Intelligent interaction device intention recognition and intelligent decision-making method based on AI large model

Abstract

The invention belongs to the technical field of intelligent interaction, and particularly relates to an intention recognition and intelligent decision-making method of intelligent interaction equipment based on an AI large model. According to the method, the intention of the user can be accurately identified, the probability of misoperation is reduced, the current input of the user is considered, decision optimization is carried out based on historical data, it is ensured that the selected scheme conforms to the habits of the user and has personalized features, the most appropriate decision scheme can be selected in combination with task complexity, execution success rate and user preference, and the user experience is improved. The efficiency and success rate of overall task execution are improved, the operation preference of the user can be learned, each decision is made to better meet the personalized requirement of the user, and intelligent and user-friendly experience is provided.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent interaction technology, and specifically relates to an intention recognition and intelligent decision-making method for an intelligent interaction device based on an AI big model. Background Art

[0002] With the rapid development of artificial intelligence technology, AI big models have become an important tool for achieving efficient and intelligent interaction. AI big models use large-scale training data and complex algorithms to achieve deep learning of data, thereby realizing intelligent decision-making in multiple fields. Intelligent interactive devices refer to devices that can interact with users naturally and efficiently, such as smart speakers, smart phones, smart wearable devices, etc. These devices can accurately understand and intelligently respond to user commands by integrating advanced speech recognition, natural language processing and other technologies. However, as user needs continue to increase, intelligent interactive devices need to more accurately identify user intentions and make more intelligent decisions to meet the diverse needs of users.

[0003] Intent recognition is one of the key technologies in intelligent interactive devices. It extracts the user's true intention by analyzing and understanding the natural language input by the user. The intention recognition method based on AI big model uses large-scale language data to train the model, enabling it to accurately identify the intention in various complex sentences. At the same time, by introducing technologies such as vector databases, the accuracy and efficiency of intent recognition can be further improved.

[0004] Intelligent decision-making is another important part of intelligent interactive devices. The intelligent decision-making method based on AI big model can realize intelligent decision-making for various complex scenarios by utilizing the powerful generalization and expression ability of the model. At the same time, by introducing technologies such as reinforcement learning, the decision-making ability of the model can be continuously optimized to make it more suitable for actual application scenarios.

[0005] Although intelligent interactive devices based on AI big models have made significant progress in intent recognition and intelligent decision-making, there are still some limitations. For example, traditional technologies mainly rely on simple pattern matching and rule engines, which are difficult to fully understand the complex needs of users. The intentions of users during the interaction process are often affected by multiple factors, resulting in insufficient accuracy in intent recognition, which in turn affects the user experience. Summary of the invention

[0006] The purpose of the present invention is to provide an intention recognition and intelligent decision-making method for an intelligent interactive device based on an AI big model, which can more accurately identify user intentions and evaluate and select candidate decisions in multiple dimensions, thereby providing an intelligent and precise decision-making execution process.

[0007] The technical solution adopted by the present invention is as follows: Intention recognition and intelligent decision-making methods for intelligent interactive devices based on AI big models, including: Acquire interaction information from multiple input sources, and acquire corresponding interaction feature information according to the multiple interaction information; Obtain interaction history feature data, obtain intention recognition results corresponding to each interaction feature information according to multiple interaction feature information and interaction history feature information, and obtain a multi-dimensional candidate decision set according to the multiple intention recognition results; Acquire multiple candidate decision information in the candidate decision set, acquire multiple execution steps corresponding to each candidate decision information, and acquire task complexity according to the multiple execution steps corresponding to each candidate decision information; Obtaining historical execution data corresponding to each candidate decision information, and obtaining the execution success rate of the corresponding candidate decision information based on the historical execution data; Obtain user preference for each candidate decision information, obtain the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information, obtain the candidate decision information corresponding to the decision value that meets the first preset condition, and execute the candidate decision information that meets the first preset condition.

[0008] In a preferred solution, the steps of obtaining interaction history feature data, obtaining an intent recognition result corresponding to each interaction feature information according to multiple interaction feature information and interaction history feature information, and obtaining a multi-dimensional candidate decision set according to the multiple intent recognition results include: Get historical data of interactive features; Acquire multiple interactive feature historical information according to the interactive feature historical data, and acquire the interactive feature historical vector corresponding to each interactive feature historical information according to the multiple interactive feature historical information; Obtaining a corresponding interaction feature vector according to each interaction feature information; Obtaining the interaction feature similarity between each interaction feature vector and each interaction feature history vector; Sort multiple interaction feature similarities in descending order, and obtain the largest interaction feature similarity and mark it as the target interaction feature similarity corresponding to each interaction feature vector; According to the similarity of each target interaction feature, the consciousness recognition result corresponding to the corresponding interaction feature history information is obtained, and marked as the intention recognition result corresponding to the interaction feature information; Based on multiple intent recognition results, the decision information corresponding to each intent recognition result is obtained, and a multi-dimensional candidate decision set is constructed by aggregation.

[0009] In a preferred solution, the step of obtaining the consciousness recognition result corresponding to the corresponding interaction feature history information according to the similarity of each target interaction feature and marking it as the intention recognition result corresponding to the interaction feature information includes: Obtaining an intention recognition table, wherein the intention recognition table includes a plurality of feature similarities and an intention recognition result corresponding to each feature similarity; According to the similarity of the target interaction features, the consciousness recognition result corresponding to the corresponding interaction feature history information is obtained from the intention recognition table; The consciousness recognition result corresponding to the corresponding interaction feature history information is marked as the intention recognition result corresponding to the interaction feature information.

[0010] In a preferred solution, the step of obtaining multiple candidate decision information in the candidate decision set, obtaining multiple execution steps corresponding to each candidate decision information, and obtaining the task complexity according to the multiple execution steps corresponding to each candidate decision information includes: Acquire information of multiple candidate decisions in the candidate decision set; Obtain multiple execution steps corresponding to each candidate decision information; Obtain multiple execution step vectors of multiple execution steps corresponding to each candidate decision information; Get the step weight of each execution step vector; The corresponding task complexity is obtained according to the multiple execution step vectors corresponding to each candidate decision information and the corresponding step weights.

[0011] In a preferred solution, the step of obtaining the historical execution data corresponding to each candidate decision information and obtaining the execution success rate of the corresponding candidate decision information according to the historical execution data includes: Obtain the historical execution data corresponding to each candidate decision information; Obtain corresponding multiple historical execution information according to each historical execution data; Get the number of execution steps in each historical execution information and the corresponding number of completed execution steps; The execution success rate of the corresponding candidate decision information is obtained according to the number of each execution step of the plurality of historical execution information and the corresponding number of completed execution steps.

[0012] In a preferred embodiment, the steps of obtaining a user preference for each candidate decision information, obtaining a decision value for each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information, obtaining candidate decision information corresponding to the decision value that meets the first preset condition, and executing the candidate decision information that meets the first preset condition include: Obtain the user preference for each candidate decision information; Obtain the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information; Get the decision threshold; Determine whether the decision value exceeds the decision threshold; If the decision value exceeds the decision threshold, the decision value exceeding the decision threshold is marked as the target decision value; Sort the target decision values ​​in descending order to obtain a target decision table; The decision information corresponding to the maximum target decision value is obtained from the target decision table, marked as the target decision information, and the target decision information is executed.

[0013] In a preferred solution, the step of obtaining the user preference of each candidate decision information includes: Construct preferred time periods; Get the number of all decision information within the preferred period; Obtain the number of times the user chooses to execute the decision information within the preferred time period; The user preference is obtained based on the number of all decision information and the number of times the user chooses to execute the decision information.

[0014] In a preferred embodiment, the step of obtaining the decision threshold comprises: Obtain decision values ​​corresponding to multiple decision information; A decision threshold is obtained according to decision values ​​of multiple decision information.

[0015] In a preferred embodiment, after the steps of obtaining the decision information corresponding to the maximum target decision value from the target decision table, marking it as the target decision information, and executing the target decision information, the method further includes: Obtaining user feedback information on the execution target decision information, wherein the user feedback information includes satisfaction and dissatisfaction; Determine whether the user feedback information is satisfactory; If the user feedback information is unsatisfactory, the decision information will be executed in the order from large to small in the target decision table.

[0016] And, the intention recognition and intelligent decision-making method terminal of the intelligent interactive device based on the AI ​​big model, including: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model.

[0017] The technical effects achieved by the present invention are: The present invention can accurately identify the user's intention and reduce the probability of misoperation. It not only considers the user's current input, but also optimizes decisions based on historical data to ensure that the selected plan is in line with user habits and has personalized characteristics. It can select the most appropriate decision plan based on task complexity, execution success rate and user preferences, improve the efficiency and success rate of overall task execution, learn the user's operating preferences, make each decision more in line with the user's personalized needs, and provide an intelligent and user-friendly experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" that appears in different places in this specification does not refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0022] Secondly, the present invention is described in detail in conjunction with schematic diagrams. When describing the embodiments of the present invention in detail, for the convenience of explanation, the schematic diagrams are only examples and should not limit the scope of protection of the present invention.

[0023] Please refer to the attached Figure 1 As shown, an intention recognition and intelligent decision-making method for intelligent interactive devices based on an AI big model is provided, including: S1. Obtaining interaction information from multiple input sources, and obtaining corresponding interaction feature information according to the multiple interaction information; S2. Obtain interaction history feature data, obtain the intent recognition result corresponding to each interaction feature information according to the multiple interaction feature information and the interaction history feature information, and obtain a multi-dimensional candidate decision set according to the multiple intent recognition results; S3, obtaining multiple candidate decision information in the candidate decision set, obtaining multiple execution steps corresponding to each candidate decision information, and obtaining task complexity according to the multiple execution steps corresponding to each candidate decision information; S4. Obtain historical execution data corresponding to each candidate decision information, and obtain the execution success rate of the corresponding candidate decision information according to the historical execution data; S5. Obtain user preference for each candidate decision information, obtain the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information, obtain the candidate decision information corresponding to the decision value that meets the first preset condition, and execute the candidate decision information that meets the first preset condition.

[0024] As in the above steps S1 to S6, firstly, the user's interaction information is obtained through multiple input sources (such as voice, gesture, touch, etc.), and the feature information of the interaction is extracted. The feature information includes any one or more combinations of the input content, method, frequency, etc., and the extracted interaction features are compared and analyzed with the historical interaction feature data. The intention is identified for each interaction feature information, and the user's intention is predicted. A multi-dimensional candidate decision set is generated based on these intentions, and multiple candidate decision information is obtained from the candidate decision set. The corresponding execution steps are matched for each decision information. The complexity of each task is obtained based on the complexity of these steps (such as the number of steps, required resources and time), and the success rate of each candidate decision in previous executions is evaluated through the obtained historical execution data. By combining the task complexity, execution success rate and user preference (the user's preferred options or past usage habits), the information value of each candidate decision is calculated, a decision value is generated, and the candidate decision information that meets the preset conditions (such as the largest decision value) is screened out and the decision is executed. This can accurately identify the user's intentions and reduce the probability of misoperation. It not only considers the user's current input, but also optimizes decisions based on historical data to ensure that the selected solution is in line with user habits and has personalized characteristics. By combining the task complexity, execution success rate and user preferences, the most appropriate decision solution can be selected to improve the efficiency and success rate of overall task execution. It can learn the user's operating preferences so that each decision is more in line with the user's personalized needs and provide an intelligent and user-friendly experience.

[0025] In a preferred embodiment, the steps of obtaining interaction history feature data, obtaining the intent recognition result corresponding to each interaction feature information according to the multiple interaction feature information and the interaction history feature information, and obtaining a multi-dimensional candidate decision set according to the multiple intent recognition results include: S201, obtaining interaction feature historical data; S202, acquiring a plurality of interaction feature history information according to the interaction feature history data, and acquiring an interaction feature history vector corresponding to each piece of interaction feature history information according to the plurality of interaction feature history information; S203, obtaining a corresponding interaction feature vector according to each interaction feature information; S204, obtaining the interaction feature similarity between each interaction feature vector and each interaction feature history vector; S205, sorting the multiple interaction feature similarities in descending order, and obtaining the largest interaction feature similarity and marking it as the target interaction feature similarity corresponding to each interaction feature vector; S206, obtaining the consciousness recognition result corresponding to the corresponding interaction feature history information according to the similarity of each target interaction feature, and marking it as the intention recognition result corresponding to the interaction feature information; S207. Obtain decision information corresponding to each intention recognition result according to multiple intention recognition results, and summarize and construct a multi-dimensional candidate decision set.

[0026] As in the above steps S201 to S207, the user's interaction feature historical data is extracted from the database or historical records. This data contains feature information of the user when interacting with the device in the past, such as usage frequency, operating habits and preferences, etc., which constitute the basis of historical interaction behavior. From the historical data, each interaction feature information is processed, extracted and converted into interaction feature historical vectors. These vectors represent the user's interaction features at different time points or scenarios. The quantized vectors are convenient for subsequent similarity calculation and pattern recognition. The corresponding feature vector is extracted from the current interaction feature information. Like the historical vector, this feature vector represents the user's behavior and features in the current operation. The similarity between the current interaction feature vector and each interaction feature historical vector is calculated. The higher the similarity, the closer the behavior pattern of the current operation is to the historical operation, which may indicate that the user's intention is similar to a certain operation in history. The calculation formula for interaction feature similarity is: , where It is expressed as interaction feature similarity, A is expressed as interaction feature vector, and B is expressed as interaction feature history vector. All calculated similarity values ​​are sorted from large to small, and the historical feature with the highest similarity is taken and marked as the target similarity of the interaction feature to ensure that the historical interaction mode that best matches the current operation is given priority. Based on the target interaction feature similarity, the corresponding historical intent recognition result is obtained, and the result is marked as the intent recognition result of the current interaction feature. According to all intent recognition results, the corresponding decision information is obtained. This decision information can be multiple feasible operation paths, options or actions. These results are summarized to generate a multi-dimensional candidate decision set for subsequent decision optimization and execution. By introducing historical interaction feature data and feature vector similarity analysis, the user's current intention can be inferred from historical behavior, which greatly improves the accuracy of intent recognition and avoids the error of judging user needs based on a single operation.

[0027] In a preferred embodiment, the step of obtaining the consciousness recognition result corresponding to the corresponding interaction feature history information according to the similarity of each target interaction feature and marking it as the intention recognition result corresponding to the interaction feature information includes: S2061, obtaining an intention recognition table, wherein the intention recognition table includes a plurality of feature similarities and an intention recognition result corresponding to each feature similarity; S2062, acquiring the consciousness recognition result corresponding to the corresponding interaction feature history information from the intention recognition table according to the target interaction feature similarity; S2063. Mark the consciousness recognition result corresponding to the corresponding interaction feature history information as the intention recognition result corresponding to the interaction feature information.

[0028] As in the above steps S2061 to S2063, an intent recognition table is prepared, which contains multiple groups of feature similarities identified in the historical interaction data and their corresponding intent recognition results. Each feature similarity is associated with a user intent recognition result in a historical context. Through this table, historical behaviors similar to the current interaction and their corresponding intentions can be quickly found. According to the target interaction feature similarity calculated previously, the record that best matches this similarity is found in the intent recognition table, and the historical intent recognition result corresponding to the similarity is extracted, ensuring that the intent that best matches the current interaction is found from the historical data, and it is used to infer the user's current intention, and the historical intent recognition result found is marked as the intention recognition result of the current interaction feature information, which reduces the complex reasoning process and makes it more efficient and accurate in identifying user intentions.

[0029] In a preferred embodiment, the step of obtaining multiple candidate decision information in the candidate decision set, obtaining multiple execution steps corresponding to each candidate decision information, and obtaining the task complexity according to the multiple execution steps corresponding to each candidate decision information includes: S301, obtaining information of multiple candidate decisions in a candidate decision set; S302, obtaining multiple execution steps corresponding to each candidate decision information; S303, obtaining multiple execution step vectors of multiple execution steps corresponding to each candidate decision information; S304, obtaining the step weight of each execution step vector; S305 . Obtain corresponding task complexity according to multiple execution step vectors corresponding to each candidate decision information and corresponding step weights.

[0030] As in the above steps S301 to S305, multiple candidate decision information is obtained from the generated candidate decision set. These candidate decision information represent multiple feasible decision paths derived from the user intention recognition results. Each candidate decision information contains a specific plan or step for executing a task. For each candidate decision information, multiple corresponding execution steps are extracted. These steps specifically define the various links or operation processes required to execute the decision. Each execution step is converted into a vector form (execution step vector). The execution step vector represents the characteristics of the step, such as time, resource requirements, complexity, etc. In this quantitative way, a weight is assigned to each execution step vector. The step weight is assigned according to the importance and execution difficulty of each execution step, which may involve considerations of multiple dimensions, such as the time required for the step, computing resources, execution risks, etc. According to the execution step vector of each candidate decision information and the corresponding step weight, the task complexity of the decision information is comprehensively calculated. The calculation formula for task complexity is: , where F represents the task complexity, i represents the number of multiple execution step vectors corresponding to each candidate decision information, i=1,2,3…n, Represented as the i-th execution step vector, Expressed as the weight of the i-th step, task complexity reflects the comprehensive difficulty of executing a decision. The more complex the task, the more steps it may involve, the higher the resource consumption or the greater the execution risk. It can analyze the execution difficulty of the task at a more fine-grained level. Combined with the step weight, the calculation of task complexity is more accurate and can reflect the actual execution differences of different decision paths.

[0031] In a preferred embodiment, the step of obtaining the historical execution data corresponding to each candidate decision information and obtaining the execution success rate of the corresponding candidate decision information according to the historical execution data includes: S401, obtaining historical execution data corresponding to each candidate decision information; S402, acquiring corresponding multiple historical execution information according to each historical execution data; S403, obtaining the number of execution steps in each historical execution information and the corresponding number of completed execution steps; S404: Obtain the execution success rate of the corresponding candidate decision information according to the number of each execution step of the plurality of historical execution information and the number of completed execution steps.

[0032] As in the above steps S401 to S404, the historical execution data corresponding to each candidate decision information is retrieved from the database. This data is the operation information recorded when similar decisions were previously executed, and includes the detailed process and results of task execution. By obtaining the historical execution data, basic data support is provided for the calculation of the execution success rate of the current decision. Each historical execution data may contain records of multiple execution instances, that is, the actual operation process when the decision is executed each time. These execution information are extracted from the historical execution data, including detailed data such as the execution steps, execution time, and resource consumption of each historical decision. For each historical execution information, the total number of execution steps contained in the information is extracted, as well as whether each execution step is successfully completed. The number of execution steps and the number of completions reflect the actual execution status of the task in each historical execution. The execution success rate of the decision is calculated based on the number of execution steps and the corresponding number of completions in each historical execution information. The execution success rate is measured by the ratio of the number of completed steps to the total number of steps. If all execution steps of a decision in history are successfully completed, the success rate of the execution is 100%; if there are unfinished steps, the success rate is reduced accordingly. The calculation formula for the execution success rate is: , where L represents the execution success rate, k represents the number of multiple historical execution information, k=1,2,3…j, It is expressed as the number of completed execution steps. Expressed as the number of execution steps, It is expressed as the ratio of the number of completed execution steps to the number of execution steps of the j-th historical execution information. It can accurately evaluate the execution success rate of each candidate decision based on the previous task execution status. This method based on historical data can greatly improve the reliability of decisions and optimize decisions in a more targeted manner.

[0033] In a preferred embodiment, the steps of obtaining the user preference of each candidate decision information, obtaining the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information, obtaining the candidate decision information corresponding to the decision value that meets the first preset condition, and executing the candidate decision information that meets the first preset condition include: S501, obtaining the user preference of each candidate decision information; S502, obtaining a decision value for each candidate decision information according to the task complexity, execution success rate, and user preference of each candidate decision information; S503, obtaining a decision threshold; S504, determining whether the decision value exceeds the decision threshold; If the decision value exceeds the decision threshold, the decision value exceeding the decision threshold is marked as the target decision value; S505, sorting the target decision values ​​in descending order to obtain a target decision table; S506: Obtain the decision information corresponding to the maximum target decision value from the target decision table, mark it as target decision information, and execute the target decision information.

[0034] As in the above steps S501 to S506, the user preference of each candidate decision information is calculated based on the user's historical behavior, interaction preference and feedback record. This preference can be quantified through machine learning algorithms or preference parameters set by the user to reflect the user's inclination towards different decision options in different scenarios. The comprehensive "decision value" of each decision is calculated based on the task complexity, execution success rate and user preference of each candidate decision information. The calculation formula of the decision value is C=F*L*P, where C represents the decision value, F represents the task complexity, L represents the execution success rate, and P represents the user preference. A minimum decision threshold is set, which represents a minimum acceptable standard. The decision value must exceed this threshold to enter the candidate execution list. This threshold may be adjusted according to the specific usage scenario or user personalized settings, and each candidate decision is judged one by one to see whether the decision value exceeds the preset value. , decision information exceeding the threshold will be marked as "target decision value", indicating that these decisions meet the execution criteria, while decisions below the threshold will be excluded from the candidate list. All target decision values ​​exceeding the decision threshold are sorted in descending order to generate a "target decision table". This table is arranged according to the priority of each candidate decision to ensure that the decision with the highest score can be selected for execution first. The decision with the highest score, that is, the decision information with the largest target decision value, is selected from the target decision table and marked as "target decision information". Subsequently, the target decision information is executed to complete the corresponding task, which can achieve highly personalized decision recommendations. This method not only takes into account the difficulty and success rate of task execution, but also fully respects the user's personal preferences, making the decision more in line with user needs. This multi-dimensional evaluation method can greatly improve the accuracy and quality of recommended decisions and reduce errors or inappropriate choices.

[0035] In a preferred embodiment, the step of obtaining the user preference of each candidate decision information includes: S5011, construct preferred time periods; S5012, obtaining the quantity of all decision information within the preferred time period; S5013, obtaining the number of times the decision information selected by the user to be executed within the preferred time period; S5014. Obtain the user preference according to the quantity of all decision information and the number of times the user chooses to execute the decision information.

[0036] As in the above steps S5011 to S5014, a "preferred period" is constructed based on the user's usage habits and operation records. This period represents the time range for collecting user behavior data. The setting of the preference period can be a fixed time period (for example, the last month) or a dynamically adjusted time range (such as the time period with the highest user operation frequency recently) to ensure that the collection of user preference data has temporal consistency and can reflect the user's selection tendency over a period of time. Within the set preference period, the total number of all candidate decision information received by the user within the period is counted. This statistical data represents all decision options faced by the user within a specific time period and is the benchmark data for calculating the preference. It is also necessary to count the number of decision information actually selected and executed by the user within the preference period. The core of this step is to extract their selection preferences from their actual operation behaviors and filter out the decision information actually executed by the user as the basis for subsequent calculations. The user's preference is calculated by the ratio of the number of decision information selected and executed by the user to the total number of all decision information within the preference period. The calculation formula for the user's preference is: , where D represents the user preference, Indicates the number of times the decision information is executed by the user. Expressed as the number of all decision information, user preference reflects the user's preference for certain decision options during the period. The higher the value, the more frequently the user chooses certain specific decision information during the period. By constructing the preference period, it is possible to accurately capture the user's operating habits and selection tendencies in different time periods, and adjust the recommended decisions in a targeted manner. At different times or scenarios, users may prefer different decisions, making the recommended decisions closer to the user's real needs. When using it, users can experience a more efficient and more expected operating process, reducing unnecessary option screening processes and improving the overall user experience.

[0037] In a preferred embodiment, the step of obtaining the decision threshold comprises: S5031. Obtain decision values ​​corresponding to multiple decision information; S5032. Obtain a decision threshold according to decision values ​​of multiple decision information.

[0038] As in the above steps S5031 to S5032, a decision value has been calculated for each candidate decision information when executing the previous steps. This decision value combines multiple factors such as task complexity, execution success rate and user preference, and represents the priority or importance of each candidate decision information. In this step, the decision values ​​of all candidate decision information are collected, and a decision threshold is calculated based on all the collected decision values. The calculation formula of the decision threshold is: , where Y represents the decision threshold, d represents the number of multiple decision values, d=1,2,3…q, Represented as the maximum value of multiple decision values, Represented as the minimum value of multiple decision values, It is expressed as the d-th decision value. By calculating and applying the decision threshold, high-priority candidate decision information can be quickly screened out, unnecessary decision steps are reduced, and decision efficiency is improved. Especially when there are a large number of candidate decision information, the use of threshold screening can effectively reduce the computational burden and improve the overall execution efficiency.

[0039] In a preferred embodiment, after the steps of obtaining the decision information corresponding to the maximum target decision value from the target decision table, marking it as the target decision information, and executing the target decision information, the method further includes: S507, obtaining user feedback information of the execution target decision information, wherein the user feedback information includes satisfaction and dissatisfaction; S508, determining whether the user feedback information is satisfactory; If the user feedback information is unsatisfactory, the decision information will be executed in the order from large to small in the target decision table.

[0040] As in the above steps S507 to S508, after executing the target decision information, user feedback information on the decision execution results is collected. Such feedback information usually includes user satisfaction evaluation, which is mainly recorded in the form of "satisfied" or "unsatisfied". The collected user feedback information is analyzed to determine whether the user feedback is "satisfied". This judgment is a key link. If the user's feedback shows dissatisfaction, other candidate decision information will be executed in order of decision values ​​from the target decision table. This process aims to try different decisions until a solution that meets user needs is found. This step effectively enables the ability to deal with dissatisfied feedback and improve user satisfaction by flexibly adjusting decision strategies.

[0041] And, the intention recognition and intelligent decision-making method terminal of the intelligent interactive device based on the AI ​​big model, including: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model.

[0042] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. Intention recognition and intelligent decision-making method for intelligent interactive devices based on AI big model, characterized in that: include: Acquire interaction information from multiple input sources, and acquire corresponding interaction feature information according to the multiple interaction information; Obtain interaction history feature data, obtain intention recognition results corresponding to each interaction feature information according to multiple interaction feature information and interaction history feature information, and obtain a multi-dimensional candidate decision set according to the multiple intention recognition results; Acquire multiple candidate decision information in the candidate decision set, acquire multiple execution steps corresponding to each candidate decision information, and acquire task complexity according to the multiple execution steps corresponding to each candidate decision information; Obtaining historical execution data corresponding to each candidate decision information, and obtaining the execution success rate of the corresponding candidate decision information based on the historical execution data; Obtain user preference for each candidate decision information, obtain the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information, obtain the candidate decision information corresponding to the decision value that meets the first preset condition, and execute the candidate decision information that meets the first preset condition.

2. The intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model according to claim 1 is characterized in that: The step of obtaining interaction history feature data, obtaining an intention recognition result corresponding to each interaction feature information according to multiple interaction feature information and interaction history feature information, and obtaining a multi-dimensional candidate decision set according to the multiple intention recognition results includes: Get historical data of interactive features; Acquire multiple interactive feature historical information according to the interactive feature historical data, and acquire the interactive feature historical vector corresponding to each interactive feature historical information according to the multiple interactive feature historical information; Obtaining a corresponding interaction feature vector according to each interaction feature information; Obtaining the interaction feature similarity between each interaction feature vector and each interaction feature history vector; Sort multiple interaction feature similarities in descending order, and obtain the largest interaction feature similarity and mark it as the target interaction feature similarity corresponding to each interaction feature vector; According to the similarity of each target interaction feature, the consciousness recognition result corresponding to the corresponding interaction feature history information is obtained, and marked as the intention recognition result corresponding to the interaction feature information; Based on multiple intent recognition results, the decision information corresponding to each intent recognition result is obtained, and a multi-dimensional candidate decision set is constructed by aggregation.

3. The intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model according to claim 2 is characterized in that: The step of obtaining the consciousness recognition result corresponding to the corresponding interaction feature history information according to the similarity of each target interaction feature, and marking it as the intention recognition result corresponding to the interaction feature information, includes: Obtaining an intention recognition table, wherein the intention recognition table includes a plurality of feature similarities and an intention recognition result corresponding to each feature similarity; According to the similarity of the target interaction features, the consciousness recognition result corresponding to the corresponding interaction feature history information is obtained from the intention recognition table; The consciousness recognition result corresponding to the corresponding interaction feature history information is marked as the intention recognition result corresponding to the interaction feature information.

4. The intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model according to claim 1 is characterized in that: The step of obtaining multiple candidate decision information in the candidate decision set, obtaining multiple execution steps corresponding to each candidate decision information, and obtaining the task complexity according to the multiple execution steps corresponding to each candidate decision information includes: Acquire information of multiple candidate decisions in the candidate decision set; Obtain multiple execution steps corresponding to each candidate decision information; Obtain multiple execution step vectors of multiple execution steps corresponding to each candidate decision information; Get the step weight of each execution step vector; The corresponding task complexity is obtained according to the multiple execution step vectors corresponding to each candidate decision information and the corresponding step weights.

5. The intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model according to claim 1 is characterized in that: The step of obtaining the historical execution data corresponding to each candidate decision information and obtaining the execution success rate of the corresponding candidate decision information according to the historical execution data includes: Obtain the historical execution data corresponding to each candidate decision information; Obtain corresponding multiple historical execution information according to each historical execution data; Get the number of execution steps in each historical execution information and the corresponding number of completed execution steps; The execution success rate of the corresponding candidate decision information is obtained according to the number of each execution step of the plurality of historical execution information and the corresponding number of completed execution steps.

6. The intention recognition and intelligent decision-making method of intelligent interactive device based on AI big model according to claim 1 is characterized in that: The steps of obtaining the user preference of each candidate decision information, obtaining the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information, obtaining the candidate decision information corresponding to the decision value that meets the first preset condition, and executing the candidate decision information that meets the first preset condition include: Obtain the user preference for each candidate decision information; Obtain the decision value of each candidate decision information according to the task complexity, execution success rate and user preference of each candidate decision information; Get the decision threshold; Determine whether the decision value exceeds the decision threshold; If the decision value exceeds the decision threshold, the decision value exceeding the decision threshold is marked as the target decision value; Sort the target decision values ​​in descending order to obtain a target decision table; The decision information corresponding to the maximum target decision value is obtained from the target decision table, marked as the target decision information, and the target decision information is executed.

7. The intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model according to claim 6 is characterized in that: The step of obtaining the user preference of each candidate decision information includes: Construct preferred time periods; Get the number of all decision information within the preferred period; Obtain the number of times the user chooses to execute the decision information within the preferred time period; The user preference is obtained based on the number of all decision information and the number of times the user chooses to execute the decision information.

8. The intention recognition and intelligent decision-making method of intelligent interactive device based on AI big model according to claim 6 is characterized in that: The step of obtaining the decision threshold comprises: Obtain decision values ​​corresponding to multiple decision information; A decision threshold is obtained according to decision values ​​of multiple decision information.

9. The intention recognition and intelligent decision-making method of intelligent interactive device based on AI big model according to claim 6 is characterized in that: After the steps of obtaining the decision information corresponding to the maximum target decision value from the target decision table, marking it as the target decision information, and executing the target decision information, the process further includes: Obtaining user feedback information on the execution target decision information, wherein the user feedback information includes satisfaction and dissatisfaction; Determine whether the user feedback information is satisfactory; If the user feedback information is unsatisfactory, the decision information will be executed in the order from large to small in the target decision table.

10. The terminal of the intention recognition and intelligent decision-making method of the intelligent interactive device based on the AI ​​big model is characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When one or more programs are executed by one or more processors, one or more processors implement the intention recognition and intelligent decision-making method of an intelligent interactive device based on an AI big model as described in any one of claims 1 to 9.

Images