Agent humanization interaction method and system based on cognitive schema tensor decomposition

Abstract

The application provides an agent personification interaction method and system based on cognitive schema tensor decomposition, the method comprising receiving multi-modal input data issued by a user, converting the multi-modal input data into structured cognitive elements, quantifying the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor; decomposing the three-dimensional cognitive schema tensor to obtain a core tensor and a factor matrix; performing double graph mapping based on the core tensor and the factor matrix to obtain a deep intention recognition result and a topic strategy; and generating a personified natural language response based on the deep intention recognition result and the topic strategy and outputting the response, which can realize smooth transition of a topic, avoid the embarrassment of traditional dialogue robots "harsh turning", and improve the personification experience.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent agent interaction, and specifically relates to a method and system for assessing default risk in credit business. Background Technology

[0002] For existing AI agents, there are three major technical pain points in human-computer interaction:

[0003] 1. Cognitive modeling fuzziness: Traditional intelligent agents are mostly based on probabilistic statistical models (such as LLM), lacking explicit modeling of human psychological cognitive processes, which makes it impossible to truly understand the deep psychological schemas behind user language.

[0004] 2. Disconnect between dialogue and thinking: Existing dialogue systems are often generated end-to-end in a "straightforward" manner, lacking the human-like thinking logic of "thinking about the intention first and then planning the topic", which leads to logical jumps or mechanical repetitions in multi-round dialogues.

[0005] 3. Poor controllability of anthropomorphic interaction: In scenarios requiring specific professional logic (such as psychological counseling and educational guidance), the illusion problem and uncontrollability of general large models are difficult to meet professional needs, and there is a lack of a mathematically quantifiable and controllable cognitive representation mechanism.

[0006] In other words, most existing technologies are "technology-heavy and cognition-light," focusing too much on the fluency of natural language generation while neglecting the construction of the agent's internal "cognitive schema." They cannot, like humans, construct structured cognitive models from information obtained through perception and then use these models for reasoning and expression. Currently, there is a lack of technical solutions for mathematically quantifying and modeling psychological cognitive schema theory through tensor decomposition. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides an anthropomorphic interaction method and system for intelligent agents based on cognitive graph tensor decomposition, which solves the technical problems in the prior art.

[0008] In a first aspect, the present invention provides the following technical solution: a human-like interaction method for intelligent agents based on cognitive schema tensor decomposition, comprising:

[0009] Receive multimodal input data sent by the user, convert the multimodal input data into structured cognitive elements, and quantize the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor;

[0010] The three-dimensional cognitive schema tensor is decomposed to obtain the core tensor and factor matrix;

[0011] A dual-graph mapping is performed based on the core tensor and the factor matrix to obtain deep intent recognition results and topic strategies;

[0012] Based on the deep intent recognition results and the topic strategy, a human-like natural language response is generated and output.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: By adjusting the scene dimension in the tensor, this invention can accurately control the speaking style and topic boundaries of the agent in different scenarios, avoiding the risk of "rambling" in general large models. At the same time, tensor decomposition can uncover potential connections between concepts, thereby identifying deep emotions that users do not directly express (such as anxiety schemas identified through complaints about trivial matters), significantly improving the accuracy of intent recognition. Relying on the weight calculation of topic schemas, this invention enables the agent to achieve smooth transitions in topics, avoiding the awkwardness of "abrupt transitions" in traditional chatbots and improving the human-like experience.

[0014] Preferably, the three-dimensional cognitive schema tensor is ,in, These represent the conceptual dimension, the relational dimension, and the scenario dimension, respectively.

[0015] Preferably, the step of decomposing the three-dimensional cognitive schema tensor to obtain the core tensor and factor matrix specifically includes:

[0016] The three-dimensional cognitive schema tensor is decomposed using a first preset formula to obtain the core tensor and factor matrix, wherein the first preset formula is:

[0017] ;

[0018] In the formula, For the core tensor, For three-dimensional cognitive schema tensors, These are the factor matrices representing the mappings of concepts, relationships, and scenarios in the latent space.

[0019] Preferably, the step of performing a bi-graph mapping based on the core tensor and the factor matrix to obtain the deep intent recognition result and topic strategy includes:

[0020] The core tensor is sliced ​​along the scene dimension to obtain a two-dimensional matrix for each scene style. Singular value decomposition is then performed on the two-dimensional matrix to obtain a left singular vector matrix, a feature matrix, and a right singular vector matrix.

[0021] Extract the first few column vectors from the left singular vector matrix to obtain the intent basis vector set for each scene style;

[0022] Identify the current dialogue state and extract feature vectors Calculate the projection coefficients between the feature vectors and the intent basis vectors in the intent basis vector set. :

[0023] ;

[0024] In the formula, For scene style The first in the intention basis vector set One intentional basis vector;

[0025] The scene style and intent basis vector corresponding to the maximum projection coefficient are selected as the target intent to obtain the deep intent recognition result;

[0026] The factor matrix that determines the mapping of concepts in the latent space Each concept prototype in Representation in the topic space:

[0027] ;

[0028] In the formula, For topic vectors, For the core tensor in The element at that location, The factor matrix of the mapping of relations in the latent space is the first... Column vectors Indicates the first Types of relational patterns;

[0029] Construct a directed graph, wherein the nodes of the directed graph are topic vectors corresponding to each concept prototype, the edges of the directed graph are transition relationships between nodes, and the weight matrix of the directed graph... for:

[0030] ;

[0031] In the formula, Representing the concept prototypes The corresponding topic vector, The attenuation coefficient is... For the concept prototype Semantic distance in the original conceptual space, Represents the weight matrix middle OK Column elements;

[0032] The current topic node is determined, and the set of topic nodes with the highest similarity to the target intent is filtered from the directed graph to obtain the target topic node set. The optimal path from the current topic node to any node in the target topic node set is found through a path search algorithm, and the topic sequence on the optimal path is output as the topic strategy.

[0033] Preferably, the step of generating and outputting a human-like natural language response based on the deep intent recognition result and the topic strategy specifically includes:

[0034] The deep intent recognition results and the topic strategy are input into a preset language generation model to output the final response along with corresponding non-linguistic instructions.

[0035] Preferably, after the step of generating and outputting a human-like natural language response based on the deep intent recognition result and the topic strategy, the method further includes:

[0036] The system receives user feedback information in response to the interaction, adjusts the parameters of the three-dimensional cognitive schema tensor in real time through a Bayesian dynamic update mechanism, and uses the updated parameters for the next round of interaction.

[0037] Secondly, the present invention provides the following technical solution: an intelligent agent anthropomorphic interaction system based on cognitive graph tensor decomposition, the system comprising:

[0038] The conversion module is used to receive multimodal input data sent by the user, convert the multimodal input data into structured cognitive elements, and quantize the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor.

[0039] The decomposition module is used to decompose the three-dimensional cognitive schema tensor to obtain the core tensor and factor matrix.

[0040] The mapping module is used to perform a dual-graph mapping based on the core tensor and the factor matrix to obtain deep intent recognition results and topic strategies;

[0041] The output module is used to generate and output a human-like natural language response based on the deep intent recognition result and the topic strategy.

[0042] Preferably, the system further includes:

[0043] The update module is used to receive user feedback information in response, adjust the parameters of the three-dimensional cognitive schema tensor in real time through a Bayesian dynamic update mechanism, and use the updated parameters for the next round of interaction.

[0044] Thirdly, the present invention provides the following technical solution: a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the above-described intelligent agent anthropomorphic interaction method based on cognitive graph tensor decomposition.

[0045] Fourthly, the present invention provides the following technical solution: a storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the above-described intelligent agent anthropomorphic interaction method based on cognitive graph tensor decomposition. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 The flowchart shows the anthropomorphic interaction method for intelligent agents based on cognitive graph tensor decomposition provided in Embodiment 1 of the present invention.

[0048] Figure 2 This is a structural block diagram of the intelligent agent anthropomorphic interaction system based on cognitive graph tensor decomposition provided in Embodiment 2 of the present invention;

[0049] Figure 3 This is a schematic diagram of the hardware structure of a computer provided for another embodiment of the present invention.

[0050] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Detailed Implementation

[0051] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0052] Example 1

[0053] In Embodiment 1 of the present invention, as Figure 1 As shown, an anthropomorphic interaction method for intelligent agents based on cognitive schema tensor decomposition includes:

[0054] S1. Receive multimodal input data sent by the user, convert the multimodal input data into structured cognitive elements, and quantize the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor.

[0055] Wherein, the three-dimensional cognitive schema tensor is ,in, These represent the conceptual dimension, the relational dimension, and the scenario dimension, respectively.

[0056] In step S1, the user's multimodal input, including text and voice intonation, is received and transformed into structured cognitive elements, which can be specifically described as follows:

[0057] By leveraging the text understanding capabilities of large models to replace traditional NLP in identifying and extracting entities (concepts) from user discourse, analyzing the user's emotional valence (positive or negative) and arousal, and using these as initial values ​​for "relationship weights," the context of the current interaction (such as time, location, and preceding topics) is identified as "scene" dimension data.

[0058] Conceptual dimension: Establish a concept dictionary and assign a unique index to each identified entity, keyword or knowledge point. The concept set covers the domain ontology, such as "anxiety", "depression" and "sleep quality" in the psychology field.

[0059] Relationship Dimension: Establishing relation type indexes, including two main categories:

[0060] Logical relationships: such as causal relationships (e.g., "stress → insomnia"), inclusion relationships (e.g., "depressive mood → low mood"), exclusion relationships, etc.;

[0061] Sentiment weight: Represented by continuous values, ranging from [-1, 1]. Negative values ​​indicate negative association, positive values ​​indicate positive association, and the magnitude of the absolute value indicates the strength of the association.

[0062] Scene Dimension: Define the categories of interaction scenes, such as "in-depth consultation", "casual chat", "teaching and tutoring", "sales dialogue", etc., with each scene corresponding to a dimension slice.

[0063] For each element in the three-dimensional cognitive schema tensor, Initialization assignment: Basic assignment: Based on the domain knowledge graph or pre-trained model, set initial values ​​for the typical strengths of common concept-relation pairs in different scenarios; Dynamic detection assignment: For new concepts or relations identified in real-time interaction, initialize according to the following rules: If concept i appears for the first time in scenario k and its association with relation j is detected, then set... , where w is the confidence score extracted from the current statement, and s is the scene adaptation coefficient (calculated based on the matching degree between scene k and the current context); sparsity processing: for concept-relationship-scene combinations with no obvious relationship, the element values ​​are kept to be 0 or a minimum value close to 0 to maintain tensor sparsity.

[0064] S2. Decompose the three-dimensional cognitive schema tensor to obtain the core tensor and factor matrix;

[0065] Step S2 includes:

[0066] The three-dimensional cognitive schema tensor is decomposed using a first preset formula to obtain the core tensor and factor matrix, wherein the first preset formula is:

[0067] ;

[0068] In the formula, For the core tensor, For three-dimensional cognitive schema tensors, These are the factor matrices representing the mappings of concepts, relationships, and scenarios in the latent space;

[0069] Specifically, the above decomposition process is Tucker decomposition, which aims to reduce the dimensionality of high-dimensional sparse cognitive tensors, extract latent and core cognitive schemas, and remove noise. The core tensor represents the agent's most core cognitive logic (i.e., the mathematical ontology of the "cognitive schema"). For example, in the scenario of "psychological counseling", the high correlation between "insomnia" and "anxiety" is shown. The factor matrix maps the projection of the original concepts, relationships and scenarios to the latent space, realizing the abstraction of cognition.

[0070] For the three factor matrices mentioned above, each column of A corresponds to a "latent concept prototype". By finding the top k primitive concepts with the highest weights in the column, semantic labels are assigned (such as "psychological stress-related concept cluster"). For B, each column corresponds to a "latent relationship pattern". By analyzing the weight distribution, the relationship type it represents is identified (such as "causal strong association pattern" and "emotional resonance pattern"). For C, each column corresponds to a "scene response style". By associating with high-frequency scenes, its style characteristics are labeled (such as "empathic listening style" and "knowledge imparting style").

[0071] S3. Perform a dual-graph mapping based on the core tensor and the factor matrix to obtain the deep intent recognition result and topic strategy;

[0072] The purpose of the dual-map mapping is to achieve the separation and linkage between "thinking" and "expression," which can be roughly divided into two steps:

[0073] Intent Schema Mapping (responsible for deep thinking): Based on the decomposed core tensor, identify the current user's deep intent clusters. For example, by analyzing tensor features, it can be identified that the user is superficially talking about "not wanting to eat," while the deep intent schema is "depressive tendencies leading to decreased physiological function." Output: target intent (such as "establishing empathy" or "exploring the causes").

[0074] Topic Schema Mapping (responsible for coherent expression): Based on the output of the intent schema, calculate the cosine similarity of different topic vectors in the factor matrix A (concept space) and construct topic transition paths: select concepts with moderate relevance to the current state (neither abrupt nor repetitive) as the next topic node, and output: specific topic strategies (such as "smoothly transitioning from the topic of food to the topic of sleep").

[0075] Step S3 includes:

[0076] S31. Slice the core tensor along the scene dimension to obtain a two-dimensional matrix under each scene style. Perform singular value decomposition on the two-dimensional matrix to obtain a left singular vector matrix, a feature matrix, and a right singular vector matrix.

[0077] S32. Extract the first few column vectors from the left singular vector matrix to obtain the intent basis vector set for each scene style;

[0078] S33. Identify the current dialogue state and extract feature vectors. Calculate the projection coefficients between the feature vectors and the intent basis vectors in the intent basis vector set. :

[0079] ;

[0080] In the formula, For scene style The first in the intention basis vector set One intentional basis vector;

[0081] S34. Select the scene style and intent basis vector corresponding to the maximum projection coefficient as the target intent to obtain the deep intent recognition result;

[0082] Specifically, after obtaining the target intent, a Hidden Markov Model (HMM) is used to model the intent transition. The state space consists of the indices of each intent basis vector, and the observation sequence consists of the projection coefficient vectors calculated in each round of dialogue. Then, the Viterbi algorithm is used to decode the most probable intent sequence and identify the intent evolution path. The fundamental purpose of these steps is to simulate the dynamics and continuity of intent in human dialogue. In a human dialogue, intent is not static but naturally evolves, deepens, or shifts as the communication deepens, emotions fluctuate, and information is supplemented. For example, from "expressing worries" to "seeking advice," and then to "confirming methods." This step aims to endow the agent with this ability to perceive the flow of intent, rather than simply recognizing a "snapshot" of intent at the current moment.

[0083] S35. Determine the factor matrix of the mapping of concepts in the latent space. Each concept prototype in Representation in the topic space:

[0084] ;

[0085] In the formula, For topic vectors, For the core tensor in The element at that location, The factor matrix of the mapping of relations in the latent space is the first... Column vectors Indicates the first Types of relational patterns;

[0086] S36. Construct a directed graph, wherein the nodes of the directed graph are topic vectors corresponding to each concept prototype, the edges of the directed graph are transition relationships between nodes, and the weight matrix of the directed graph... for:

[0087] ;

[0088] In the formula, Representing the concept prototypes The corresponding topic vector, The attenuation coefficient is... For the concept prototype Semantic distance in the original conceptual space, Represents the weight matrix middle OK Column elements;

[0089] S37. Determine the current topic node, filter the set of topic nodes with the highest similarity to the target intent from the directed graph to obtain the target topic node set, find the optimal path from the current topic node to any node in the target topic node set through the path search algorithm, and output the topic sequence on the optimal path as the topic strategy.

[0090] S4. Generate and output a human-like natural language response based on the deep intent recognition result and the topic strategy.

[0091] Specifically, step S4 is as follows:

[0092] The deep intent recognition results and the topic strategy are input into a preset language generation model to output the final response along with corresponding non-linguistic instructions.

[0093] Non-verbal instructions include facial expression mapping: selecting appropriate facial expressions and gestures from the behavior library based on the emotional tone and dialogue content; timing alignment: ensuring that the key points of non-verbal behavior are aligned with the position of verbal accents; and personality consistency: adjusting the amplitude and frequency of behavior based on the set personality characteristics of the agent.

[0094] In another embodiment of the present invention, the method further includes:

[0095] S5. Receive user feedback information in response, adjust the parameters of the three-dimensional cognitive schema tensor in real time through the Bayesian dynamic update mechanism, and use the updated parameters for the next round of interaction.

[0096] Specifically, the purpose of step S5 is to simulate the dynamic changes in human cognition and adjust the three-dimensional cognitive schema tensor in real time based on interactive feedback. The general steps are as follows:

[0097] A Bayesian network is constructed, with "target intent" as a latent variable and "topic strategy" and user feedback as observed variables. Posterior probability update: when a user shows resistance (negative feedback) to a topic, the system uses the Bayesian formula to update the corresponding weight value in the tensor in reverse (such as reducing the fit of a topic in this scenario), realizing the agent's "instant learning". Memory storage: the updated core tensor parameters are stored to form a personalized long-term memory for the specific user.

[0098] To further illustrate this, the present invention provides the following practical scenarios:

[0099] Scenario 1: Age-friendly intelligent companion robot – “Associative nostalgic chat”;

[0100] Business Background: Elderly people living alone experience memory decline, tend to jump around in conversations, and crave emotional companionship;

[0101] User behavior: An elderly person had just finished talking about their grandson in the morning, and in the afternoon, while looking out the window, they suddenly said, "Oh, the osmanthus cake from that year was so delicious."

[0102] System "thinking" process (technical path):

[0103] Scene awareness: Recognizes that it is afternoon tea time and autumn (the flowering season of osmanthus).

[0104] Dynamic memory retrieval: The system retrieved "long-term memory slices" in the cognitive tensor and found that "osmanthus cake" and the three concept nodes "deceased spouse", "Mid-Autumn Festival" and "grandson's childhood" constituted a high-weight "happy memory schema".

[0105] Smooth transition of topics: Instead of mechanically asking "Where did you buy the osmanthus cake?", the system uses the association path in the tensor to connect the "grandson" discussed in the morning with the "osmanthus cake" now.

[0106] System anthropomorphic reaction:

[0107] Reply: "Yes, I remember you mentioned that it was your husband's best dish (evoking deep memories). When your grandson came back last time, did you also want him to try this flavor? (naturally connecting to the morning's topic, building coherence)";

[0108] Business value: The intelligent agent demonstrated strong "memory" and "human touch," not only catching up on tangent topics but also providing emotional comfort, significantly increasing users' dependence on the robot.

[0109] Scenario 2: K12 Personalized AI Tutor – “Resilient Teaching”;

[0110] User behavior: A student shouted into the microphone, "This question is too hard! I'm not doing it! What a lousy system!"

[0111] System "thinking" process (technical path):

[0112] Multimodal perception: Voice intonation analysis detected high arousal emotions of "anger / frustration"; operational behavior showed frantic mouse clicking.

[0113] Cognitive schema adjustment: The system updates the "current learning state tensor" in real time and detects a sharp drop in the weight of the student's "resilience" dimension. The system determines that continuing to explain the knowledge points (conventional logic) at this point would lead to user churn.

[0114] Strategy shift: The "Intent Schema" is shifted from "knowledge transfer" to "emotional guidance"; the "Topic Schema" difficulty gradient is reduced.

[0115] System anthropomorphic reaction:

[0116] Reply: (Pauses push notifications, tone shifts to calm and encouraging) "This question is indeed the 'big boss' of this chapter, and many students get stuck on their first attempt (normalizing frustration). Let's not look at this question yet. I'll show you a magic trick, a simpler example, and you'll definitely see through its trick at a glance. (Using simplified guidance to spark interest)"

[0117] Business value: It simulates the classroom management skills of an excellent human teacher, preventing students from dropping out of class due to frustration, and improving course completion rate and learning confidence.

[0118] Scenario 3: High-end car intelligent sales consultant – “Lifestyle mapping”;

[0119] Business background: A customer was looking at cars in the showroom, but wasn't interested in the specifications and kept talking about family trips.

[0120] User behavior: A customer asked, "Is the trunk of this car big? Can it fit my fishing gear and my child's bicycle?"

[0121] System "thinking" process (technical path):

[0122] 1. Needs Attribution: On the surface, the question asked about "spatial parameters," but the system used tensor decomposition to uncover the underlying "lifestyle schema"—this was a customer who valued "family time" and "outdoor freedom."

[0123] Script restructuring: The system has suppressed data-driven responses that directly state "500 liters of volume" (low anthropomorphism).

[0124] Scene adaptation: Activate cognitive weights in the "outdoor camping" scenario.

[0125] System anthropomorphic reaction:

[0126] Reply: "Absolutely no problem. Not only can it fit your entire fishing gear, but the rear seats can even be folded down to create a double bed. Imagine camping by the lake with your kids on the weekend, opening the trunk and sitting in the car to stargaze—this car is your mobile observation deck. (Describe the scene, not just list specifications)"

[0127] Business value: The intelligent agent has transformed from a "manual repeater" into a "top salesperson," greatly boosting purchase conversion rates by creating scenarios that evoke emotional resonance.

[0128] The first embodiment of this invention provides an anthropomorphic interaction method for intelligent agents based on cognitive graph tensor decomposition. By adjusting the scene dimension in the tensor, this invention can precisely control the speaking style and topic boundaries of the intelligent agent in different scenarios, avoiding the risk of "rambling" in general large models. At the same time, tensor decomposition can uncover potential connections between concepts, thereby identifying deep emotions that users do not directly express (such as anxiety graphs identified through complaints about trivial matters), significantly improving the accuracy of intent recognition. Relying on the weight calculation of topic graphs, this invention enables the intelligent agent to achieve smooth transitions in topics, avoiding the awkwardness of "abrupt transitions" in traditional chatbots and improving the anthropomorphic experience.

[0129] Example 2

[0130] like Figure 2 As shown, in Embodiment 2 of the present invention, an anthropomorphic interactive system for intelligent agents based on cognitive graph tensor decomposition is provided. The system includes:

[0131] The conversion module 1 is used to receive multimodal input data sent by the user, convert the multimodal input data into structured cognitive elements, and quantize the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor.

[0132] Decomposition module 2 is used to decompose the three-dimensional cognitive schema tensor to obtain the core tensor and factor matrix;

[0133] Mapping module 3 is used to perform a dual-graph mapping based on the core tensor and the factor matrix to obtain deep intent recognition results and topic strategies;

[0134] Output module 4 is used to generate and output an anthropomorphic natural language response based on the deep intent recognition result and the topic strategy.

[0135] The system also includes:

[0136] The update module 5 is used to receive user feedback information in response, adjust the parameters of the three-dimensional cognitive schema tensor in real time through a Bayesian dynamic update mechanism, and use the updated parameters for the next round of interaction.

[0137] Wherein, the three-dimensional cognitive schema tensor is ,in, These represent the conceptual dimension, the relational dimension, and the scenario dimension, respectively.

[0138] Specifically, the decomposition module 2 is used for:

[0139] The three-dimensional cognitive schema tensor is decomposed using a first preset formula to obtain the core tensor and factor matrix, wherein the first preset formula is:

[0140] ;

[0141] In the formula, For the core tensor, For three-dimensional cognitive schema tensors, These are the factor matrices representing the mappings of concepts, relationships, and scenarios in the latent space.

[0142] Specifically, the mapping module 3 is used for:

[0143] The core tensor is sliced ​​along the scene dimension to obtain a two-dimensional matrix for each scene style. Singular value decomposition is then performed on the two-dimensional matrix to obtain a left singular vector matrix, a feature matrix, and a right singular vector matrix.

[0144] Extract the first few column vectors from the left singular vector matrix to obtain the intent basis vector set for each scene style;

[0145] Identify the current dialogue state and extract feature vectors Calculate the projection coefficients between the feature vectors and the intent basis vectors in the intent basis vector set. :

[0146] ;

[0147] In the formula, For scene style The first in the intention basis vector set One intentional basis vector;

[0148] The scene style and intent basis vector corresponding to the maximum projection coefficient are selected as the target intent to obtain the deep intent recognition result;

[0149] The factor matrix that determines the mapping of concepts in the latent space Each concept prototype in Representation in the topic space:

[0150] ;

[0151] In the formula, For topic vectors, For the core tensor in The element at that location, The factor matrix of the mapping of relations in the latent space is the first... Column vectors Indicates the first Types of relational patterns;

[0152] Construct a directed graph, wherein the nodes of the directed graph are topic vectors corresponding to each concept prototype, the edges of the directed graph are transition relationships between nodes, and the weight matrix of the directed graph... for:

[0153] ;

[0154] In the formula, Representing the concept prototypes The corresponding topic vector, The attenuation coefficient is... For the concept prototype Semantic distance in the original conceptual space, Represents the weight matrix middle OK Column elements;

[0155] The current topic node is determined, and the set of topic nodes with the highest similarity to the target intent is filtered from the directed graph to obtain the target topic node set. The optimal path from the current topic node to any node in the target topic node set is found through a path search algorithm, and the topic sequence on the optimal path is output as the topic strategy.

[0156] Specifically, the output module 4 is used for:

[0157] The deep intent recognition results and the topic strategy are input into a preset language generation model to output the final response along with corresponding non-linguistic instructions.

[0158] In other embodiments of the present invention, the present invention provides the following technical solution: a computer, including a memory 102, a processor 101, and a computer program stored in the memory 102 and executable on the processor 101, wherein the processor 101 executes the computer program to implement the above-described intelligent agent anthropomorphic interaction method based on cognitive graph tensor decomposition.

[0159] Specifically, the processor 101 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0160] The memory 102 may include a large-capacity memory for data or instructions. For example, and not limitingly, the memory 102 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 102 may include removable or non-removable (or fixed) media. Where appropriate, the memory 102 may be internal or external to a data processing device. In a particular embodiment, the memory 102 is non-volatile memory. In a particular embodiment, the memory 102 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0161] The memory 102 can be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 101.

[0162] The processor 101 reads and executes the computer program instructions stored in the memory 102 to implement the above-mentioned intelligent agent anthropomorphic interaction method based on cognitive schema tensor decomposition.

[0163] In some embodiments, the computer may further include a communication interface 103 and a bus 100. For example, Figure 3 As shown, the processor 101, memory 102, and communication interface 103 are connected through bus 100 and complete communication with each other.

[0164] The communication interface 103 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 103 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0165] Bus 100 includes hardware, software, or both, that couples components of a computer device together. Bus 100 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 100 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 100 may include one or more buses. Although specific buses are described and illustrated in the embodiments of the present invention, the present invention is contemplated by any suitable bus or interconnect.

[0166] The computer can execute the anthropomorphic interaction method of the present invention based on cognitive graph tensor decomposition based on the acquired intelligent agent anthropomorphic interaction system, thereby realizing intelligent agent anthropomorphic interaction based on cognitive graph tensor decomposition.

[0167] In some further embodiments of the present invention, in conjunction with the above-described intelligent agent anthropomorphic interaction method based on cognitive graph tensor decomposition, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-described intelligent agent anthropomorphic interaction method based on cognitive graph tensor decomposition.

[0168] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0169] More specific examples of readable media (a non-exhaustive list) include: electrical connections (electronic devices) with one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0170] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0171] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0172] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A human-like interaction method for intelligent agents based on cognitive schema tensor decomposition, characterized in that, include: Receive multimodal input data sent by the user, convert the multimodal input data into structured cognitive elements, and quantize the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor; The three-dimensional cognitive schema tensor is decomposed to obtain the core tensor and factor matrix; A dual-graph mapping is performed based on the core tensor and the factor matrix to obtain deep intent recognition results and topic strategies; Based on the deep intent recognition results and the topic strategy, generate and output a human-like natural language response; The three-dimensional cognitive schema tensor is: ,in, These represent the concept dimension, relationship dimension, and scenario dimension, respectively. The specific steps for decomposing the three-dimensional cognitive schema tensor to obtain the core tensor and factor matrix are as follows: The three-dimensional cognitive schema tensor is decomposed using a first preset formula to obtain the core tensor and factor matrix, wherein the first preset formula is: ； In the formula, For the core tensor, For three-dimensional cognitive schema tensors, These are the factor matrices representing the mappings of concepts, relationships, and scenarios in the latent space; The step of performing a dual-graph mapping based on the core tensor and the factor matrix to obtain the deep intent recognition result and topic strategy includes: The core tensor is sliced ​​along the scene dimension to obtain a two-dimensional matrix for each scene style. Singular value decomposition is then performed on the two-dimensional matrix to obtain a left singular vector matrix, a feature matrix, and a right singular vector matrix. Extract the first few column vectors from the left singular vector matrix to obtain the intent basis vector set for each scene style; Identify the current dialogue state and extract feature vectors Calculate the projection coefficients between the feature vectors and the intent basis vectors in the intent basis vector set. : ； In the formula, For scene style The first in the intention basis vector set One intentional basis vector; The scene style and intent basis vector corresponding to the maximum projection coefficient are selected as the target intent to obtain the deep intent recognition result; The factor matrix that determines the mapping of concepts in the latent space Each concept prototype in Representation in the topic space: ； In the formula, For topic vectors, For the core tensor in The element at that location, The factor matrix of the mapping of relations in the latent space is the first... Column vectors Indicates the first Types of relational patterns; Construct a directed graph, wherein the nodes of the directed graph are topic vectors corresponding to each concept prototype, the edges of the directed graph are transition relationships between nodes, and the weight matrix of the directed graph... for: ； In the formula, Representing the concept prototypes The corresponding topic vector, The attenuation coefficient is... For the concept prototype Semantic distance in the original conceptual space, Represents the weight matrix middle OK Column elements; The current topic node is determined, and the set of topic nodes with the highest similarity to the target intent is filtered from the directed graph to obtain the target topic node set. The optimal path from the current topic node to any node in the target topic node set is found through a path search algorithm, and the topic sequence on the optimal path is output as the topic strategy.

2. The intelligent agent anthropomorphic interaction method based on cognitive schema tensor decomposition according to claim 1, characterized in that, The specific steps for generating and outputting a human-like natural language response based on the deep intent recognition result and the topic strategy are as follows: The deep intent recognition results and the topic strategy are input into a preset language generation model to output the final response along with corresponding non-linguistic instructions.

3. The anthropomorphic interaction method for intelligent agents based on cognitive schema tensor decomposition according to claim 1, characterized in that, After the step of generating and outputting a human-like natural language response based on the deep intent recognition result and the topic strategy, the method further includes: The system receives user feedback information in response to the interaction, adjusts the parameters of the three-dimensional cognitive schema tensor in real time through a Bayesian dynamic update mechanism, and uses the updated parameters for the next round of interaction.

4. A human-like interactive system for intelligent agents based on cognitive graph tensor decomposition, wherein the system employs the human-like interactive method for intelligent agents based on cognitive graph tensor decomposition as described in claim 1, characterized in that, The system includes: The conversion module is used to receive multimodal input data sent by the user, convert the multimodal input data into structured cognitive elements, and quantize the cognitive elements into mathematical expressions to obtain a three-dimensional cognitive schema tensor. The decomposition module is used to decompose the three-dimensional cognitive schema tensor to obtain the core tensor and factor matrix. The mapping module is used to perform a dual-graph mapping based on the core tensor and the factor matrix to obtain deep intent recognition results and topic strategies. The output module is used to generate and output a human-like natural language response based on the deep intent recognition result and the topic strategy.

5. The intelligent agent anthropomorphic interaction system based on cognitive schema tensor decomposition according to claim 4, characterized in that, The system also includes: The update module is used to receive user feedback information in response, adjust the parameters of the three-dimensional cognitive schema tensor in real time through a Bayesian dynamic update mechanism, and use the updated parameters for the next round of interaction.

6. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent agent anthropomorphic interaction method based on cognitive schema tensor decomposition as described in any one of claims 1 to 3.

7. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the intelligent agent anthropomorphic interaction method based on cognitive graph tensor decomposition as described in any one of claims 1 to 3.

Images