: collecting the behavior data of residence time, song ordering frequency and interaction intensity of a user in a virtual environment through a sensor, and processing the behavior data through a long and short term memory network, obtaining a user dynamic preference vector; matching is carried out according to the user dynamic preference vector and a preset scene type library, if the matching degree is higher than a threshold value, it is judged that the current scene preference is stable, otherwise, it is judged that potential switching intentions exist, and potential switching intention probability distribution is obtained through the judgment; after the probability distribution of the potential switching intention is obtained, a support vector machine classifier is adopted to train association features among multi-scene data; the invention aims to solve the problems of resource loading delay and unsmooth interface switching caused by rapid change of user scene preference in a virtual environment in the prior art.

The invention provides a multi-modal data joint query analysis method and system supporting natural language interaction, and relates to the technical field of data query analysis. Historical operation data, audio data and text data of a user on an intelligent search platform supporting natural language interaction are collected; modeling the historical operation data to obtain a user preference vector, performing voice recognition and text standardization processing on the audio data to obtain standard query data, and integrating the standard query data and the text data into context information; an attention mechanism-based algorithm is used for semantic understanding and intention recognition to obtain an intention feature vector, and the intention feature vector is fused with a user preference vector to obtain a classification result; and finally, based on the result, querying in a preset multi-modal database through a collaborative filtering algorithm to obtain a joint query result, so that personalized accurate query of the multi-modal data under natural language interaction can be realized, and the result fits the intention and long-term preference of the user.

The invention discloses a cross-domain AI knowledge aggregation method based on collaborative filtering. The method comprises the steps that S1, multi-source heterogeneous AI knowledge data and user behavior data are collected and preprocessed; s2, constructing a double-tower cross-domain embedded network, and outputting a cross-domain semantic fusion sequence; s3, modeling and analyzing user preferences through the improved Bi-GRU network, and generating user behavior preference vectors; s4, performing semantic diffusion and neighborhood reasoning on the cold start user, and complementing interest features; s5, adopting a double-tower recall structure and an XGBoost model to sort and generate a cross-domain recommendation list; s6, constructing a context rule base to execute context adaptability judgment, and generating a matched knowledge aggregation recommendation list; and S7, performing incremental learning according to user feedback information, and dynamically updating the double-tower cross-domain embedded network and the improved Bi-GRU network. According to the method, the knowledge matching precision, the cold start adaptability and the scene adaptation capability of cross-domain recommendation content are improved.

The invention discloses a personalized learning scheme recommendation method and system based on artificial intelligence, and relates to the technical field of intelligent recommendation, and the method comprises the steps: collecting a learning behavior track and a knowledge mastering progress, carrying out the normalization processing, and generating structural feature input; performing multi-dimensional feature mapping and weight training on the structured feature input by using a reinforcement learning model to generate a learning preference vector; matching the learning preference vector with the weighted feature interaction matrix, calculating an adaptive score, and generating a candidate resource set; and performing iterative fusion on the candidate resource set and the real-time interaction feedback vector to generate a feedback fusion adaptive score matrix, updating adaptive scores by adopting a dynamic adjustment method, forming an optimized resource sequence, converting the optimized resource sequence into a task execution path, and generating personalized learning scheme recommendation. According to the invention, the accuracy of learning resource matching and the timeliness of personalized recommendation are improved.

The invention discloses a federal privacy protection sequence recommendation method based on knowledge increment collaboration, and relates to the field of data privacy protection. The invention aims to solve the problems of privacy exposure and low recommendation accuracy of the existing federal privacy protection sequence recommendation method. The method comprises the following steps: acquiring historical interaction articles of a user, inputting the historical interaction articles of the user into an optimized client, sending a global embedding matrix to the optimized client by an optimized server, acquiring a score of each article in a candidate pool by the optimized client by utilizing the global embedding matrix, and storing the score of each article in the candidate pool in the candidate pool, and forming a recommendation list by the L articles with the highest scores, and outputting the recommendation list. According to the method, the preference vector and other high-entropy features of the user are always kept in the local of the client, so that a channel of leakage of sensitive information through a gradient path is fundamentally cut off, and privacy exposure of the user is avoided. The method is used for sequence recommendation.

The invention relates to the field of communication, and discloses an intelligent traffic scheduling optimization method, device and equipment and a storage medium, and the method is used for 5G network intelligent traffic scheduling optimization. The method comprises the following steps: receiving a service request, analyzing the service request to obtain a 5G standard service type, an application scene and a 5G network slice identifier, then obtaining an SLA parameter corresponding to the service request and a candidate link required by service transmission, and converting the SLA parameter into an SLA preference vector; acquiring global data of the candidate link, and predicting by using the network state prediction model to obtain a predicted network state of the candidate link; the SLA preference vector is used as optimization target guidance, the predicted network state is used as a constraint condition, and a global optimization routing strategy is calculated through a multi-target optimization algorithm; and generating a control message packet based on the global optimization routing strategy, issuing the control message packet to the network forwarding equipment, and adjusting parameters of the network state prediction model and the multi-objective optimization algorithm based on the actual performance data after execution.

The invention belongs to the technical field of workshop production scheduling, and discloses a flexible job workshop multi-target scheduling method and system based on preference driving, and the method comprises the steps: defining a target function and a constraint condition based on obtained information, and constructing a workshop scheduling model; converting a workshop scheduling problem into a Markov decision problem, designing a reward function, and creating a preference pool; selecting a preference vector from the preference pool, and respectively inputting the preference vector and the system state into a strategy network and a value network for iterative training to obtain a trained strategy network and a trained value network; and obtaining current preferences for different dispatches and a current state of the system, inputting the current preferences and the current state of the system into the trained strategy network, obtaining action probability distribution in a given state, and selecting a dispatching action with the maximum probability to obtain a corresponding optimal dispatching scheme. According to the method, the requirements of tire enterprises in different situations are met, high flexibility is achieved, the nonlinear target utility function is designed, and a tire workshop scheduling scheme with higher quality can be found out easily.

The invention discloses an APP interface visual communication adaptive optimization method and system based on user satisfaction, and the method comprises the steps: constructing a visual element-satisfaction association prediction model, taking historical satisfaction score, operation feedback and scene feature data as a training basis, and carrying out the training through a gradient boosting tree algorithm; collecting real-time scene features and initial interaction data of a target user, and generating candidate visual element parameters and prediction scores; constructing an instant preference vector, and screening a to-be-verified visual scheme in combination with cosine similarity; a local gray level display scheme is adopted, and multi-modal feedback is collected; constructing an evaluation matrix, and calculating a real-time satisfaction comprehensive score by using an analytic hierarchy process; determining a scheme and incrementally updating the model if the standard is reached, and regenerating parameters if the standard is not reached; the system comprises six components such as a model iteration engine and a real-time data collector, and dynamic adaptation is achieved cooperatively. According to the method, user preferences and scenes can be accurately matched, the visual experience and the operation efficiency are improved, continuous optimization can be achieved through model iteration, and the method is suitable for various APPs needing personalized interfaces.

The invention relates to the technical field of user interaction, and discloses an interpretable news recommendation method and device. According to the method, the user preference vector is obtained based on comprehensive analysis of the attributes, the long-term behaviors and the short-term behaviors of the user, so that more reliable analysis can be given from the attributes and the behaviors of the user. The semantic features of the candidate news are subjected to semantic enhancement in combination with the preset knowledge base, so that higher semantic understanding depth is achieved. The vector similarity can provide a news recommendation basis from the perspective of feature analysis, the semantic association degree can provide a news recommendation basis from the perspective of semantic analysis, the deep semantic association between the user and the news can be captured through combination of the vector similarity and the semantic association degree, and the reliability of news recommendation is finally improved. The first matching information is matched features, the second matching information is matched text descriptions, and the first matching information and the second matching information are combined with each other, so that the news recommendation thought or reason can be displayed for the user, and the news recommendation method has interpretability.

The invention provides a resource recommendation method and system based on a large model, and the method comprises the steps: S1, obtaining the historical behavior information and the current query information of a user, and converting the historical behavior information and the current query information of the user into a user preference vector and a user demand vector; s2, based on the user preference vector and the user demand vector, using a language large model to extract a keyword set vector; s3, extracting a candidate resource set from a resource library by using a vector similarity retrieval method according to the user preference vector, the keyword set vector and the user demand vector; s4, reordering the candidate resource set by adopting a bootstrap strategy to obtain a candidate resource sequence; and S5, updating the user preference vector according to the first N items of the candidate resource sequence, then repeating the steps S2-S5 until the maximum number of iterations, and then recommending the first N items of resources. According to the method, the problems of poor generalization ability and insufficient adaptation in the prior art are solved.

The invention provides a household service supply and demand intelligent docking management method and system based on big data analysis. The method comprises the following steps: collecting original data; processing and splicing the family structure information and the historical order information to obtain a comprehensive feature vector, and inputting the comprehensive feature vector into a multi-layer perceptron to obtain a life cycle stage vector; projecting the life cycle stage vector to a space consistent with a preference vector to obtain a stage service mask; carrying out dimension-by-dimension multiplication operation on the stage service mask, the preference vector and the attention weight to obtain a preference fusion vector; constructing a first service capability vector for each service staff, and reconstructing the preference fusion vector and the service capability vector to obtain a second service capability vector; and calculating an average value to obtain a comprehensive score, sorting the comprehensive scores of all candidate service personnel according to a score descending order, and finally selecting the candidate service personnel with the highest comprehensive score and meeting a threshold condition as an order sending object.

The invention discloses a tea space color matching method, device and equipment based on user preferences and a storage medium, and relates to the technical field of intelligent design, and the tea space color matching method based on the user preferences comprises the following steps: obtaining a tea ceremony type label, a tea ware picture, a tea space intention picture, an artistic conception text description and other multi-modal data input by a user; analyzing and generating a tea culture semantic vector, and extracting key features in the image by using an improved convolutional neural network to generate an image feature vector; and then, fusing the two into a user preference vector through a multi-modal attention mechanism, and finally, calculating the similarity between the vector and a color matching scheme feature vector in a preset database, and determining a most matched target tea space color matching scheme. According to the method, the personalized color matching scheme conforming to tea space culture attributes can be generated according to user interests and preferences, and efficient and accurate personalized recommendation is realized.

The application discloses a kind of based on key value memory network's complex electromechanical system design knowledge active recommendation method.First, the scene feature semantic information extraction of software platform log file in the design process of complex electromechanical system is carried out, and scene ontology is established based on scene feature semantic information, and system scene knowledge base is formed by scene ontology and original knowledge base;Then, the scene maximum frequent sequence is used to describe the scene sequence feature similarity between designers;Again, the knowledge item interaction sequence of all designers is learned, and the knowledge item sequence preference vector corresponding to all designers is obtained;Further, input to key value memory network, and the initial design knowledge active recommendation sequence is obtained;Finally, after knowledge item selection, the final design knowledge active recommendation sequence of each designer is obtained.The application obtains design knowledge active recommendation sequence more in line with the needs of designers, and then improves the efficiency of complex electromechanical system design.

The invention relates to resource allocation in a wireless communication network. The present disclosure proposes a network device for efficient multi-target radio resource allocation for user equipment. The network device includes a preference module and a policy module. The preference module is configured to determine a global preference vector, where the global preference vector describes an overall weight of each network performance metric in a set of network performance metrics for one or more user devices, and provide the global preference vector to the policy module. And the strategy module is used for acquiring the global preference vector from the preference module and making a resource allocation decision based on the global preference vector.

The application discloses an intelligent agent generated interactive system and method applied to hospital operation management, relates to the technical field of hospital operation management, and captures each interactive operation of a user for an operation entity from a hospital heterogeneous system and records each operation as an atomic interactive event; analyzes the atomic interactive event to generate a user long-term preference vector and stores the user long-term preference vector; when a natural language query is received, a target operation entity identifier and an analysis index are analyzed; the analyzed index is matched with a candidate analysis model, an execution model is generated, and an index analysis result set is output; the data type and the dimension quantity of the index analysis result set are analyzed, and a basic UI component requirement is generated; the UI component requirement and a UI component preference are fused and decided, a target UI component selection is generated, and a final interactive interface is rendered and generated; and the frequency distribution is updated according to the user interaction. Personalized decision is supported, and the use experience of the user is improved.

The application discloses a real-time dynamic packing optimization method and system for high hybrid SKU, relates to the technical field of logistics management, and comprises the following steps: acquiring a loading task and a current packing state; analyzing field-specific language business rules to generate executable constraint functions, graph construction metadata and arbitration metadata; constructing constraint dependency graphs, geometric graphs and heterogeneous state graphs; updating the heterogeneous state graph when an order is changed or a container is changed, and determining an influence domain based on logical association and geometric adjacency; inputting the updated heterogeneous state graph into a constraint-aware graph neural network to output a priority score and a position preference vector; generating an action set in the influence domain, pruning and locally re-optimizing to obtain an updated packing scheme. The scheme realizes rule structured expression, incremental solution of the influence domain and local rapid repair, reduces the dynamic rearrangement calculation amount and response time delay, and improves the packing compliance, stability and loading rate.

The invention discloses a product personalized recommendation method and system based on multi-source data fusion. The method comprises the following steps: acquiring multi-dimensional behavior data of a user, associating financial product attributes with user spatio-temporal characteristics to construct a product association map, fusing multi-source semantic information, and performing noise elimination to generate a unified semantic vector; after reaching the standard, extracting time sequence interaction information to construct a deep customization preference vector, analyzing a sub-graph semantic coincidence degree to obtain semantic analysis strength, and generating a cross-scene association path identification potential preference point set; screening the core preference subsets to obtain a recommendation list, and matching the user motivation components to determine a final recommendation result. According to the method, accurate personalized recommendation driven by multi-source data can be realized, and high-quality recommendation requirements of deep customization and cross-scene adaptation in a complex scene are met.

The application provides an information recommendation method, a training method and device of a research report information scoring prediction model, equipment and a storage medium. It relates to the technical field of machine learning. The method comprises: obtaining a behavior sequence of a target object, and dividing the behavior sequence into multiple session information; performing encoding processing on each session information to obtain a first feature vector of each session information; performing behavior-to-behavior time distance feature extraction on multiple first feature vectors of multiple session information to obtain a second feature vector; performing real intention spatial distance feature extraction on the second feature vector based on a multi-head self-attention mechanism to obtain a preference vector of the target object; and obtaining a research report information recommendation list of the target object from at least one candidate research report information according to the preference vector, and recommending the research report information recommendation list to the target object. The method can improve the accuracy of the recommendation result.

The present application relates to the technical field of artificial intelligence and recommendation system, in particular to a kind of dynamic interaction perception and multilayer semantic compression generative recommendation method, comprising the following steps: obtaining the original attribute information of article and carrying out semantic extension, generating enhanced article description text;The historical interaction sequence of user is obtained, and the historical interaction sequence is carried out time feature extraction, multi-factor weighting and gate memory state update, to generate the dynamic preference vector of user;The historical interaction sequence is divided into multiple blocks, and the block-level abstract of each block is generated using a large language model and recursively compressed to obtain a global preference abstract;Question and answer samples are constructed, and a large language model is trained based on the question and answer samples to obtain the trained large language model.The present application improves the modeling capability of the recommendation system for user interest evolution through dynamic interaction perception and multilayer semantic compression, reduces long sequence input redundancy, improves the inference efficiency of the large model, and enhances the accuracy and interpretability of the recommendation results.

The invention relates to a personalized design generation method and system based on a user portrait, equipment and a medium. The method comprises the following steps: analyzing historical interaction data of a target user to generate a structured user preference vector; constructing a dual-path evaluation model, and generating an initial design image through a conditional diffusion model based on the structured user preference vector and the design task text description; inputting the initial design image into the double-path evaluation model to obtain a corresponding subjective matching degree score and a corresponding objective quality score, and calculating a comprehensive reward score by combining the calculated fusion weight coefficient; with maximization of the comprehensive reward score as a target, the conditional diffusion model is updated, and a personalized design generation model is obtained; and generating a personalized design image through the personalized design generation model and the design request of the target user. By adopting the method, the core problem that subjective preference and objective quality are difficult to quantitatively align and dynamically balance during personalized generation in the prior art can be effectively solved.

The invention relates to a multi-target vehicle path optimization method based on preference perception expert mixed deep reinforcement learning, and belongs to the field of path optimization. The method comprises the following steps: inputting a problem instance and a preference vector to an encoder; the encoder outputs a node embedding set through a multi-head attention mechanism; the node embedding set is input into an adapter, a decoder is modulated according to the preference vector, adaptive features are obtained, and the adaptive features are used for each time step in the decoder; the state information of the decoder is represented by a context vector, and at each time step of the decoder, the inference nodes are modulated according to the adaptive features and the score level until all the nodes are accessed, and the selection probability distribution is obtained. The problem that efficient and extensible real-time reasoning and generalization ability is difficult to achieve on the premise of guaranteeing performance due to the fact that an effective zero sample preference generalization mechanism is lacked, preference modeling is rough, parameter efficiency is low and a training strategy is unstable is solved.